diff --git "a/site/search/search_index.json" "b/site/search/search_index.json" --- "a/site/search/search_index.json" +++ "b/site/search/search_index.json" @@ -1 +1 @@ -{"config":{"lang":["en"],"separator":"[\\s\\-,:!=\\[\\]()\"`/]+|\\.(?!\\d)|&[lg]t;|&","pipeline":["stopWordFilter"],"fields":{"title":{"boost":1000.0},"text":{"boost":1.0},"tags":{"boost":1000000.0}}},"docs":[{"location":"","title":"The DETERMINATOR","text":"

Generalist Deep Research Agent - Stops at Nothing Until Finding Precise Answers

The DETERMINATOR is a powerful generalist deep research agent system that uses iterative search-and-judge loops to comprehensively investigate any research question. It stops at nothing until finding precise answers, only stopping at configured limits (budget, time, iterations).

Key Features: - Generalist: Handles queries from any domain (medical, technical, business, scientific, etc.) - Automatic Source Selection: Automatically determines if medical knowledge sources (PubMed, ClinicalTrials.gov) are needed - Multi-Source Search: Web search, PubMed, ClinicalTrials.gov, Europe PMC, RAG - Iterative Refinement: Continues searching and refining until precise answers are found - Evidence Synthesis: Comprehensive reports with proper citations

Important: The DETERMINATOR is a research tool that synthesizes evidence. It cannot provide medical advice or answer medical questions directly.

"},{"location":"#features","title":"Features","text":""},{"location":"#quick-start","title":"Quick Start","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync dependencies\nuv sync\n\n# Start the Gradio app\nuv run gradio run src/app.py\n

Open your browser to http://localhost:7860.

For detailed installation and setup instructions, see the Getting Started Guide.

"},{"location":"#architecture","title":"Architecture","text":"

The DETERMINATOR uses a Vertical Slice Architecture:

  1. Search Slice: Retrieving evidence from multiple sources (web, PubMed, ClinicalTrials.gov, Europe PMC, RAG) based on query analysis
  2. Judge Slice: Evaluating evidence quality using LLMs
  3. Orchestrator Slice: Managing the research loop and UI

The system supports three main research patterns:

Learn more about the Architecture.

"},{"location":"#documentation","title":"Documentation","text":""},{"location":"#links","title":"Links","text":""},{"location":"LICENSE/","title":"License","text":"

DeepCritical is licensed under the MIT License.

"},{"location":"LICENSE/#mit-license","title":"MIT License","text":"

Copyright (c) 2024 DeepCritical Team

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/","title":"MkDocs & Material UI Improvement Assessment","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#current-configuration-analysis","title":"Current Configuration Analysis","text":"

Your current mkdocs.yml already includes many excellent features: - \u2705 Material theme with light/dark mode toggle - \u2705 Navigation tabs, sections, expand, and top navigation - \u2705 Search with suggestions and highlighting - \u2705 Code annotation and copy buttons - \u2705 Mermaid diagram support - \u2705 Code include plugin - \u2705 Minification for performance - \u2705 Comprehensive markdown extensions

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#recommended-improvements","title":"Recommended Improvements","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#1-versioning--multi-version-documentation--high-priority","title":"1. Versioning & Multi-Version Documentation \u2b50 High Priority","text":"

If you plan to maintain multiple versions or branches:

plugins:\n  - search\n  - mermaid2\n  - codeinclude\n  - minify:\n      minify_html: true\n      minify_js: true\n      minify_css: true\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n  # Optional: For versioning\n  # - versioning:\n  #     version: ['dev', 'main']\n

Benefits: Shows when pages were last updated, helps users understand document freshness.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#2-git-integration--revision-information--high-priority","title":"2. Git Integration & Revision Information \u2b50 High Priority","text":"

Add revision dates and authors to pages:

plugins:\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n      fallback_to_build_date: true\n  - git-committers:\n      repository: DeepCritical/GradioDemo\n      branch: dev\n

Benefits: Users see when content was last updated, builds trust in documentation freshness.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#3-enhanced-navigation-features--high-priority","title":"3. Enhanced Navigation Features \u2b50 High Priority","text":"

Add breadcrumbs and improve navigation:

theme:\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.indexes  # Add index pages\n    - navigation.instant  # Instant page loads\n    - navigation.tracking  # Track scroll position\n    - navigation.smooth  # Smooth scrolling\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link  # Link to specific tabs\n    - content.tooltips  # Tooltips for abbreviations\n

Benefits: Better UX, easier navigation, professional feel.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#4-content-tabs-for-code-examples--high-priority","title":"4. Content Tabs for Code Examples \u2b50 High Priority","text":"

Perfect for showing multiple code examples (Python, TypeScript, etc.):

markdown_extensions:\n  - pymdownx.tabbed:\n      alternate_style: true\n      combine_header_slug: true  # Add this\n

Usage in docs: 

=== \"Python\"\n    ```python\n    def example():\n        pass\n    ```\n\n=== \"TypeScript\"\n    ```typescript\n    function example() {}\n    ```\n

Benefits: Clean way to show multiple implementations without cluttering pages.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#5-enhanced-admonitions--medium-priority","title":"5. Enhanced Admonitions \u2b50 Medium Priority","text":"

Add more admonition types and better styling:

markdown_extensions:\n  - admonition\n  - pymdownx.details\n  - pymdownx.superfences:\n      custom_fences:\n        - name: mermaid\n          class: mermaid\n          format: !!python/name:pymdownx.superfences.fence_code_format\n        # Add custom admonition fences\n        - name: danger\n          class: danger\n          format: !!python/name:pymdownx.superfences.fence_code_format\n

Usage: 

!!! danger \"Important\"\n    This is a critical warning.\n

Benefits: Better visual hierarchy for warnings, tips, and important information.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#6-math-formula-support--medium-priority-if-needed","title":"6. Math Formula Support \u2b50 Medium Priority (if needed)","text":"

If your documentation includes mathematical formulas:

markdown_extensions:\n  - pymdownx.arithmatex:\n      generic: true\n  - pymdownx.superfences:\n      custom_fences:\n        - name: math\n          class: arithmetic\n          format: !!python/name:pymdownx.superfences.fence_code_format\n\nextra_javascript:\n  - javascripts/mathjax.js\n  - https://polyfill.io/v3/polyfill.min.js?features=es6\n  - https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js\n

Benefits: Essential for scientific/technical documentation with formulas.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#7-better-code-highlighting--medium-priority","title":"7. Better Code Highlighting \u2b50 Medium Priority","text":"

Add more language support and better themes:

markdown_extensions:\n  - pymdownx.highlight:\n      anchor_linenums: true\n      line_spans: __span\n      pygments_lang_class: true\n      use_pygments: true\n      noclasses: false  # Use CSS classes instead of inline styles\n

Benefits: Better syntax highlighting, more language support.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#8-social-links-enhancement--low-priority","title":"8. Social Links Enhancement \u2b50 Low Priority","text":"

Add more social platforms and better icons:

extra:\n  social:\n    - icon: fontawesome/brands/github\n      link: https://github.com/DeepCritical/GradioDemo\n      name: GitHub\n    - icon: fontawesome/brands/twitter\n      link: https://twitter.com/yourhandle\n      name: Twitter\n    - icon: material/web\n      link: https://huggingface.co/spaces/DataQuests/DeepCritical\n      name: HuggingFace Space\n    - icon: fontawesome/brands/discord\n      link: https://discord.gg/yourserver\n      name: Discord\n

Benefits: Better community engagement, more ways to connect.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#9-analytics-integration--medium-priority","title":"9. Analytics Integration \u2b50 Medium Priority","text":"

Add privacy-respecting analytics:

extra:\n  analytics:\n    provider: google\n    property: G-XXXXXXXXXX\n  # Or use privacy-focused alternative:\n  # analytics:\n  #   provider: plausible\n  #   domain: yourdomain.com\n

Benefits: Understand how users interact with your documentation.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#10-custom-cssjs-for-branding--low-priority","title":"10. Custom CSS/JS for Branding \u2b50 Low Priority","text":"

Add custom styling:

extra_css:\n  - stylesheets/extra.css\n\nextra_javascript:\n  - javascripts/extra.js\n

Benefits: Customize appearance, add interactive features.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#11-better-table-of-contents--medium-priority","title":"11. Better Table of Contents \u2b50 Medium Priority","text":"

Enhance TOC with more options:

markdown_extensions:\n  - toc:\n      permalink: true\n      permalink_title: \"Anchor link to this section\"\n      baselevel: 1\n      toc_depth: 3\n      slugify: !!python/object/apply:pymdownx.slugs.slugify\n        kwds:\n          case: lower\n

Benefits: Better navigation within long pages, SEO-friendly anchor links.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#12-image-optimization--medium-priority","title":"12. Image Optimization \u2b50 Medium Priority","text":"

Add image handling plugin:

plugins:\n  - search\n  - mermaid2\n  - codeinclude\n  - minify:\n      minify_html: true\n      minify_js: true\n      minify_css: true\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n  # Optional: Image optimization\n  # - awesome-pages  # For better page organization\n

Benefits: Faster page loads, better mobile experience.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#13-keyboard-shortcuts--low-priority","title":"13. Keyboard Shortcuts \u2b50 Low Priority","text":"

Enable keyboard navigation:

theme:\n  keyboard_shortcuts:\n    search: true\n    previous: true\n    next: true\n

Benefits: Power users can navigate faster.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#14-print-styles--low-priority","title":"14. Print Styles \u2b50 Low Priority","text":"

Better printing experience:

theme:\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.indexes\n    - navigation.instant\n    - navigation.tracking\n    - navigation.smooth\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link\n    - content.tooltips\n    - content.action.edit  # Edit button\n    - content.action.view  # View source\n

Benefits: Users can print documentation cleanly.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#15-better-search-configuration--medium-priority","title":"15. Better Search Configuration \u2b50 Medium Priority","text":"

Enhance search capabilities:

plugins:\n  - search:\n      lang:\n        - en\n      separator: '[\\s\\-,:!=\\[\\]()\"`/]+|\\.(?!\\d)|&[lg]t;|&'\n      prebuild_index: true  # For faster search\n      indexing: full  # Full-text indexing\n

Benefits: Faster, more accurate search results.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#16-api-documentation-enhancements--high-priority-for-your-api-docs","title":"16. API Documentation Enhancements \u2b50 High Priority (for your API docs)","text":"

Since you have extensive API documentation, consider:

markdown_extensions:\n  - pymdownx.superfences:\n      custom_fences:\n        - name: mermaid\n          class: mermaid\n          format: !!python/name:pymdownx.superfences.fence_code_format\n      preserve_tabs: true\n  # Add API-specific features\n  - attr_list\n  - md_in_html\n  - pymdownx.caret\n  - pymdownx.tilde\n

Benefits: Better formatting for API endpoints, parameters, responses.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#17-blognews-section--low-priority-if-needed","title":"17. Blog/News Section \u2b50 Low Priority (if needed)","text":"

If you want to add a blog:

plugins:\n  - blog:\n      blog_dir: blog\n      blog_description: \"News and updates\"\n      post_date_format: full\n      post_url_format: '{slug}'\n      archive: true\n

Benefits: Keep users updated with changelog, announcements.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#18-tags-and-categories--low-priority","title":"18. Tags and Categories \u2b50 Low Priority","text":"

Organize content with tags:

markdown_extensions:\n  - meta\n

Then in frontmatter: 

---\ntags:\n  - api\n  - agents\n  - getting-started\n---\n

Benefits: Better content organization, related content discovery.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#19-better-mobile-experience--high-priority","title":"19. Better Mobile Experience \u2b50 High Priority","text":"

Ensure mobile optimization:

theme:\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.instant  # Helps on mobile\n    - navigation.tracking\n    - navigation.smooth\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link\n    - content.tooltips\n    - toc.integrate  # Better mobile TOC\n

Benefits: Better experience for mobile users (growing segment).

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#20-feedback-mechanism--medium-priority","title":"20. Feedback Mechanism \u2b50 Medium Priority","text":"

Add feedback buttons:

extra:\n  feedback:\n    title: \"Was this page helpful?\"\n    ratings:\n      - icon: material/thumb-up-outline\n        name: \"This page was helpful\"\n      - icon: material/thumb-down-outline\n        name: \"This page could be improved\"\n

Benefits: Understand what content needs improvement.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#priority-recommendations","title":"Priority Recommendations","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#immediate-high-impact-easy-implementation","title":"Immediate (High Impact, Easy Implementation)","text":"
  1. \u2705 Git revision dates - Shows content freshness
  2. \u2705 Enhanced navigation features - Better UX
  3. \u2705 Content tabs - Perfect for code examples
  4. \u2705 Better search configuration - Faster search
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#short-term-high-impact-medium-effort","title":"Short-term (High Impact, Medium Effort)","text":"
  1. \u2705 API documentation enhancements - Better API docs
  2. \u2705 Enhanced admonitions - Better visual hierarchy
  3. \u2705 Mobile optimization - Better mobile experience
  4. \u2705 Analytics - Understand user behavior
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#long-term-nice-to-have","title":"Long-term (Nice to Have)","text":"
  1. \u26a0\ufe0f Versioning - If you need multiple versions
  2. \u26a0\ufe0f Math formulas - If you have mathematical content
  3. \u26a0\ufe0f Blog section - If you want to publish updates
  4. \u26a0\ufe0f Custom CSS/JS - For advanced customization
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#implementation-example","title":"Implementation Example","text":"

Here's an enhanced mkdocs.yml with the high-priority improvements:

site_name: The DETERMINATOR\nsite_description: Generalist Deep Research Agent that Stops at Nothing\nsite_author: The DETERMINATOR Team\nsite_url: https://deepcritical.github.io/GradioDemo/\n\nrepo_name: DeepCritical/GradioDemo\nrepo_url: https://github.com/DeepCritical/GradioDemo\nedit_uri: edit/dev/docs/\n\nstrict: false\n\ntheme:\n  name: material\n  palette:\n    - scheme: default\n      primary: orange\n      accent: red\n      toggle:\n        icon: material/brightness-7\n        name: Switch to dark mode\n    - scheme: slate\n      primary: orange\n      accent: red\n      toggle:\n        icon: material/brightness-4\n        name: Switch to light mode\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.indexes\n    - navigation.instant\n    - navigation.tracking\n    - navigation.smooth\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link\n    - content.tooltips\n    - toc.integrate\n  icon:\n    repo: fontawesome/brands/github\n  language: en\n\nplugins:\n  - search:\n      lang:\n        - en\n      separator: '[\\s\\-,:!=\\[\\]()\"`/]+|\\.(?!\\d)|&[lg]t;|&'\n      prebuild_index: true\n      indexing: full\n  - mermaid2\n  - codeinclude\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n      fallback_to_build_date: true\n  - minify:\n      minify_html: true\n      minify_js: true\n      minify_css: true\n\nmarkdown_extensions:\n  - dev.docs_plugins:\n      base_path: \".\"\n  - pymdownx.highlight:\n      anchor_linenums: true\n      line_spans: __span\n      pygments_lang_class: true\n      use_pygments: true\n      noclasses: false\n  - pymdownx.inlinehilite\n  - pymdownx.superfences:\n      custom_fences:\n        - name: mermaid\n          class: mermaid\n          format: !!python/name:pymdownx.superfences.fence_code_format\n      preserve_tabs: true\n  - pymdownx.tabbed:\n      alternate_style: true\n      combine_header_slug: true\n  - pymdownx.tasklist:\n      custom_checkbox: true\n  - pymdownx.emoji:\n      emoji_generator: !!python/name:pymdownx.emoji.to_svg\n      emoji_index: !!python/name:pymdownx.emoji.twemoji\n  - pymdownx.snippets\n  - admonition\n  - pymdownx.details\n  - attr_list\n  - md_in_html\n  - tables\n  - meta\n  - toc:\n      permalink: true\n      permalink_title: \"Anchor link to this section\"\n      baselevel: 1\n      toc_depth: 3\n      slugify: !!python/object/apply:pymdownx.slugs.slugify\n        kwds:\n          case: lower\n\nnav:\n  - Home: index.md\n  - Overview:\n    - overview/architecture.md\n    - overview/features.md\n  - Getting Started:\n    - getting-started/installation.md\n    - getting-started/quick-start.md\n    - getting-started/mcp-integration.md\n    - getting-started/examples.md\n  - Configuration:\n    - configuration/index.md\n  - Architecture:\n    - \"Graph Orchestration\": architecture/graph_orchestration.md\n    - \"Workflow Diagrams\": architecture/workflow-diagrams.md\n    - \"Agents\": architecture/agents.md\n    - \"Orchestrators\": architecture/orchestrators.md\n    - \"Tools\": architecture/tools.md\n    - \"Middleware\": architecture/middleware.md\n    - \"Services\": architecture/services.md\n  - API Reference:\n    - api/agents.md\n    - api/tools.md\n    - api/orchestrators.md\n    - api/services.md\n    - api/models.md\n  - Contributing:\n    - contributing/index.md\n    - contributing/code-quality.md\n    - contributing/code-style.md\n    - contributing/error-handling.md\n    - contributing/implementation-patterns.md\n    - contributing/prompt-engineering.md\n    - contributing/testing.md\n  - License: LICENSE.md\n  - Team: team.md\n\nextra:\n  social:\n    - icon: fontawesome/brands/github\n      link: https://github.com/DeepCritical/GradioDemo\n      name: GitHub\n    - icon: material/web\n      link: https://huggingface.co/spaces/DataQuests/DeepCritical\n      name: HuggingFace Space\n  version:\n    provider: mike\n  generator:\n    enabled: false\n\ncopyright: Copyright © 2024 DeepCritical Team\n
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#additional-documentation-improvements","title":"Additional Documentation Improvements","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#content-structure","title":"Content Structure","text":"
  1. Add a changelog page - Keep users informed of updates
  2. Add a FAQ section - Address common questions
  3. Add a glossary - Define technical terms
  4. Add a troubleshooting guide - Help users solve common issues
  5. Add video tutorials - Embed videos for complex topics
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#visual-enhancements","title":"Visual Enhancements","text":"
  1. Add diagrams - Use more Mermaid diagrams for complex flows
  2. Add screenshots - Visual guides for UI features
  3. Add code examples - More practical examples
  4. Add comparison tables - Compare different approaches/options
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#seo--discoverability","title":"SEO & Discoverability","text":"
  1. Add meta descriptions - Better search engine results
  2. Add Open Graph tags - Better social media sharing
  3. Add sitemap - Help search engines index your docs
  4. Add robots.txt - Control search engine crawling
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#next-steps","title":"Next Steps","text":"
  1. Review this assessment
  2. Prioritize features based on your needs
  3. Test changes in a branch
  4. Gather user feedback
  5. Iterate and improve
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#resources","title":"Resources","text":""},{"location":"team/","title":"Team","text":"

DeepCritical is developed by a team of researchers and developers working on AI-assisted research.

"},{"location":"team/#team-members","title":"Team Members","text":""},{"location":"team/#zj","title":"ZJ","text":""},{"location":"team/#mario-aderman","title":"Mario Aderman","text":""},{"location":"team/#joseph-pollack","title":"Joseph Pollack","text":""},{"location":"team/#virat-chauran","title":"Virat Chauran","text":""},{"location":"team/#anna-bossler","title":"Anna Bossler","text":""},{"location":"team/#about","title":"About","text":"

The DeepCritical team met online in the Alzheimer's Critical Literature Review Group in the Hugging Science initiative. We're building the agent framework we want to use for AI-assisted research to turn the vast amounts of clinical data into cures.

"},{"location":"team/#contributing","title":"Contributing","text":"

We welcome contributions! See the Contributing Guide for details.

"},{"location":"team/#links","title":"Links","text":""},{"location":"api/agents/","title":"Agents API Reference","text":"

This page documents the API for DeepCritical agents.

"},{"location":"api/agents/#knowledgegapagent","title":"KnowledgeGapAgent","text":"

Module: src.agents.knowledge_gap

Purpose: Evaluates research state and identifies knowledge gaps.

"},{"location":"api/agents/#methods","title":"Methods","text":""},{"location":"api/agents/#evaluate","title":"evaluate","text":"

Evaluates research completeness and identifies outstanding knowledge gaps.

Parameters: - query: Research query string - background_context: Background context for the query (default: \"\") - conversation_history: History of actions, findings, and thoughts as string (default: \"\") - iteration: Current iteration number (default: 0) - time_elapsed_minutes: Elapsed time in minutes (default: 0.0) - max_time_minutes: Maximum time limit in minutes (default: 10)

Returns: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

"},{"location":"api/agents/#toolselectoragent","title":"ToolSelectorAgent","text":"

Module: src.agents.tool_selector

Purpose: Selects appropriate tools for addressing knowledge gaps.

"},{"location":"api/agents/#methods_1","title":"Methods","text":""},{"location":"api/agents/#select_tools","title":"select_tools","text":"

Selects tools for addressing a knowledge gap.

Parameters: - gap: The knowledge gap to address - query: Research query string - background_context: Optional background context (default: \"\") - conversation_history: History of actions, findings, and thoughts as string (default: \"\")

Returns: AgentSelectionPlan with list of AgentTask objects.

"},{"location":"api/agents/#writeragent","title":"WriterAgent","text":"

Module: src.agents.writer

Purpose: Generates final reports from research findings.

"},{"location":"api/agents/#methods_2","title":"Methods","text":""},{"location":"api/agents/#write_report","title":"write_report","text":"

Generates a markdown report from research findings.

Parameters: - query: Research query string - findings: Research findings to include in report - output_length: Optional description of desired output length (default: \"\") - output_instructions: Optional additional instructions for report generation (default: \"\")

Returns: Markdown string with numbered citations.

"},{"location":"api/agents/#longwriteragent","title":"LongWriterAgent","text":"

Module: src.agents.long_writer

Purpose: Long-form report generation with section-by-section writing.

"},{"location":"api/agents/#methods_3","title":"Methods","text":""},{"location":"api/agents/#write_next_section","title":"write_next_section","text":"

Writes the next section of a long-form report.

Parameters: - original_query: The original research query - report_draft: Current report draft as string (all sections written so far) - next_section_title: Title of the section to write - next_section_draft: Draft content for the next section

Returns: LongWriterOutput with formatted section and references.

"},{"location":"api/agents/#write_report_1","title":"write_report","text":"

Generates final report from draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Complete report draft

Returns: Final markdown report string.

"},{"location":"api/agents/#proofreaderagent","title":"ProofreaderAgent","text":"

Module: src.agents.proofreader

Purpose: Proofreads and polishes report drafts.

"},{"location":"api/agents/#methods_4","title":"Methods","text":""},{"location":"api/agents/#proofread","title":"proofread","text":"

Proofreads and polishes a report draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Report draft to proofread

Returns: Polished markdown string.

"},{"location":"api/agents/#thinkingagent","title":"ThinkingAgent","text":"

Module: src.agents.thinking

Purpose: Generates observations from conversation history.

"},{"location":"api/agents/#methods_5","title":"Methods","text":""},{"location":"api/agents/#generate_observations","title":"generate_observations","text":"

Generates observations from conversation history.

Parameters: - query: Research query string - background_context: Optional background context (default: \"\") - conversation_history: History of actions, findings, and thoughts as string (default: \"\") - iteration: Current iteration number (default: 1)

Returns: Observation string.

"},{"location":"api/agents/#inputparseragent","title":"InputParserAgent","text":"

Module: src.agents.input_parser

Purpose: Parses and improves user queries, detects research mode.

"},{"location":"api/agents/#methods_6","title":"Methods","text":""},{"location":"api/agents/#parse","title":"parse","text":"

Parses and improves a user query.

Parameters: - query: Original query string

Returns: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: \"iterative\" or \"deep\" - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"api/agents/#factory-functions","title":"Factory Functions","text":"

All agents have factory functions in src.agent_factory.agents:

Parameters: - model: Optional Pydantic AI model. If None, uses get_model() from settings. - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars)

Returns: Agent instance.

"},{"location":"api/agents/#see-also","title":"See Also","text":""},{"location":"api/models/","title":"Models API Reference","text":"

This page documents the Pydantic models used throughout DeepCritical.

"},{"location":"api/models/#evidence","title":"Evidence","text":"

Module: src.utils.models

Purpose: Represents evidence from search results.

Fields: - citation: Citation information (title, URL, date, authors) - content: Evidence text content - relevance: Relevance score (0.0-1.0) - metadata: Additional metadata dictionary

"},{"location":"api/models/#citation","title":"Citation","text":"

Module: src.utils.models

Purpose: Citation information for evidence.

Fields: - source: Source name (e.g., \"pubmed\", \"clinicaltrials\", \"europepmc\", \"web\", \"rag\") - title: Article/trial title - url: Source URL - date: Publication date (YYYY-MM-DD or \"Unknown\") - authors: List of authors (optional)

"},{"location":"api/models/#knowledgegapoutput","title":"KnowledgeGapOutput","text":"

Module: src.utils.models

Purpose: Output from knowledge gap evaluation.

Fields: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

"},{"location":"api/models/#agentselectionplan","title":"AgentSelectionPlan","text":"

Module: src.utils.models

Purpose: Plan for tool/agent selection.

Fields: - tasks: List of agent tasks to execute

"},{"location":"api/models/#agenttask","title":"AgentTask","text":"

Module: src.utils.models

Purpose: Individual agent task.

Fields: - gap: The knowledge gap being addressed (optional) - agent: Name of agent to use - query: The specific query for the agent - entity_website: The website of the entity being researched, if known (optional)

"},{"location":"api/models/#reportdraft","title":"ReportDraft","text":"

Module: src.utils.models

Purpose: Draft structure for long-form reports.

Fields: - sections: List of report sections

"},{"location":"api/models/#reportsection","title":"ReportSection","text":"

Module: src.utils.models

Purpose: Individual section in a report draft.

Fields: - section_title: The title of the section - section_content: The content of the section

"},{"location":"api/models/#parsedquery","title":"ParsedQuery","text":"

Module: src.utils.models

Purpose: Parsed and improved query.

Fields: - original_query: Original query string - improved_query: Refined query string - research_mode: Research mode (\"iterative\" or \"deep\") - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"api/models/#conversation","title":"Conversation","text":"

Module: src.utils.models

Purpose: Conversation history with iterations.

Fields: - history: List of iteration data

"},{"location":"api/models/#iterationdata","title":"IterationData","text":"

Module: src.utils.models

Purpose: Data for a single iteration.

Fields: - gap: The gap addressed in the iteration - tool_calls: The tool calls made - findings: The findings collected from tool calls - thought: The thinking done to reflect on the success of the iteration and next steps

"},{"location":"api/models/#agentevent","title":"AgentEvent","text":"

Module: src.utils.models

Purpose: Event emitted during research execution.

Fields: - type: Event type (e.g., \"started\", \"search_complete\", \"complete\") - iteration: Iteration number (optional) - data: Event data dictionary

"},{"location":"api/models/#budgetstatus","title":"BudgetStatus","text":"

Module: src.utils.models

Purpose: Current budget status.

Fields: - tokens_used: Total tokens used - tokens_limit: Token budget limit - time_elapsed_seconds: Time elapsed in seconds - time_limit_seconds: Time budget limit (default: 600.0 seconds / 10 minutes) - iterations: Number of iterations completed - iterations_limit: Maximum iterations (default: 10) - iteration_tokens: Tokens used per iteration (iteration number -> token count)

"},{"location":"api/models/#see-also","title":"See Also","text":""},{"location":"api/orchestrators/","title":"Orchestrators API Reference","text":"

This page documents the API for DeepCritical orchestrators.

"},{"location":"api/orchestrators/#iterativeresearchflow","title":"IterativeResearchFlow","text":"

Module: src.orchestrator.research_flow

Purpose: Single-loop research with search-judge-synthesize cycles.

"},{"location":"api/orchestrators/#methods","title":"Methods","text":""},{"location":"api/orchestrators/#run","title":"run","text":"

Runs iterative research flow.

Parameters: - query: Research query string - background_context: Background context (default: \"\") - output_length: Optional description of desired output length (default: \"\") - output_instructions: Optional additional instructions for report generation (default: \"\")

Returns: Final report string.

Note: max_iterations, max_time_minutes, and token_budget are constructor parameters, not run() parameters.

"},{"location":"api/orchestrators/#deepresearchflow","title":"DeepResearchFlow","text":"

Module: src.orchestrator.research_flow

Purpose: Multi-section parallel research with planning and synthesis.

"},{"location":"api/orchestrators/#methods_1","title":"Methods","text":""},{"location":"api/orchestrators/#run_1","title":"run","text":"

Runs deep research flow.

Parameters: - query: Research query string

Returns: Final report string.

Note: max_iterations_per_section, max_time_minutes, and token_budget are constructor parameters, not run() parameters.

"},{"location":"api/orchestrators/#graphorchestrator","title":"GraphOrchestrator","text":"

Module: src.orchestrator.graph_orchestrator

Purpose: Graph-based execution using Pydantic AI agents as nodes.

"},{"location":"api/orchestrators/#methods_2","title":"Methods","text":""},{"location":"api/orchestrators/#run_2","title":"run","text":"

Runs graph-based research orchestration.

Parameters: - query: Research query string

Yields: AgentEvent objects during graph execution.

Note: research_mode and use_graph are constructor parameters, not run() parameters.

"},{"location":"api/orchestrators/#orchestrator-factory","title":"Orchestrator Factory","text":"

Module: src.orchestrator_factory

Purpose: Factory for creating orchestrators.

"},{"location":"api/orchestrators/#functions","title":"Functions","text":""},{"location":"api/orchestrators/#create_orchestrator","title":"create_orchestrator","text":"

Creates an orchestrator instance.

Parameters: - search_handler: Search handler protocol implementation (optional, required for simple mode) - judge_handler: Judge handler protocol implementation (optional, required for simple mode) - config: Configuration object (optional) - mode: Orchestrator mode (\"simple\", \"advanced\", \"magentic\", \"iterative\", \"deep\", \"auto\", or None for auto-detect) - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars)

Returns: Orchestrator instance.

Raises: - ValueError: If requirements not met

Modes: - \"simple\": Legacy orchestrator - \"advanced\" or \"magentic\": Magentic orchestrator (requires OpenAI API key) - None: Auto-detect based on API key availability

"},{"location":"api/orchestrators/#magenticorchestrator","title":"MagenticOrchestrator","text":"

Module: src.orchestrator_magentic

Purpose: Multi-agent coordination using Microsoft Agent Framework.

"},{"location":"api/orchestrators/#methods_3","title":"Methods","text":""},{"location":"api/orchestrators/#run_3","title":"run","text":"

Runs Magentic orchestration.

Parameters: - query: Research query string

Yields: AgentEvent objects converted from Magentic events.

Note: max_rounds and max_stalls are constructor parameters, not run() parameters.

Requirements: - agent-framework-core package - OpenAI API key

"},{"location":"api/orchestrators/#see-also","title":"See Also","text":""},{"location":"api/services/","title":"Services API Reference","text":"

This page documents the API for DeepCritical services.

"},{"location":"api/services/#embeddingservice","title":"EmbeddingService","text":"

Module: src.services.embeddings

Purpose: Local sentence-transformers for semantic search and deduplication.

"},{"location":"api/services/#methods","title":"Methods","text":""},{"location":"api/services/#embed","title":"embed","text":"

Generates embedding for a text string.

Parameters: - text: Text to embed

Returns: Embedding vector as list of floats.

"},{"location":"api/services/#embed_batch","title":"embed_batch","text":"
async def embed_batch(self, texts: list[str]) -> list[list[float]]\n

Generates embeddings for multiple texts.

Parameters: - texts: List of texts to embed

Returns: List of embedding vectors.

"},{"location":"api/services/#similarity","title":"similarity","text":"
async def similarity(self, text1: str, text2: str) -> float\n

Calculates similarity between two texts.

Parameters: - text1: First text - text2: Second text

Returns: Similarity score (0.0-1.0).

"},{"location":"api/services/#find_duplicates","title":"find_duplicates","text":"
async def find_duplicates(\n    self,\n    texts: list[str],\n    threshold: float = 0.85\n) -> list[tuple[int, int]]\n

Finds duplicate texts based on similarity threshold.

Parameters: - texts: List of texts to check - threshold: Similarity threshold (default: 0.85)

Returns: List of (index1, index2) tuples for duplicate pairs.

"},{"location":"api/services/#add_evidence","title":"add_evidence","text":"
async def add_evidence(\n    self,\n    evidence_id: str,\n    content: str,\n    metadata: dict[str, Any]\n) -> None\n

Adds evidence to vector store for semantic search.

Parameters: - evidence_id: Unique identifier for the evidence - content: Evidence text content - metadata: Additional metadata dictionary

"},{"location":"api/services/#search_similar","title":"search_similar","text":"
async def search_similar(\n    self,\n    query: str,\n    n_results: int = 5\n) -> list[dict[str, Any]]\n

Finds semantically similar evidence.

Parameters: - query: Search query string - n_results: Number of results to return (default: 5)

Returns: List of dictionaries with id, content, metadata, and distance keys.

"},{"location":"api/services/#deduplicate","title":"deduplicate","text":"
async def deduplicate(\n    self,\n    new_evidence: list[Evidence],\n    threshold: float = 0.9\n) -> list[Evidence]\n

Removes semantically duplicate evidence.

Parameters: - new_evidence: List of evidence items to deduplicate - threshold: Similarity threshold (default: 0.9, where 0.9 = 90% similar is duplicate)

Returns: List of unique evidence items (not already in vector store).

"},{"location":"api/services/#factory-function","title":"Factory Function","text":""},{"location":"api/services/#get_embedding_service","title":"get_embedding_service","text":"
@lru_cache(maxsize=1)\ndef get_embedding_service() -> EmbeddingService\n

Returns singleton EmbeddingService instance.

"},{"location":"api/services/#llamaindexragservice","title":"LlamaIndexRAGService","text":"

Module: src.services.rag

Purpose: Retrieval-Augmented Generation using LlamaIndex.

"},{"location":"api/services/#methods_1","title":"Methods","text":""},{"location":"api/services/#ingest_evidence","title":"ingest_evidence","text":"

Ingests evidence into RAG service.

Parameters: - evidence_list: List of Evidence objects to ingest

Note: Supports multiple embedding providers (OpenAI, local sentence-transformers, Hugging Face).

"},{"location":"api/services/#retrieve","title":"retrieve","text":"
def retrieve(\n    self,\n    query: str,\n    top_k: int | None = None\n) -> list[dict[str, Any]]\n

Retrieves relevant documents for a query.

Parameters: - query: Search query string - top_k: Number of top results to return (defaults to similarity_top_k from constructor)

Returns: List of dictionaries with text, score, and metadata keys.

"},{"location":"api/services/#query","title":"query","text":"
def query(\n    self,\n    query_str: str,\n    top_k: int | None = None\n) -> str\n

Queries RAG service and returns synthesized response.

Parameters: - query_str: Query string - top_k: Number of results to use (defaults to similarity_top_k from constructor)

Returns: Synthesized response string.

Raises: - ConfigurationError: If no LLM API key is available for query synthesis

"},{"location":"api/services/#ingest_documents","title":"ingest_documents","text":"
def ingest_documents(self, documents: list[Any]) -> None\n

Ingests raw LlamaIndex Documents.

Parameters: - documents: List of LlamaIndex Document objects

"},{"location":"api/services/#clear_collection","title":"clear_collection","text":"
def clear_collection(self) -> None\n

Clears all documents from the collection.

"},{"location":"api/services/#factory-function_1","title":"Factory Function","text":""},{"location":"api/services/#get_rag_service","title":"get_rag_service","text":"
def get_rag_service(\n    collection_name: str = \"deepcritical_evidence\",\n    oauth_token: str | None = None,\n    **kwargs: Any\n) -> LlamaIndexRAGService\n

Get or create a RAG service instance.

Parameters: - collection_name: Name of the ChromaDB collection (default: \"deepcritical_evidence\") - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars) - **kwargs: Additional arguments for LlamaIndexRAGService (e.g., use_openai_embeddings=False)

Returns: Configured LlamaIndexRAGService instance.

Note: By default, uses local embeddings (sentence-transformers) which require no API keys.

"},{"location":"api/services/#statisticalanalyzer","title":"StatisticalAnalyzer","text":"

Module: src.services.statistical_analyzer

Purpose: Secure execution of AI-generated statistical code.

"},{"location":"api/services/#methods_2","title":"Methods","text":""},{"location":"api/services/#analyze","title":"analyze","text":"
async def analyze(\n    self,\n    query: str,\n    evidence: list[Evidence],\n    hypothesis: dict[str, Any] | None = None\n) -> AnalysisResult\n

Analyzes a research question using statistical methods.

Parameters: - query: The research question - evidence: List of Evidence objects to analyze - hypothesis: Optional hypothesis dict with drug, target, pathway, effect, confidence keys

Returns: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - confidence: Confidence in verdict (0.0-1.0) - statistical_evidence: Summary of statistical findings - code_generated: Python code that was executed - execution_output: Output from code execution - key_takeaways: Key takeaways from analysis - limitations: List of limitations

Note: Requires Modal credentials for sandbox execution.

"},{"location":"api/services/#see-also","title":"See Also","text":""},{"location":"api/tools/","title":"Tools API Reference","text":"

This page documents the API for DeepCritical search tools.

"},{"location":"api/tools/#searchtool-protocol","title":"SearchTool Protocol","text":"

All tools implement the SearchTool protocol:

class SearchTool(Protocol):\n    @property\n    def name(self) -> str: ...\n    \n    async def search(\n        self, \n        query: str, \n        max_results: int = 10\n    ) -> list[Evidence]: ...\n
"},{"location":"api/tools/#pubmedtool","title":"PubMedTool","text":"

Module: src.tools.pubmed

Purpose: Search peer-reviewed biomedical literature from PubMed.

"},{"location":"api/tools/#properties","title":"Properties","text":""},{"location":"api/tools/#name","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"pubmed\"

"},{"location":"api/tools/#methods","title":"Methods","text":""},{"location":"api/tools/#search","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches PubMed for articles.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with PubMed articles.

Raises: - SearchError: If search fails (timeout, HTTP error, XML parsing error) - RateLimitError: If rate limit is exceeded (429 status code)

Note: Uses NCBI E-utilities (ESearch \u2192 EFetch). Rate limit: 0.34s between requests. Handles single vs. multiple articles.

"},{"location":"api/tools/#clinicaltrialstool","title":"ClinicalTrialsTool","text":"

Module: src.tools.clinicaltrials

Purpose: Search ClinicalTrials.gov for interventional studies.

"},{"location":"api/tools/#properties_1","title":"Properties","text":""},{"location":"api/tools/#name_1","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"clinicaltrials\"

"},{"location":"api/tools/#methods_1","title":"Methods","text":""},{"location":"api/tools/#search_1","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches ClinicalTrials.gov for trials.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with clinical trials.

Note: Only returns interventional studies with status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION. Uses requests library (NOT httpx - WAF blocks httpx). Runs in thread pool for async compatibility.

Raises: - SearchError: If search fails (HTTP error, request exception)

"},{"location":"api/tools/#europepmctool","title":"EuropePMCTool","text":"

Module: src.tools.europepmc

Purpose: Search Europe PMC for preprints and peer-reviewed articles.

"},{"location":"api/tools/#properties_2","title":"Properties","text":""},{"location":"api/tools/#name_2","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"europepmc\"

"},{"location":"api/tools/#methods_2","title":"Methods","text":""},{"location":"api/tools/#search_2","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches Europe PMC for articles and preprints.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with articles/preprints.

Note: Includes both preprints (marked with [PREPRINT - Not peer-reviewed]) and peer-reviewed articles. Handles preprint markers. Builds URLs from DOI or PMID.

Raises: - SearchError: If search fails (HTTP error, connection error)

"},{"location":"api/tools/#ragtool","title":"RAGTool","text":"

Module: src.tools.rag_tool

Purpose: Semantic search within collected evidence.

"},{"location":"api/tools/#initialization","title":"Initialization","text":"
def __init__(\n    self,\n    rag_service: LlamaIndexRAGService | None = None,\n    oauth_token: str | None = None\n) -> None\n

Parameters: - rag_service: Optional RAG service instance. If None, will be lazy-initialized. - oauth_token: Optional OAuth token from HuggingFace login (for RAG LLM)

"},{"location":"api/tools/#properties_3","title":"Properties","text":""},{"location":"api/tools/#name_3","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"rag\"

"},{"location":"api/tools/#methods_3","title":"Methods","text":""},{"location":"api/tools/#search_3","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches collected evidence using semantic similarity.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects from collected evidence.

Raises: - ConfigurationError: If RAG service is unavailable

Note: Requires evidence to be ingested into RAG service first. Wraps LlamaIndexRAGService. Returns Evidence from RAG results.

"},{"location":"api/tools/#searchhandler","title":"SearchHandler","text":"

Module: src.tools.search_handler

Purpose: Orchestrates parallel searches across multiple tools.

"},{"location":"api/tools/#initialization_1","title":"Initialization","text":"
def __init__(\n    self,\n    tools: list[SearchTool],\n    timeout: float = 30.0,\n    include_rag: bool = False,\n    auto_ingest_to_rag: bool = True,\n    oauth_token: str | None = None\n) -> None\n

Parameters: - tools: List of search tools to use - timeout: Timeout for each search in seconds (default: 30.0) - include_rag: Whether to include RAG tool in searches (default: False) - auto_ingest_to_rag: Whether to automatically ingest results into RAG (default: True) - oauth_token: Optional OAuth token from HuggingFace login (for RAG LLM)

"},{"location":"api/tools/#methods_4","title":"Methods","text":""},{"location":"api/tools/#execute","title":"execute","text":"

Searches multiple tools in parallel.

Parameters: - query: Search query string - max_results_per_tool: Maximum results per tool (default: 10)

Returns: SearchResult with: - query: The search query - evidence: Aggregated list of evidence - sources_searched: List of source names searched - total_found: Total number of results - errors: List of error messages from failed tools

Raises: - SearchError: If search times out

Note: Uses asyncio.gather() for parallel execution. Handles tool failures gracefully (returns errors in SearchResult.errors). Automatically ingests evidence into RAG if enabled.

"},{"location":"api/tools/#see-also","title":"See Also","text":""},{"location":"architecture/agents/","title":"Agents Architecture","text":"

DeepCritical uses Pydantic AI agents for all AI-powered operations. All agents follow a consistent pattern and use structured output types.

"},{"location":"architecture/agents/#agent-pattern","title":"Agent Pattern","text":""},{"location":"architecture/agents/#pydantic-ai-agents","title":"Pydantic AI Agents","text":"

Pydantic AI agents use the Agent class with the following structure:

Note: Factory functions accept an optional oauth_token parameter for HuggingFace authentication, which takes priority over environment variables.

"},{"location":"architecture/agents/#model-initialization","title":"Model Initialization","text":"

Agents use get_model() from src/agent_factory/judges.py if no model is provided. This supports:

The model selection is based on the configured LLM_PROVIDER in settings.

"},{"location":"architecture/agents/#error-handling","title":"Error Handling","text":"

Agents return fallback values on failure rather than raising exceptions:

All errors are logged with context using structlog.

"},{"location":"architecture/agents/#input-validation","title":"Input Validation","text":"

All agents validate inputs:

"},{"location":"architecture/agents/#output-types","title":"Output Types","text":"

Agents use structured output types from src/utils/models.py:

For text output (writer agents), agents return str directly.

"},{"location":"architecture/agents/#agent-types","title":"Agent Types","text":""},{"location":"architecture/agents/#knowledge-gap-agent","title":"Knowledge Gap Agent","text":"

File: src/agents/knowledge_gap.py

Purpose: Evaluates research state and identifies knowledge gaps.

Output: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

Methods: - async def evaluate(query, background_context, conversation_history, iteration, time_elapsed_minutes, max_time_minutes) -> KnowledgeGapOutput

"},{"location":"architecture/agents/#tool-selector-agent","title":"Tool Selector Agent","text":"

File: src/agents/tool_selector.py

Purpose: Selects appropriate tools for addressing knowledge gaps.

Output: AgentSelectionPlan with list of AgentTask objects.

Available Agents: - WebSearchAgent: General web search for fresh information - SiteCrawlerAgent: Research specific entities/companies - RAGAgent: Semantic search within collected evidence

"},{"location":"architecture/agents/#writer-agent","title":"Writer Agent","text":"

File: src/agents/writer.py

Purpose: Generates final reports from research findings.

Output: Markdown string with numbered citations.

Methods: - async def write_report(query, findings, output_length, output_instructions) -> str

Features: - Validates inputs - Truncates very long findings (max 50000 chars) with warning - Retry logic for transient failures (3 retries) - Citation validation before returning

"},{"location":"architecture/agents/#long-writer-agent","title":"Long Writer Agent","text":"

File: src/agents/long_writer.py

Purpose: Long-form report generation with section-by-section writing.

Input/Output: Uses ReportDraft models.

Methods: - async def write_next_section(query, draft, section_title, section_content) -> LongWriterOutput - async def write_report(query, report_title, report_draft) -> str

Features: - Writes sections iteratively - Aggregates references across sections - Reformats section headings and references - Deduplicates and renumbers references

"},{"location":"architecture/agents/#proofreader-agent","title":"Proofreader Agent","text":"

File: src/agents/proofreader.py

Purpose: Proofreads and polishes report drafts.

Input: ReportDraft Output: Polished markdown string

Methods: - async def proofread(query, report_title, report_draft) -> str

Features: - Removes duplicate content across sections - Adds executive summary if multiple sections - Preserves all references and citations - Improves flow and readability

"},{"location":"architecture/agents/#thinking-agent","title":"Thinking Agent","text":"

File: src/agents/thinking.py

Purpose: Generates observations from conversation history.

Output: Observation string

Methods: - async def generate_observations(query, background_context, conversation_history) -> str

"},{"location":"architecture/agents/#input-parser-agent","title":"Input Parser Agent","text":"

File: src/agents/input_parser.py

Purpose: Parses and improves user queries, detects research mode.

Output: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: \"iterative\" or \"deep\" - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"architecture/agents/#magentic-agents","title":"Magentic Agents","text":"

The following agents use the BaseAgent pattern from agent-framework and are used exclusively with MagenticOrchestrator:

"},{"location":"architecture/agents/#hypothesis-agent","title":"Hypothesis Agent","text":"

File: src/agents/hypothesis_agent.py

Purpose: Generates mechanistic hypotheses based on evidence.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Uses internal Pydantic AI Agent with HypothesisAssessment output type - Accesses shared evidence_store for evidence - Uses embedding service for diverse evidence selection (MMR algorithm) - Stores hypotheses in shared context

"},{"location":"architecture/agents/#search-agent","title":"Search Agent","text":"

File: src/agents/search_agent.py

Purpose: Wraps SearchHandler as an agent for Magentic orchestrator.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Executes searches via SearchHandlerProtocol - Deduplicates evidence using embedding service - Searches for semantically related evidence - Updates shared evidence store

"},{"location":"architecture/agents/#analysis-agent","title":"Analysis Agent","text":"

File: src/agents/analysis_agent.py

Purpose: Performs statistical analysis using Modal sandbox.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Wraps StatisticalAnalyzer service - Analyzes evidence and hypotheses - Returns verdict (SUPPORTED/REFUTED/INCONCLUSIVE) - Stores analysis results in shared context

"},{"location":"architecture/agents/#report-agent-magentic","title":"Report Agent (Magentic)","text":"

File: src/agents/report_agent.py

Purpose: Generates structured scientific reports from evidence and hypotheses.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Uses internal Pydantic AI Agent with ResearchReport output type - Accesses shared evidence store and hypotheses - Validates citations before returning - Formats report as markdown

"},{"location":"architecture/agents/#judge-agent","title":"Judge Agent","text":"

File: src/agents/judge_agent.py

Purpose: Evaluates evidence quality and determines if sufficient for synthesis.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse - async def run_stream(messages, thread, **kwargs) -> AsyncIterable[AgentRunResponseUpdate]

Features: - Wraps JudgeHandlerProtocol - Accesses shared evidence store - Returns JudgeAssessment with sufficient flag, confidence, and recommendation

"},{"location":"architecture/agents/#agent-patterns","title":"Agent Patterns","text":"

DeepCritical uses two distinct agent patterns:

"},{"location":"architecture/agents/#1-pydantic-ai-agents-traditional-pattern","title":"1. Pydantic AI Agents (Traditional Pattern)","text":"

These agents use the Pydantic AI Agent class directly and are used in iterative and deep research flows:

"},{"location":"architecture/agents/#2-magentic-agents-agent-framework-pattern","title":"2. Magentic Agents (Agent-Framework Pattern)","text":"

These agents use the BaseAgent class from agent-framework and are used in Magentic orchestrator:

Note: Magentic agents are used exclusively with the MagenticOrchestrator and follow the agent-framework protocol for multi-agent coordination.

"},{"location":"architecture/agents/#factory-functions","title":"Factory Functions","text":"

All agents have factory functions in src/agent_factory/agents.py:

Factory functions: - Use get_model() if no model provided - Accept oauth_token parameter for HuggingFace authentication - Raise ConfigurationError if creation fails - Log agent creation

"},{"location":"architecture/agents/#see-also","title":"See Also","text":""},{"location":"architecture/graph_orchestration/","title":"Graph Orchestration Architecture","text":""},{"location":"architecture/graph_orchestration/#overview","title":"Overview","text":"

DeepCritical implements a graph-based orchestration system for research workflows using Pydantic AI agents as nodes. This enables better parallel execution, conditional routing, and state management compared to simple agent chains.

"},{"location":"architecture/graph_orchestration/#graph-patterns","title":"Graph Patterns","text":""},{"location":"architecture/graph_orchestration/#iterative-research-graph","title":"Iterative Research Graph","text":"

The iterative research graph follows this pattern:

[Input] \u2192 [Thinking] \u2192 [Knowledge Gap] \u2192 [Decision: Complete?]\n                                              \u2193 No          \u2193 Yes\n                                    [Tool Selector]    [Writer]\n                                              \u2193\n                                    [Execute Tools] \u2192 [Loop Back]\n

Node IDs: thinking \u2192 knowledge_gap \u2192 continue_decision \u2192 tool_selector/writer \u2192 execute_tools \u2192 (loop back to thinking)

Special Node Handling: - execute_tools: State node that uses search_handler to execute searches and add evidence to workflow state - continue_decision: Decision node that routes based on research_complete flag from KnowledgeGapOutput

"},{"location":"architecture/graph_orchestration/#deep-research-graph","title":"Deep Research Graph","text":"

The deep research graph follows this pattern:

[Input] \u2192 [Planner] \u2192 [Store Plan] \u2192 [Parallel Loops] \u2192 [Collect Drafts] \u2192 [Synthesizer]\n                                        \u2193         \u2193         \u2193\n                                     [Loop1]  [Loop2]  [Loop3]\n

Node IDs: planner \u2192 store_plan \u2192 parallel_loops \u2192 collect_drafts \u2192 synthesizer

Special Node Handling: - planner: Agent node that creates ReportPlan with report outline - store_plan: State node that stores ReportPlan in context for parallel loops - parallel_loops: Parallel node that executes IterativeResearchFlow instances for each section - collect_drafts: State node that collects section drafts from parallel loops - synthesizer: Agent node that calls LongWriterAgent.write_report() directly with ReportDraft

"},{"location":"architecture/graph_orchestration/#deep-research","title":"Deep Research","text":"
\nsequenceDiagram\n    actor User\n    participant GraphOrchestrator\n    participant InputParser\n    participant GraphBuilder\n    participant GraphExecutor\n    participant Agent\n    participant BudgetTracker\n    participant WorkflowState\n\n    User->>GraphOrchestrator: run(query)\n    GraphOrchestrator->>InputParser: detect_research_mode(query)\n    InputParser-->>GraphOrchestrator: mode (iterative/deep)\n    GraphOrchestrator->>GraphBuilder: build_graph(mode)\n    GraphBuilder-->>GraphOrchestrator: ResearchGraph\n    GraphOrchestrator->>WorkflowState: init_workflow_state()\n    GraphOrchestrator->>BudgetTracker: create_budget()\n    GraphOrchestrator->>GraphExecutor: _execute_graph(graph)\n    \n    loop For each node in graph\n        GraphExecutor->>Agent: execute_node(agent_node)\n        Agent->>Agent: process_input\n        Agent-->>GraphExecutor: result\n        GraphExecutor->>WorkflowState: update_state(result)\n        GraphExecutor->>BudgetTracker: add_tokens(used)\n        GraphExecutor->>BudgetTracker: check_budget()\n        alt Budget exceeded\n            GraphExecutor->>GraphOrchestrator: emit(error_event)\n        else Continue\n            GraphExecutor->>GraphOrchestrator: emit(progress_event)\n        end\n    end\n    \n    GraphOrchestrator->>User: AsyncGenerator[AgentEvent]\n
"},{"location":"architecture/graph_orchestration/#iterative-research","title":"Iterative Research","text":"
sequenceDiagram\n    participant IterativeFlow\n    participant ThinkingAgent\n    participant KnowledgeGapAgent\n    participant ToolSelector\n    participant ToolExecutor\n    participant JudgeHandler\n    participant WriterAgent\n\n    IterativeFlow->>IterativeFlow: run(query)\n    \n    loop Until complete or max_iterations\n        IterativeFlow->>ThinkingAgent: generate_observations()\n        ThinkingAgent-->>IterativeFlow: observations\n        \n        IterativeFlow->>KnowledgeGapAgent: evaluate_gaps()\n        KnowledgeGapAgent-->>IterativeFlow: KnowledgeGapOutput\n        \n        alt Research complete\n            IterativeFlow->>WriterAgent: create_final_report()\n            WriterAgent-->>IterativeFlow: final_report\n        else Gaps remain\n            IterativeFlow->>ToolSelector: select_agents(gap)\n            ToolSelector-->>IterativeFlow: AgentSelectionPlan\n            \n            IterativeFlow->>ToolExecutor: execute_tool_tasks()\n            ToolExecutor-->>IterativeFlow: ToolAgentOutput[]\n            \n            IterativeFlow->>JudgeHandler: assess_evidence()\n            JudgeHandler-->>IterativeFlow: should_continue\n        end\n    end
"},{"location":"architecture/graph_orchestration/#graph-structure","title":"Graph Structure","text":""},{"location":"architecture/graph_orchestration/#nodes","title":"Nodes","text":"

Graph nodes represent different stages in the research workflow:

  1. Agent Nodes: Execute Pydantic AI agents
  2. Input: Prompt/query
  3. Output: Structured or unstructured response

  4. Examples: KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent

  5. State Nodes: Update or read workflow state

  6. Input: Current state
  7. Output: Updated state

  8. Examples: Update evidence, update conversation history

  9. Decision Nodes: Make routing decisions based on conditions

  10. Input: Current state/results
  11. Output: Next node ID

  12. Examples: Continue research vs. complete research

  13. Parallel Nodes: Execute multiple nodes concurrently

  14. Input: List of node IDs
  15. Output: Aggregated results
  16. Examples: Parallel iterative research loops
"},{"location":"architecture/graph_orchestration/#edges","title":"Edges","text":"

Edges define transitions between nodes:

  1. Sequential Edges: Always traversed (no condition)
  2. From: Source node
  3. To: Target node

  4. Condition: None (always True)

  5. Conditional Edges: Traversed based on condition

  6. From: Source node
  7. To: Target node
  8. Condition: Callable that returns bool

  9. Example: If research complete \u2192 go to writer, else \u2192 continue loop

  10. Parallel Edges: Used for parallel execution branches

  11. From: Parallel node
  12. To: Multiple target nodes
  13. Execution: All targets run concurrently
"},{"location":"architecture/graph_orchestration/#state-management","title":"State Management","text":"

State is managed via WorkflowState using ContextVar for thread-safe isolation:

State transitions occur at state nodes, which update the global workflow state.

"},{"location":"architecture/graph_orchestration/#execution-flow","title":"Execution Flow","text":"
  1. Graph Construction: Build graph from nodes and edges using create_iterative_graph() or create_deep_graph()
  2. Graph Validation: Ensure graph is valid (no cycles, all nodes reachable) via ResearchGraph.validate_structure()
  3. Graph Execution: Traverse graph from entry node using GraphOrchestrator._execute_graph()
  4. Node Execution: Execute each node based on type:
  5. Agent Nodes: Call agent.run() with transformed input
  6. State Nodes: Update workflow state via state_updater function
  7. Decision Nodes: Evaluate decision_function to get next node ID
  8. Parallel Nodes: Execute all parallel nodes concurrently via asyncio.gather()
  9. Edge Evaluation: Determine next node(s) based on edges and conditions
  10. Parallel Execution: Use asyncio.gather() for parallel nodes
  11. State Updates: Update state at state nodes via GraphExecutionContext.update_state()
  12. Event Streaming: Yield AgentEvent objects during execution for UI
"},{"location":"architecture/graph_orchestration/#graphexecutioncontext","title":"GraphExecutionContext","text":"

The GraphExecutionContext class manages execution state during graph traversal:

Methods: - set_node_result(node_id, result): Store result from node execution - get_node_result(node_id): Retrieve stored result - has_visited(node_id): Check if node was visited - mark_visited(node_id): Mark node as visited - update_state(updater, data): Update workflow state

"},{"location":"architecture/graph_orchestration/#conditional-routing","title":"Conditional Routing","text":"

Decision nodes evaluate conditions and return next node IDs:

"},{"location":"architecture/graph_orchestration/#parallel-execution","title":"Parallel Execution","text":"

Parallel nodes execute multiple nodes concurrently:

"},{"location":"architecture/graph_orchestration/#budget-enforcement","title":"Budget Enforcement","text":"

Budget constraints are enforced at decision nodes:

If any budget is exceeded, execution routes to exit node.

"},{"location":"architecture/graph_orchestration/#error-handling","title":"Error Handling","text":"

Errors are handled at multiple levels:

  1. Node Level: Catch errors in individual node execution
  2. Graph Level: Handle errors during graph traversal
  3. State Level: Rollback state changes on error

Errors are logged and yield error events for UI.

"},{"location":"architecture/graph_orchestration/#backward-compatibility","title":"Backward Compatibility","text":"

Graph execution is optional via feature flag:

This allows gradual migration and fallback if needed.

"},{"location":"architecture/graph_orchestration/#see-also","title":"See Also","text":""},{"location":"architecture/middleware/","title":"Middleware Architecture","text":"

DeepCritical uses middleware for state management, budget tracking, and workflow coordination.

"},{"location":"architecture/middleware/#state-management","title":"State Management","text":""},{"location":"architecture/middleware/#workflowstate","title":"WorkflowState","text":"

File: src/middleware/state_machine.py

Purpose: Thread-safe state management for research workflows

Implementation: Uses ContextVar for thread-safe isolation

State Components: - evidence: list[Evidence]: Collected evidence from searches - conversation: Conversation: Iteration history (gaps, tool calls, findings, thoughts) - embedding_service: Any: Embedding service for semantic search

Methods: - add_evidence(new_evidence: list[Evidence]) -> int: Adds evidence with URL-based deduplication. Returns the number of new items added (excluding duplicates). - async search_related(query: str, n_results: int = 5) -> list[Evidence]: Semantic search for related evidence using embedding service

Initialization:

Access:

"},{"location":"architecture/middleware/#workflow-manager","title":"Workflow Manager","text":"

File: src/middleware/workflow_manager.py

Purpose: Coordinates parallel research loops

Methods: - async add_loop(loop_id: str, query: str) -> ResearchLoop: Add a new research loop to manage - async run_loops_parallel(loop_configs: list[dict], loop_func: Callable, judge_handler: Any | None = None, budget_tracker: Any | None = None) -> list[Any]: Run multiple research loops in parallel. Takes configuration dicts and a loop function. - async update_loop_status(loop_id: str, status: LoopStatus, error: str | None = None): Update loop status - async sync_loop_evidence_to_state(loop_id: str): Synchronize evidence from a specific loop to global state

Features: - Uses asyncio.gather() for parallel execution - Handles errors per loop (doesn't fail all if one fails) - Tracks loop status: pending, running, completed, failed, cancelled - Evidence deduplication across parallel loops

Usage: 

from src.middleware.workflow_manager import WorkflowManager\n\nmanager = WorkflowManager()\nawait manager.add_loop(\"loop1\", \"Research query 1\")\nawait manager.add_loop(\"loop2\", \"Research query 2\")\n\nasync def run_research(config: dict) -> str:\n    loop_id = config[\"loop_id\"]\n    query = config[\"query\"]\n    # ... research logic ...\n    return \"report\"\n\nresults = await manager.run_loops_parallel(\n    loop_configs=[\n        {\"loop_id\": \"loop1\", \"query\": \"Research query 1\"},\n        {\"loop_id\": \"loop2\", \"query\": \"Research query 2\"},\n    ],\n    loop_func=run_research,\n)\n

"},{"location":"architecture/middleware/#budget-tracker","title":"Budget Tracker","text":"

File: src/middleware/budget_tracker.py

Purpose: Tracks and enforces resource limits

Budget Components: - Tokens: LLM token usage - Time: Elapsed time in seconds - Iterations: Number of iterations

Methods: - create_budget(loop_id: str, tokens_limit: int = 100000, time_limit_seconds: float = 600.0, iterations_limit: int = 10) -> BudgetStatus: Create a budget for a specific loop - add_tokens(loop_id: str, tokens: int): Add token usage to a loop's budget - start_timer(loop_id: str): Start time tracking for a loop - update_timer(loop_id: str): Update elapsed time for a loop - increment_iteration(loop_id: str): Increment iteration count for a loop - check_budget(loop_id: str) -> tuple[bool, str]: Check if a loop's budget has been exceeded. Returns (exceeded: bool, reason: str) - can_continue(loop_id: str) -> bool: Check if a loop can continue based on budget

Token Estimation: - estimate_tokens(text: str) -> int: ~4 chars per token - estimate_llm_call_tokens(prompt: str, response: str) -> int: Estimate LLM call tokens

Usage: 

from src.middleware.budget_tracker import BudgetTracker\n\ntracker = BudgetTracker()\nbudget = tracker.create_budget(\n    loop_id=\"research_loop\",\n    tokens_limit=100000,\n    time_limit_seconds=600,\n    iterations_limit=10\n)\ntracker.start_timer(\"research_loop\")\n# ... research operations ...\ntracker.add_tokens(\"research_loop\", 5000)\ntracker.update_timer(\"research_loop\")\nexceeded, reason = tracker.check_budget(\"research_loop\")\nif exceeded:\n    # Budget exceeded, stop research\n    pass\nif not tracker.can_continue(\"research_loop\"):\n    # Budget exceeded, stop research\n    pass\n

"},{"location":"architecture/middleware/#models","title":"Models","text":"

All middleware models are defined in src/utils/models.py:

"},{"location":"architecture/middleware/#thread-safety","title":"Thread Safety","text":"

All middleware components use ContextVar for thread-safe isolation:

"},{"location":"architecture/middleware/#see-also","title":"See Also","text":""},{"location":"architecture/orchestrators/","title":"Orchestrators Architecture","text":"

DeepCritical supports multiple orchestration patterns for research workflows.

"},{"location":"architecture/orchestrators/#research-flows","title":"Research Flows","text":""},{"location":"architecture/orchestrators/#iterativeresearchflow","title":"IterativeResearchFlow","text":"

File: src/orchestrator/research_flow.py

Pattern: Generate observations \u2192 Evaluate gaps \u2192 Select tools \u2192 Execute \u2192 Judge \u2192 Continue/Complete

Agents Used: - KnowledgeGapAgent: Evaluates research completeness - ToolSelectorAgent: Selects tools for addressing gaps - ThinkingAgent: Generates observations - WriterAgent: Creates final report - JudgeHandler: Assesses evidence sufficiency

Features: - Tracks iterations, time, budget - Supports graph execution (use_graph=True) and agent chains (use_graph=False) - Iterates until research complete or constraints met

Usage:

"},{"location":"architecture/orchestrators/#deepresearchflow","title":"DeepResearchFlow","text":"

File: src/orchestrator/research_flow.py

Pattern: Planner \u2192 Parallel iterative loops per section \u2192 Synthesizer

Agents Used: - PlannerAgent: Breaks query into report sections - IterativeResearchFlow: Per-section research (parallel) - LongWriterAgent or ProofreaderAgent: Final synthesis

Features: - Uses WorkflowManager for parallel execution - Budget tracking per section and globally - State synchronization across parallel loops - Supports graph execution and agent chains

Usage:

"},{"location":"architecture/orchestrators/#graph-orchestrator","title":"Graph Orchestrator","text":"

File: src/orchestrator/graph_orchestrator.py

Purpose: Graph-based execution using Pydantic AI agents as nodes

Features: - Uses graph execution (use_graph=True) or agent chains (use_graph=False) as fallback - Routes based on research mode (iterative/deep/auto) - Streams AgentEvent objects for UI - Uses GraphExecutionContext to manage execution state

Node Types: - Agent Nodes: Execute Pydantic AI agents - State Nodes: Update or read workflow state - Decision Nodes: Make routing decisions - Parallel Nodes: Execute multiple nodes concurrently

Edge Types: - Sequential Edges: Always traversed - Conditional Edges: Traversed based on condition - Parallel Edges: Used for parallel execution branches

Special Node Handling:

The GraphOrchestrator has special handling for certain nodes:

GraphExecutionContext:

The orchestrator uses GraphExecutionContext to manage execution state: - Tracks current node, visited nodes, and node results - Manages workflow state and budget tracker - Provides methods to store and retrieve node execution results

"},{"location":"architecture/orchestrators/#orchestrator-factory","title":"Orchestrator Factory","text":"

File: src/orchestrator_factory.py

Purpose: Factory for creating orchestrators

Modes: - Simple: Legacy orchestrator (backward compatible) - Advanced: Magentic orchestrator (requires OpenAI API key) - Auto-detect: Chooses based on API key availability

Usage:

"},{"location":"architecture/orchestrators/#magentic-orchestrator","title":"Magentic Orchestrator","text":"

File: src/orchestrator_magentic.py

Purpose: Multi-agent coordination using Microsoft Agent Framework

Features: - Uses agent-framework-core - ChatAgent pattern with internal LLMs per agent - MagenticBuilder with participants: - searcher: SearchAgent (wraps SearchHandler) - hypothesizer: HypothesisAgent (generates hypotheses) - judge: JudgeAgent (evaluates evidence) - reporter: ReportAgent (generates final report) - Manager orchestrates agents via chat client (OpenAI or HuggingFace) - Event-driven: converts Magentic events to AgentEvent for UI streaming via _process_event() method - Supports max rounds, stall detection, and reset handling

Event Processing:

The orchestrator processes Magentic events and converts them to AgentEvent: - MagenticOrchestratorMessageEvent \u2192 AgentEvent with type based on message content - MagenticAgentMessageEvent \u2192 AgentEvent with type based on agent name - MagenticAgentDeltaEvent \u2192 AgentEvent for streaming updates - MagenticFinalResultEvent \u2192 AgentEvent with type \"complete\"

Requirements: - agent-framework-core package - OpenAI API key or HuggingFace authentication

"},{"location":"architecture/orchestrators/#hierarchical-orchestrator","title":"Hierarchical Orchestrator","text":"

File: src/orchestrator_hierarchical.py

Purpose: Hierarchical orchestrator using middleware and sub-teams

Features: - Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge - Adapts Magentic ChatAgent to SubIterationTeam protocol - Event-driven via asyncio.Queue for coordination - Supports sub-iteration patterns for complex research tasks

"},{"location":"architecture/orchestrators/#legacy-simple-mode","title":"Legacy Simple Mode","text":"

File: src/legacy_orchestrator.py

Purpose: Linear search-judge-synthesize loop

Features: - Uses SearchHandlerProtocol and JudgeHandlerProtocol - Generator-based design yielding AgentEvent objects - Backward compatibility for simple use cases

"},{"location":"architecture/orchestrators/#state-initialization","title":"State Initialization","text":"

All orchestrators must initialize workflow state:

"},{"location":"architecture/orchestrators/#event-streaming","title":"Event Streaming","text":"

All orchestrators yield AgentEvent objects:

Event Types: - started: Research started - searching: Search in progress - search_complete: Search completed - judging: Evidence evaluation in progress - judge_complete: Evidence evaluation completed - looping: Iteration in progress - hypothesizing: Generating hypotheses - analyzing: Statistical analysis in progress - analysis_complete: Statistical analysis completed - synthesizing: Synthesizing results - complete: Research completed - error: Error occurred - streaming: Streaming update (delta events)

Event Structure:

"},{"location":"architecture/orchestrators/#see-also","title":"See Also","text":""},{"location":"architecture/services/","title":"Services Architecture","text":"

DeepCritical provides several services for embeddings, RAG, and statistical analysis.

"},{"location":"architecture/services/#embedding-service","title":"Embedding Service","text":"

File: src/services/embeddings.py

Purpose: Local sentence-transformers for semantic search and deduplication

Features: - No API Key Required: Uses local sentence-transformers models - Async-Safe: All operations use run_in_executor() to avoid blocking the event loop - ChromaDB Storage: In-memory vector storage for embeddings - Deduplication: 0.9 similarity threshold by default (90% similarity = duplicate, configurable)

Model: Configurable via settings.local_embedding_model (default: all-MiniLM-L6-v2)

Methods: - async def embed(text: str) -> list[float]: Generate embeddings (async-safe via run_in_executor()) - async def embed_batch(texts: list[str]) -> list[list[float]]: Batch embedding (more efficient) - async def add_evidence(evidence_id: str, content: str, metadata: dict[str, Any]) -> None: Add evidence to vector store - async def search_similar(query: str, n_results: int = 5) -> list[dict[str, Any]]: Find semantically similar evidence - async def deduplicate(new_evidence: list[Evidence], threshold: float = 0.9) -> list[Evidence]: Remove semantically duplicate evidence

Usage: 

from src.services.embeddings import get_embedding_service\n\nservice = get_embedding_service()\nembedding = await service.embed(\"text to embed\")\n

"},{"location":"architecture/services/#llamaindex-rag-service","title":"LlamaIndex RAG Service","text":"

File: src/services/llamaindex_rag.py

Purpose: Retrieval-Augmented Generation using LlamaIndex

Features: - Multiple Embedding Providers: OpenAI embeddings (requires OPENAI_API_KEY) or local sentence-transformers (no API key) - Multiple LLM Providers: HuggingFace LLM (preferred) or OpenAI LLM (fallback) for query synthesis - ChromaDB Storage: Vector database for document storage (supports in-memory mode) - Metadata Preservation: Preserves source, title, URL, date, authors - Lazy Initialization: Graceful fallback if dependencies not available

Initialization Parameters: - use_openai_embeddings: bool | None: Force OpenAI embeddings (None = auto-detect) - use_in_memory: bool: Use in-memory ChromaDB client (useful for tests) - oauth_token: str | None: Optional OAuth token from HuggingFace login (takes priority over env vars)

Methods: - async def ingest_evidence(evidence: list[Evidence]) -> None: Ingest evidence into RAG - async def retrieve(query: str, top_k: int = 5) -> list[Document]: Retrieve relevant documents - async def query(query: str, top_k: int = 5) -> str: Query with RAG

Usage: 

from src.services.llamaindex_rag import get_rag_service\n\nservice = get_rag_service(\n    use_openai_embeddings=False,  # Use local embeddings\n    use_in_memory=True,  # Use in-memory ChromaDB\n    oauth_token=token  # Optional HuggingFace token\n)\nif service:\n    documents = await service.retrieve(\"query\", top_k=5)\n

"},{"location":"architecture/services/#statistical-analyzer","title":"Statistical Analyzer","text":"

File: src/services/statistical_analyzer.py

Purpose: Secure execution of AI-generated statistical code

Features: - Modal Sandbox: Secure, isolated execution environment - Code Generation: Generates Python code via LLM - Library Pinning: Version-pinned libraries in SANDBOX_LIBRARIES - Network Isolation: block_network=True by default

Libraries Available: - pandas, numpy, scipy - matplotlib, scikit-learn - statsmodels

Output: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - code: Generated analysis code - output: Execution output - error: Error message if execution failed

Usage: 

from src.services.statistical_analyzer import StatisticalAnalyzer\n\nanalyzer = StatisticalAnalyzer()\nresult = await analyzer.analyze(\n    hypothesis=\"Metformin reduces cancer risk\",\n    evidence=evidence_list\n)\n

"},{"location":"architecture/services/#singleton-pattern","title":"Singleton Pattern","text":"

Services use singleton patterns for lazy initialization:

EmbeddingService: Uses a global variable pattern:

LlamaIndexRAGService: Direct instantiation (no caching):

This ensures: - Single instance per process - Lazy initialization - No dependencies required at import time

"},{"location":"architecture/services/#service-availability","title":"Service Availability","text":"

Services check availability before use:

from src.utils.config import settings\n\nif settings.modal_available:\n    # Use Modal sandbox\n    pass\n\nif settings.has_openai_key:\n    # Use OpenAI embeddings for RAG\n    pass\n
"},{"location":"architecture/services/#see-also","title":"See Also","text":""},{"location":"architecture/tools/","title":"Tools Architecture","text":"

DeepCritical implements a protocol-based search tool system for retrieving evidence from multiple sources.

"},{"location":"architecture/tools/#searchtool-protocol","title":"SearchTool Protocol","text":"

All tools implement the SearchTool protocol from src/tools/base.py:

"},{"location":"architecture/tools/#rate-limiting","title":"Rate Limiting","text":"

All tools use the @retry decorator from tenacity:

Tools with API rate limits implement _rate_limit() method and use shared rate limiters from src/tools/rate_limiter.py.

"},{"location":"architecture/tools/#error-handling","title":"Error Handling","text":"

Tools raise custom exceptions:

Tools handle HTTP errors (429, 500, timeout) and return empty lists on non-critical errors (with warning logs).

"},{"location":"architecture/tools/#query-preprocessing","title":"Query Preprocessing","text":"

Tools use preprocess_query() from src/tools/query_utils.py to:

"},{"location":"architecture/tools/#evidence-conversion","title":"Evidence Conversion","text":"

All tools convert API responses to Evidence objects with:

Missing fields are handled gracefully with defaults.

"},{"location":"architecture/tools/#tool-implementations","title":"Tool Implementations","text":""},{"location":"architecture/tools/#pubmed-tool","title":"PubMed Tool","text":"

File: src/tools/pubmed.py

API: NCBI E-utilities (ESearch \u2192 EFetch)

Rate Limiting: - 0.34s between requests (3 req/sec without API key) - 0.1s between requests (10 req/sec with NCBI API key)

Features: - XML parsing with xmltodict - Handles single vs. multiple articles - Query preprocessing - Evidence conversion with metadata extraction

"},{"location":"architecture/tools/#clinicaltrials-tool","title":"ClinicalTrials Tool","text":"

File: src/tools/clinicaltrials.py

API: ClinicalTrials.gov API v2

Important: Uses requests library (NOT httpx) because WAF blocks httpx TLS fingerprint.

Execution: Runs in thread pool: await asyncio.to_thread(requests.get, ...)

Filtering: - Only interventional studies - Status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION

Features: - Parses nested JSON structure - Extracts trial metadata - Evidence conversion

"},{"location":"architecture/tools/#europe-pmc-tool","title":"Europe PMC Tool","text":"

File: src/tools/europepmc.py

API: Europe PMC REST API

Features: - Handles preprint markers: [PREPRINT - Not peer-reviewed] - Builds URLs from DOI or PMID - Checks pubTypeList for preprint detection - Includes both preprints and peer-reviewed articles

"},{"location":"architecture/tools/#rag-tool","title":"RAG Tool","text":"

File: src/tools/rag_tool.py

Purpose: Semantic search within collected evidence

Implementation: Wraps LlamaIndexRAGService

Features: - Returns Evidence from RAG results - Handles evidence ingestion - Semantic similarity search - Metadata preservation

"},{"location":"architecture/tools/#search-handler","title":"Search Handler","text":"

File: src/tools/search_handler.py

Purpose: Orchestrates parallel searches across multiple tools

Initialization Parameters: - tools: list[SearchTool]: List of search tools to use - timeout: float = 30.0: Timeout for each search in seconds - include_rag: bool = False: Whether to include RAG tool in searches - auto_ingest_to_rag: bool = True: Whether to automatically ingest results into RAG - oauth_token: str | None = None: Optional OAuth token from HuggingFace login (for RAG LLM)

Methods: - async def execute(query: str, max_results_per_tool: int = 10) -> SearchResult: Execute search across all tools in parallel

Features: - Uses asyncio.gather() with return_exceptions=True for parallel execution - Aggregates results into SearchResult with evidence and metadata - Handles tool failures gracefully (continues with other tools) - Deduplicates results by URL - Automatically ingests results into RAG if auto_ingest_to_rag=True - Can add RAG tool dynamically via add_rag_tool() method

"},{"location":"architecture/tools/#tool-registration","title":"Tool Registration","text":"

Tools are registered in the search handler:

from src.tools.pubmed import PubMedTool\nfrom src.tools.clinicaltrials import ClinicalTrialsTool\nfrom src.tools.europepmc import EuropePMCTool\nfrom src.tools.search_handler import SearchHandler\n\nsearch_handler = SearchHandler(\n    tools=[\n        PubMedTool(),\n        ClinicalTrialsTool(),\n        EuropePMCTool(),\n    ],\n    include_rag=True,  # Include RAG tool for semantic search\n    auto_ingest_to_rag=True,  # Automatically ingest results into RAG\n    oauth_token=token  # Optional HuggingFace token for RAG LLM\n)\n\n# Execute search\nresult = await search_handler.execute(\"query\", max_results_per_tool=10)\n
"},{"location":"architecture/tools/#see-also","title":"See Also","text":""},{"location":"architecture/workflow-diagrams/","title":"DeepCritical Workflow - Simplified Magentic Architecture","text":"

Architecture Pattern: Microsoft Magentic Orchestration Design Philosophy: Simple, dynamic, manager-driven coordination Key Innovation: Intelligent manager replaces rigid sequential phases

"},{"location":"architecture/workflow-diagrams/#1-high-level-magentic-workflow","title":"1. High-Level Magentic Workflow","text":"
flowchart TD\n    Start([User Query]) --> Manager[Magentic Manager<br/>Plan \u2022 Select \u2022 Assess \u2022 Adapt]\n\n    Manager -->|Plans| Task1[Task Decomposition]\n    Task1 --> Manager\n\n    Manager -->|Selects & Executes| HypAgent[Hypothesis Agent]\n    Manager -->|Selects & Executes| SearchAgent[Search Agent]\n    Manager -->|Selects & Executes| AnalysisAgent[Analysis Agent]\n    Manager -->|Selects & Executes| ReportAgent[Report Agent]\n\n    HypAgent -->|Results| Manager\n    SearchAgent -->|Results| Manager\n    AnalysisAgent -->|Results| Manager\n    ReportAgent -->|Results| Manager\n\n    Manager -->|Assesses Quality| Decision{Good Enough?}\n    Decision -->|No - Refine| Manager\n    Decision -->|No - Different Agent| Manager\n    Decision -->|No - Stalled| Replan[Reset Plan]\n    Replan --> Manager\n\n    Decision -->|Yes| Synthesis[Synthesize Final Result]\n    Synthesis --> Output([Research Report])\n\n    style Start fill:#e1f5e1\n    style Manager fill:#ffe6e6\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style Decision fill:#ffd6d6\n    style Synthesis fill:#d4edda\n    style Output fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#2-magentic-manager-the-6-phase-cycle","title":"2. Magentic Manager: The 6-Phase Cycle","text":"
flowchart LR\n    P1[1. Planning<br/>Analyze task<br/>Create strategy] --> P2[2. Agent Selection<br/>Pick best agent<br/>for subtask]\n    P2 --> P3[3. Execution<br/>Run selected<br/>agent with tools]\n    P3 --> P4[4. Assessment<br/>Evaluate quality<br/>Check progress]\n    P4 --> Decision{Quality OK?<br/>Progress made?}\n    Decision -->|Yes| P6[6. Synthesis<br/>Combine results<br/>Generate report]\n    Decision -->|No| P5[5. Iteration<br/>Adjust plan<br/>Try again]\n    P5 --> P2\n    P6 --> Done([Complete])\n\n    style P1 fill:#fff4e6\n    style P2 fill:#ffe6e6\n    style P3 fill:#e6f3ff\n    style P4 fill:#ffd6d6\n    style P5 fill:#fff3cd\n    style P6 fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#3-simplified-agent-architecture","title":"3. Simplified Agent Architecture","text":"
graph TB\n    subgraph \"Orchestration Layer\"\n        Manager[Magentic Manager<br/>\u2022 Plans workflow<br/>\u2022 Selects agents<br/>\u2022 Assesses quality<br/>\u2022 Adapts strategy]\n        SharedContext[(Shared Context<br/>\u2022 Hypotheses<br/>\u2022 Search Results<br/>\u2022 Analysis<br/>\u2022 Progress)]\n        Manager <--> SharedContext\n    end\n\n    subgraph \"Specialist Agents\"\n        HypAgent[Hypothesis Agent<br/>\u2022 Domain understanding<br/>\u2022 Hypothesis generation<br/>\u2022 Testability refinement]\n        SearchAgent[Search Agent<br/>\u2022 Multi-source search<br/>\u2022 RAG retrieval<br/>\u2022 Result ranking]\n        AnalysisAgent[Analysis Agent<br/>\u2022 Evidence extraction<br/>\u2022 Statistical analysis<br/>\u2022 Code execution]\n        ReportAgent[Report Agent<br/>\u2022 Report assembly<br/>\u2022 Visualization<br/>\u2022 Citation formatting]\n    end\n\n    subgraph \"MCP Tools\"\n        WebSearch[Web Search<br/>PubMed \u2022 arXiv \u2022 bioRxiv]\n        CodeExec[Code Execution<br/>Sandboxed Python]\n        RAG[RAG Retrieval<br/>Vector DB \u2022 Embeddings]\n        Viz[Visualization<br/>Charts \u2022 Graphs]\n    end\n\n    Manager -->|Selects & Directs| HypAgent\n    Manager -->|Selects & Directs| SearchAgent\n    Manager -->|Selects & Directs| AnalysisAgent\n    Manager -->|Selects & Directs| ReportAgent\n\n    HypAgent --> SharedContext\n    SearchAgent --> SharedContext\n    AnalysisAgent --> SharedContext\n    ReportAgent --> SharedContext\n\n    SearchAgent --> WebSearch\n    SearchAgent --> RAG\n    AnalysisAgent --> CodeExec\n    ReportAgent --> CodeExec\n    ReportAgent --> Viz\n\n    style Manager fill:#ffe6e6\n    style SharedContext fill:#ffe6f0\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style WebSearch fill:#e6f3ff\n    style CodeExec fill:#e6f3ff\n    style RAG fill:#e6f3ff\n    style Viz fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#4-dynamic-workflow-example","title":"4. Dynamic Workflow Example","text":"
sequenceDiagram\n    participant User\n    participant Manager\n    participant HypAgent\n    participant SearchAgent\n    participant AnalysisAgent\n    participant ReportAgent\n\n    User->>Manager: \"Research protein folding in Alzheimer's\"\n\n    Note over Manager: PLAN: Generate hypotheses \u2192 Search \u2192 Analyze \u2192 Report\n\n    Manager->>HypAgent: Generate 3 hypotheses\n    HypAgent-->>Manager: Returns 3 hypotheses\n    Note over Manager: ASSESS: Good quality, proceed\n\n    Manager->>SearchAgent: Search literature for hypothesis 1\n    SearchAgent-->>Manager: Returns 15 papers\n    Note over Manager: ASSESS: Good results, continue\n\n    Manager->>SearchAgent: Search for hypothesis 2\n    SearchAgent-->>Manager: Only 2 papers found\n    Note over Manager: ASSESS: Insufficient, refine search\n\n    Manager->>SearchAgent: Refined query for hypothesis 2\n    SearchAgent-->>Manager: Returns 12 papers\n    Note over Manager: ASSESS: Better, proceed\n\n    Manager->>AnalysisAgent: Analyze evidence for all hypotheses\n    AnalysisAgent-->>Manager: Returns analysis with code\n    Note over Manager: ASSESS: Complete, generate report\n\n    Manager->>ReportAgent: Create comprehensive report\n    ReportAgent-->>Manager: Returns formatted report\n    Note over Manager: SYNTHESIZE: Combine all results\n\n    Manager->>User: Final Research Report
"},{"location":"architecture/workflow-diagrams/#5-manager-decision-logic","title":"5. Manager Decision Logic","text":"
flowchart TD\n    Start([Manager Receives Task]) --> Plan[Create Initial Plan]\n\n    Plan --> Select[Select Agent for Next Subtask]\n    Select --> Execute[Execute Agent]\n    Execute --> Collect[Collect Results]\n\n    Collect --> Assess[Assess Quality & Progress]\n\n    Assess --> Q1{Quality Sufficient?}\n    Q1 -->|No| Q2{Same Agent Can Fix?}\n    Q2 -->|Yes| Feedback[Provide Specific Feedback]\n    Feedback --> Execute\n    Q2 -->|No| Different[Try Different Agent]\n    Different --> Select\n\n    Q1 -->|Yes| Q3{Task Complete?}\n    Q3 -->|No| Q4{Making Progress?}\n    Q4 -->|Yes| Select\n    Q4 -->|No - Stalled| Replan[Reset Plan & Approach]\n    Replan --> Plan\n\n    Q3 -->|Yes| Synth[Synthesize Final Result]\n    Synth --> Done([Return Report])\n\n    style Start fill:#e1f5e1\n    style Plan fill:#fff4e6\n    style Select fill:#ffe6e6\n    style Execute fill:#e6f3ff\n    style Assess fill:#ffd6d6\n    style Q1 fill:#ffe6e6\n    style Q2 fill:#ffe6e6\n    style Q3 fill:#ffe6e6\n    style Q4 fill:#ffe6e6\n    style Synth fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#6-hypothesis-agent-workflow","title":"6. Hypothesis Agent Workflow","text":"
flowchart LR\n    Input[Research Query] --> Domain[Identify Domain<br/>& Key Concepts]\n    Domain --> Context[Retrieve Background<br/>Knowledge]\n    Context --> Generate[Generate 3-5<br/>Initial Hypotheses]\n    Generate --> Refine[Refine for<br/>Testability]\n    Refine --> Rank[Rank by<br/>Quality Score]\n    Rank --> Output[Return Top<br/>Hypotheses]\n\n    Output --> Struct[Hypothesis Structure:<br/>\u2022 Statement<br/>\u2022 Rationale<br/>\u2022 Testability Score<br/>\u2022 Data Requirements<br/>\u2022 Expected Outcomes]\n\n    style Input fill:#e1f5e1\n    style Output fill:#fff4e6\n    style Struct fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#7-search-agent-workflow","title":"7. Search Agent Workflow","text":"
flowchart TD\n    Input[Hypotheses] --> Strategy[Formulate Search<br/>Strategy per Hypothesis]\n\n    Strategy --> Multi[Multi-Source Search]\n\n    Multi --> PubMed[PubMed Search<br/>via MCP]\n    Multi --> ArXiv[arXiv Search<br/>via MCP]\n    Multi --> BioRxiv[bioRxiv Search<br/>via MCP]\n\n    PubMed --> Aggregate[Aggregate Results]\n    ArXiv --> Aggregate\n    BioRxiv --> Aggregate\n\n    Aggregate --> Filter[Filter & Rank<br/>by Relevance]\n    Filter --> Dedup[Deduplicate<br/>Cross-Reference]\n    Dedup --> Embed[Embed Documents<br/>via MCP]\n    Embed --> Vector[(Vector DB)]\n    Vector --> RAGRetrieval[RAG Retrieval<br/>Top-K per Hypothesis]\n    RAGRetrieval --> Output[Return Contextualized<br/>Search Results]\n\n    style Input fill:#fff4e6\n    style Multi fill:#ffe6e6\n    style Vector fill:#ffe6f0\n    style Output fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#8-analysis-agent-workflow","title":"8. Analysis Agent Workflow","text":"
flowchart TD\n    Input1[Hypotheses] --> Extract\n    Input2[Search Results] --> Extract[Extract Evidence<br/>per Hypothesis]\n\n    Extract --> Methods[Determine Analysis<br/>Methods Needed]\n\n    Methods --> Branch{Requires<br/>Computation?}\n    Branch -->|Yes| GenCode[Generate Python<br/>Analysis Code]\n    Branch -->|No| Qual[Qualitative<br/>Synthesis]\n\n    GenCode --> Execute[Execute Code<br/>via MCP Sandbox]\n    Execute --> Interpret1[Interpret<br/>Results]\n    Qual --> Interpret2[Interpret<br/>Findings]\n\n    Interpret1 --> Synthesize[Synthesize Evidence<br/>Across Sources]\n    Interpret2 --> Synthesize\n\n    Synthesize --> Verdict[Determine Verdict<br/>per Hypothesis]\n    Verdict --> Support[\u2022 Supported<br/>\u2022 Refuted<br/>\u2022 Inconclusive]\n    Support --> Gaps[Identify Knowledge<br/>Gaps & Limitations]\n    Gaps --> Output[Return Analysis<br/>Report]\n\n    style Input1 fill:#fff4e6\n    style Input2 fill:#e6f3ff\n    style Execute fill:#ffe6e6\n    style Output fill:#e6ffe6
"},{"location":"architecture/workflow-diagrams/#9-report-agent-workflow","title":"9. Report Agent Workflow","text":"
flowchart TD\n    Input1[Query] --> Assemble\n    Input2[Hypotheses] --> Assemble\n    Input3[Search Results] --> Assemble\n    Input4[Analysis] --> Assemble[Assemble Report<br/>Sections]\n\n    Assemble --> Exec[Executive Summary]\n    Assemble --> Intro[Introduction]\n    Assemble --> Methods[Methods]\n    Assemble --> Results[Results per<br/>Hypothesis]\n    Assemble --> Discussion[Discussion]\n    Assemble --> Future[Future Directions]\n    Assemble --> Refs[References]\n\n    Results --> VizCheck{Needs<br/>Visualization?}\n    VizCheck -->|Yes| GenViz[Generate Viz Code]\n    GenViz --> ExecViz[Execute via MCP<br/>Create Charts]\n    ExecViz --> Combine\n    VizCheck -->|No| Combine[Combine All<br/>Sections]\n\n    Exec --> Combine\n    Intro --> Combine\n    Methods --> Combine\n    Discussion --> Combine\n    Future --> Combine\n    Refs --> Combine\n\n    Combine --> Format[Format Output]\n    Format --> MD[Markdown]\n    Format --> PDF[PDF]\n    Format --> JSON[JSON]\n\n    MD --> Output[Return Final<br/>Report]\n    PDF --> Output\n    JSON --> Output\n\n    style Input1 fill:#e1f5e1\n    style Input2 fill:#fff4e6\n    style Input3 fill:#e6f3ff\n    style Input4 fill:#e6ffe6\n    style Output fill:#d4edda
"},{"location":"architecture/workflow-diagrams/#10-data-flow--event-streaming","title":"10. Data Flow & Event Streaming","text":"
flowchart TD\n    User[\ud83d\udc64 User] -->|Research Query| UI[Gradio UI]\n    UI -->|Submit| Manager[Magentic Manager]\n\n    Manager -->|Event: Planning| UI\n    Manager -->|Select Agent| HypAgent[Hypothesis Agent]\n    HypAgent -->|Event: Delta/Message| UI\n    HypAgent -->|Hypotheses| Context[(Shared Context)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| SearchAgent[Search Agent]\n    SearchAgent -->|MCP Request| WebSearch[Web Search Tool]\n    WebSearch -->|Results| SearchAgent\n    SearchAgent -->|Event: Delta/Message| UI\n    SearchAgent -->|Documents| Context\n    SearchAgent -->|Embeddings| VectorDB[(Vector DB)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| AnalysisAgent[Analysis Agent]\n    AnalysisAgent -->|MCP Request| CodeExec[Code Execution Tool]\n    CodeExec -->|Results| AnalysisAgent\n    AnalysisAgent -->|Event: Delta/Message| UI\n    AnalysisAgent -->|Analysis| Context\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| ReportAgent[Report Agent]\n    ReportAgent -->|MCP Request| CodeExec\n    ReportAgent -->|Event: Delta/Message| UI\n    ReportAgent -->|Report| Context\n\n    Manager -->|Event: Final Result| UI\n    UI -->|Display| User\n\n    style User fill:#e1f5e1\n    style UI fill:#e6f3ff\n    style Manager fill:#ffe6e6\n    style Context fill:#ffe6f0\n    style VectorDB fill:#ffe6f0\n    style WebSearch fill:#f0f0f0\n    style CodeExec fill:#f0f0f0
"},{"location":"architecture/workflow-diagrams/#11-mcp-tool-architecture","title":"11. MCP Tool Architecture","text":"
graph TB\n    subgraph \"Agent Layer\"\n        Manager[Magentic Manager]\n        HypAgent[Hypothesis Agent]\n        SearchAgent[Search Agent]\n        AnalysisAgent[Analysis Agent]\n        ReportAgent[Report Agent]\n    end\n\n    subgraph \"MCP Protocol Layer\"\n        Registry[MCP Tool Registry<br/>\u2022 Discovers tools<br/>\u2022 Routes requests<br/>\u2022 Manages connections]\n    end\n\n    subgraph \"MCP Servers\"\n        Server1[Web Search Server<br/>localhost:8001<br/>\u2022 PubMed<br/>\u2022 arXiv<br/>\u2022 bioRxiv]\n        Server2[Code Execution Server<br/>localhost:8002<br/>\u2022 Sandboxed Python<br/>\u2022 Package management]\n        Server3[RAG Server<br/>localhost:8003<br/>\u2022 Vector embeddings<br/>\u2022 Similarity search]\n        Server4[Visualization Server<br/>localhost:8004<br/>\u2022 Chart generation<br/>\u2022 Plot rendering]\n    end\n\n    subgraph \"External Services\"\n        PubMed[PubMed API]\n        ArXiv[arXiv API]\n        BioRxiv[bioRxiv API]\n        Modal[Modal Sandbox]\n        ChromaDB[(ChromaDB)]\n    end\n\n    SearchAgent -->|Request| Registry\n    AnalysisAgent -->|Request| Registry\n    ReportAgent -->|Request| Registry\n\n    Registry --> Server1\n    Registry --> Server2\n    Registry --> Server3\n    Registry --> Server4\n\n    Server1 --> PubMed\n    Server1 --> ArXiv\n    Server1 --> BioRxiv\n    Server2 --> Modal\n    Server3 --> ChromaDB\n\n    style Manager fill:#ffe6e6\n    style Registry fill:#fff4e6\n    style Server1 fill:#e6f3ff\n    style Server2 fill:#e6f3ff\n    style Server3 fill:#e6f3ff\n    style Server4 fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#12-progress-tracking--stall-detection","title":"12. Progress Tracking & Stall Detection","text":"
stateDiagram-v2\n    [*] --> Initialization: User Query\n\n    Initialization --> Planning: Manager starts\n\n    Planning --> AgentExecution: Select agent\n\n    AgentExecution --> Assessment: Collect results\n\n    Assessment --> QualityCheck: Evaluate output\n\n    QualityCheck --> AgentExecution: Poor quality<br/>(retry < max_rounds)\n    QualityCheck --> Planning: Poor quality<br/>(try different agent)\n    QualityCheck --> NextAgent: Good quality<br/>(task incomplete)\n    QualityCheck --> Synthesis: Good quality<br/>(task complete)\n\n    NextAgent --> AgentExecution: Select next agent\n\n    state StallDetection <<choice>>\n    Assessment --> StallDetection: Check progress\n    StallDetection --> Planning: No progress<br/>(stall count < max)\n    StallDetection --> ErrorRecovery: No progress<br/>(max stalls reached)\n\n    ErrorRecovery --> PartialReport: Generate partial results\n    PartialReport --> [*]\n\n    Synthesis --> FinalReport: Combine all outputs\n    FinalReport --> [*]\n\n    note right of QualityCheck\n        Manager assesses:\n        \u2022 Output completeness\n        \u2022 Quality metrics\n        \u2022 Progress made\n    end note\n\n    note right of StallDetection\n        Stall = no new progress\n        after agent execution\n        Triggers plan reset\n    end note
"},{"location":"architecture/workflow-diagrams/#13-gradio-ui-integration","title":"13. Gradio UI Integration","text":"
graph TD\n    App[Gradio App<br/>DeepCritical Research Agent]\n\n    App --> Input[Input Section]\n    App --> Status[Status Section]\n    App --> Output[Output Section]\n\n    Input --> Query[Research Question<br/>Text Area]\n    Input --> Controls[Controls]\n    Controls --> MaxHyp[Max Hypotheses: 1-10]\n    Controls --> MaxRounds[Max Rounds: 5-20]\n    Controls --> Submit[Start Research Button]\n\n    Status --> Log[Real-time Event Log<br/>\u2022 Manager planning<br/>\u2022 Agent selection<br/>\u2022 Execution updates<br/>\u2022 Quality assessment]\n    Status --> Progress[Progress Tracker<br/>\u2022 Current agent<br/>\u2022 Round count<br/>\u2022 Stall count]\n\n    Output --> Tabs[Tabbed Results]\n    Tabs --> Tab1[Hypotheses Tab<br/>Generated hypotheses with scores]\n    Tabs --> Tab2[Search Results Tab<br/>Papers & sources found]\n    Tabs --> Tab3[Analysis Tab<br/>Evidence & verdicts]\n    Tabs --> Tab4[Report Tab<br/>Final research report]\n    Tab4 --> Download[Download Report<br/>MD / PDF / JSON]\n\n    Submit -.->|Triggers| Workflow[Magentic Workflow]\n    Workflow -.->|MagenticOrchestratorMessageEvent| Log\n    Workflow -.->|MagenticAgentDeltaEvent| Log\n    Workflow -.->|MagenticAgentMessageEvent| Log\n    Workflow -.->|MagenticFinalResultEvent| Tab4\n\n    style App fill:#e1f5e1\n    style Input fill:#fff4e6\n    style Status fill:#e6f3ff\n    style Output fill:#e6ffe6\n    style Workflow fill:#ffe6e6
"},{"location":"architecture/workflow-diagrams/#14-complete-system-context","title":"14. Complete System Context","text":"
graph LR\n    User[\ud83d\udc64 Researcher<br/>Asks research questions] -->|Submits query| DC[DeepCritical<br/>Magentic Workflow]\n\n    DC -->|Literature search| PubMed[PubMed API<br/>Medical papers]\n    DC -->|Preprint search| ArXiv[arXiv API<br/>Scientific preprints]\n    DC -->|Biology search| BioRxiv[bioRxiv API<br/>Biology preprints]\n    DC -->|Agent reasoning| Claude[Claude API<br/>Sonnet 4 / Opus]\n    DC -->|Code execution| Modal[Modal Sandbox<br/>Safe Python env]\n    DC -->|Vector storage| Chroma[ChromaDB<br/>Embeddings & RAG]\n\n    DC -->|Deployed on| HF[HuggingFace Spaces<br/>Gradio 6.0]\n\n    PubMed -->|Results| DC\n    ArXiv -->|Results| DC\n    BioRxiv -->|Results| DC\n    Claude -->|Responses| DC\n    Modal -->|Output| DC\n    Chroma -->|Context| DC\n\n    DC -->|Research report| User\n\n    style User fill:#e1f5e1\n    style DC fill:#ffe6e6\n    style PubMed fill:#e6f3ff\n    style ArXiv fill:#e6f3ff\n    style BioRxiv fill:#e6f3ff\n    style Claude fill:#ffd6d6\n    style Modal fill:#f0f0f0\n    style Chroma fill:#ffe6f0\n    style HF fill:#d4edda
"},{"location":"architecture/workflow-diagrams/#15-workflow-timeline-simplified","title":"15. Workflow Timeline (Simplified)","text":"
gantt\n    title DeepCritical Magentic Workflow - Typical Execution\n    dateFormat mm:ss\n    axisFormat %M:%S\n\n    section Manager Planning\n    Initial planning         :p1, 00:00, 10s\n\n    section Hypothesis Agent\n    Generate hypotheses      :h1, after p1, 30s\n    Manager assessment       :h2, after h1, 5s\n\n    section Search Agent\n    Search hypothesis 1      :s1, after h2, 20s\n    Search hypothesis 2      :s2, after s1, 20s\n    Search hypothesis 3      :s3, after s2, 20s\n    RAG processing          :s4, after s3, 15s\n    Manager assessment      :s5, after s4, 5s\n\n    section Analysis Agent\n    Evidence extraction     :a1, after s5, 15s\n    Code generation        :a2, after a1, 20s\n    Code execution         :a3, after a2, 25s\n    Synthesis              :a4, after a3, 20s\n    Manager assessment     :a5, after a4, 5s\n\n    section Report Agent\n    Report assembly        :r1, after a5, 30s\n    Visualization          :r2, after r1, 15s\n    Formatting             :r3, after r2, 10s\n\n    section Manager Synthesis\n    Final synthesis        :f1, after r3, 10s
"},{"location":"architecture/workflow-diagrams/#key-differences-from-original-design","title":"Key Differences from Original Design","text":"Aspect Original (Judge-in-Loop) New (Magentic) Control Flow Fixed sequential phases Dynamic agent selection Quality Control Separate Judge Agent Manager assessment built-in Retry Logic Phase-level with feedback Agent-level with adaptation Flexibility Rigid 4-phase pipeline Adaptive workflow Complexity 5 agents (including Judge) 4 agents (no Judge) Progress Tracking Manual state management Built-in round/stall detection Agent Coordination Sequential handoff Manager-driven dynamic selection Error Recovery Retry same phase Try different agent or replan"},{"location":"architecture/workflow-diagrams/#simplified-design-principles","title":"Simplified Design Principles","text":"
  1. Manager is Intelligent: LLM-powered manager handles planning, selection, and quality assessment
  2. No Separate Judge: Manager's assessment phase replaces dedicated Judge Agent
  3. Dynamic Workflow: Agents can be called multiple times in any order based on need
  4. Built-in Safety: max_round_count (15) and max_stall_count (3) prevent infinite loops
  5. Event-Driven UI: Real-time streaming updates to Gradio interface
  6. MCP-Powered Tools: All external capabilities via Model Context Protocol
  7. Shared Context: Centralized state accessible to all agents
  8. Progress Awareness: Manager tracks what's been done and what's needed
"},{"location":"architecture/workflow-diagrams/#legend","title":"Legend","text":""},{"location":"architecture/workflow-diagrams/#implementation-highlights","title":"Implementation Highlights","text":"

Simple 4-Agent Setup:

Manager handles quality assessment in its instructions: - Checks hypothesis quality (testable, novel, clear) - Validates search results (relevant, authoritative, recent) - Assesses analysis soundness (methodology, evidence, conclusions) - Ensures report completeness (all sections, proper citations)

No separate Judge Agent needed - manager does it all!

Document Version: 2.0 (Magentic Simplified) Last Updated: 2025-11-24 Architecture: Microsoft Magentic Orchestration Pattern Agents: 4 (Hypothesis, Search, Analysis, Report) + 1 Manager License: MIT

"},{"location":"architecture/workflow-diagrams/#see-also","title":"See Also","text":""},{"location":"configuration/","title":"Configuration Guide","text":""},{"location":"configuration/#overview","title":"Overview","text":"

DeepCritical uses Pydantic Settings for centralized configuration management. All settings are defined in the Settings class in src/utils/config.py and can be configured via environment variables or a .env file.

The configuration system provides:

"},{"location":"configuration/#quick-start","title":"Quick Start","text":"
  1. Create a .env file in the project root
  2. Set at least one LLM API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, or HF_TOKEN)
  3. Optionally configure other services as needed
  4. The application will automatically load and validate your configuration
"},{"location":"configuration/#configuration-system-architecture","title":"Configuration System Architecture","text":""},{"location":"configuration/#settings-class","title":"Settings Class","text":"

The [Settings][settings-class] class extends BaseSettings from pydantic_settings and defines all application configuration:

View source

"},{"location":"configuration/#singleton-instance","title":"Singleton Instance","text":"

A global settings instance is available for import:

View source

"},{"location":"configuration/#usage-pattern","title":"Usage Pattern","text":"

Access configuration throughout the codebase:

from src.utils.config import settings\n\n# Check if API keys are available\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\n# Access configuration values\nmax_iterations = settings.max_iterations\nweb_search_provider = settings.web_search_provider\n
"},{"location":"configuration/#required-configuration","title":"Required Configuration","text":""},{"location":"configuration/#llm-provider","title":"LLM Provider","text":"

You must configure at least one LLM provider. The system supports:

"},{"location":"configuration/#openai-configuration","title":"OpenAI Configuration","text":"
LLM_PROVIDER=openai\nOPENAI_API_KEY=your_openai_api_key_here\nOPENAI_MODEL=gpt-5.1\n

The default model is defined in the Settings class:

"},{"location":"configuration/#anthropic-configuration","title":"Anthropic Configuration","text":"
LLM_PROVIDER=anthropic\nANTHROPIC_API_KEY=your_anthropic_api_key_here\nANTHROPIC_MODEL=claude-sonnet-4-5-20250929\n

The default model is defined in the Settings class:

"},{"location":"configuration/#huggingface-configuration","title":"HuggingFace Configuration","text":"

HuggingFace can work without an API key for public models, but an API key provides higher rate limits:

# Option 1: Using HF_TOKEN (preferred)\nHF_TOKEN=your_huggingface_token_here\n\n# Option 2: Using HUGGINGFACE_API_KEY (alternative)\nHUGGINGFACE_API_KEY=your_huggingface_api_key_here\n\n# Default model\nHUGGINGFACE_MODEL=meta-llama/Llama-3.1-8B-Instruct\n

The HuggingFace token can be set via either environment variable:

"},{"location":"configuration/#optional-configuration","title":"Optional Configuration","text":""},{"location":"configuration/#embedding-configuration","title":"Embedding Configuration","text":"

DeepCritical supports multiple embedding providers for semantic search and RAG:

# Embedding Provider: \"openai\", \"local\", or \"huggingface\"\nEMBEDDING_PROVIDER=local\n\n# OpenAI Embedding Model (used by LlamaIndex RAG)\nOPENAI_EMBEDDING_MODEL=text-embedding-3-small\n\n# Local Embedding Model (sentence-transformers, used by EmbeddingService)\nLOCAL_EMBEDDING_MODEL=all-MiniLM-L6-v2\n\n# HuggingFace Embedding Model\nHUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2\n

The embedding provider configuration:

Note: OpenAI embeddings require OPENAI_API_KEY. The local provider (default) uses sentence-transformers and requires no API key.

"},{"location":"configuration/#web-search-configuration","title":"Web Search Configuration","text":"

DeepCritical supports multiple web search providers:

# Web Search Provider: \"serper\", \"searchxng\", \"brave\", \"tavily\", or \"duckduckgo\"\n# Default: \"duckduckgo\" (no API key required)\nWEB_SEARCH_PROVIDER=duckduckgo\n\n# Serper API Key (for Google search via Serper)\nSERPER_API_KEY=your_serper_api_key_here\n\n# SearchXNG Host URL (for self-hosted search)\nSEARCHXNG_HOST=http://localhost:8080\n\n# Brave Search API Key\nBRAVE_API_KEY=your_brave_api_key_here\n\n# Tavily API Key\nTAVILY_API_KEY=your_tavily_api_key_here\n

The web search provider configuration:

Note: DuckDuckGo is the default and requires no API key, making it ideal for development and testing.

"},{"location":"configuration/#pubmed-configuration","title":"PubMed Configuration","text":"

PubMed search supports optional NCBI API key for higher rate limits:

# NCBI API Key (optional, for higher rate limits: 10 req/sec vs 3 req/sec)\nNCBI_API_KEY=your_ncbi_api_key_here\n

The PubMed tool uses this configuration:

"},{"location":"configuration/#agent-configuration","title":"Agent Configuration","text":"

Control agent behavior and research loop execution:

# Maximum iterations per research loop (1-50, default: 10)\nMAX_ITERATIONS=10\n\n# Search timeout in seconds\nSEARCH_TIMEOUT=30\n\n# Use graph-based execution for research flows\nUSE_GRAPH_EXECUTION=false\n

The agent configuration fields:

"},{"location":"configuration/#budget--rate-limiting-configuration","title":"Budget & Rate Limiting Configuration","text":"

Control resource limits for research loops:

# Default token budget per research loop (1000-1000000, default: 100000)\nDEFAULT_TOKEN_LIMIT=100000\n\n# Default time limit per research loop in minutes (1-120, default: 10)\nDEFAULT_TIME_LIMIT_MINUTES=10\n\n# Default iterations limit per research loop (1-50, default: 10)\nDEFAULT_ITERATIONS_LIMIT=10\n

The budget configuration with validation:

"},{"location":"configuration/#rag-service-configuration","title":"RAG Service Configuration","text":"

Configure the Retrieval-Augmented Generation service:

# ChromaDB collection name for RAG\nRAG_COLLECTION_NAME=deepcritical_evidence\n\n# Number of top results to retrieve from RAG (1-50, default: 5)\nRAG_SIMILARITY_TOP_K=5\n\n# Automatically ingest evidence into RAG\nRAG_AUTO_INGEST=true\n

The RAG configuration:

"},{"location":"configuration/#chromadb-configuration","title":"ChromaDB Configuration","text":"

Configure the vector database for embeddings and RAG:

# ChromaDB storage path\nCHROMA_DB_PATH=./chroma_db\n\n# Whether to persist ChromaDB to disk\nCHROMA_DB_PERSIST=true\n\n# ChromaDB server host (for remote ChromaDB, optional)\nCHROMA_DB_HOST=localhost\n\n# ChromaDB server port (for remote ChromaDB, optional)\nCHROMA_DB_PORT=8000\n

The ChromaDB configuration:

"},{"location":"configuration/#external-services","title":"External Services","text":""},{"location":"configuration/#modal-configuration","title":"Modal Configuration","text":"

Modal is used for secure sandbox execution of statistical analysis:

# Modal Token ID (for Modal sandbox execution)\nMODAL_TOKEN_ID=your_modal_token_id_here\n\n# Modal Token Secret\nMODAL_TOKEN_SECRET=your_modal_token_secret_here\n

The Modal configuration:

"},{"location":"configuration/#logging-configuration","title":"Logging Configuration","text":"

Configure structured logging:

# Log Level: \"DEBUG\", \"INFO\", \"WARNING\", or \"ERROR\"\nLOG_LEVEL=INFO\n

The logging configuration:

Logging is configured via the configure_logging() function:

"},{"location":"configuration/#configuration-properties","title":"Configuration Properties","text":"

The Settings class provides helpful properties for checking configuration state:

"},{"location":"configuration/#api-key-availability","title":"API Key Availability","text":"

Check which API keys are available:

Usage:

from src.utils.config import settings\n\n# Check API key availability\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\nif settings.has_anthropic_key:\n    # Use Anthropic\n    pass\n\nif settings.has_huggingface_key:\n    # Use HuggingFace\n    pass\n\nif settings.has_any_llm_key:\n    # At least one LLM is available\n    pass\n
"},{"location":"configuration/#service-availability","title":"Service Availability","text":"

Check if external services are configured:

Usage:

from src.utils.config import settings\n\n# Check service availability\nif settings.modal_available:\n    # Use Modal sandbox\n    pass\n\nif settings.web_search_available:\n    # Web search is configured\n    pass\n
"},{"location":"configuration/#api-key-retrieval","title":"API Key Retrieval","text":"

Get the API key for the configured provider:

For OpenAI-specific operations (e.g., Magentic mode):

"},{"location":"configuration/#configuration-usage-in-codebase","title":"Configuration Usage in Codebase","text":"

The configuration system is used throughout the codebase:

"},{"location":"configuration/#llm-factory","title":"LLM Factory","text":"

The LLM factory uses settings to create appropriate models:

"},{"location":"configuration/#embedding-service","title":"Embedding Service","text":"

The embedding service uses local embedding model configuration:

"},{"location":"configuration/#orchestrator-factory","title":"Orchestrator Factory","text":"

The orchestrator factory uses settings to determine mode:

"},{"location":"configuration/#environment-variables-reference","title":"Environment Variables Reference","text":""},{"location":"configuration/#required-at-least-one-llm","title":"Required (at least one LLM)","text":""},{"location":"configuration/#llm-configuration-variables","title":"LLM Configuration Variables","text":""},{"location":"configuration/#embedding-configuration-variables","title":"Embedding Configuration Variables","text":""},{"location":"configuration/#web-search-configuration-variables","title":"Web Search Configuration Variables","text":""},{"location":"configuration/#pubmed-configuration-variables","title":"PubMed Configuration Variables","text":""},{"location":"configuration/#agent-configuration-variables","title":"Agent Configuration Variables","text":""},{"location":"configuration/#budget-configuration-variables","title":"Budget Configuration Variables","text":""},{"location":"configuration/#rag-configuration-variables","title":"RAG Configuration Variables","text":""},{"location":"configuration/#chromadb-configuration-variables","title":"ChromaDB Configuration Variables","text":""},{"location":"configuration/#external-services-variables","title":"External Services Variables","text":""},{"location":"configuration/#logging-configuration-variables","title":"Logging Configuration Variables","text":""},{"location":"configuration/#validation","title":"Validation","text":"

Settings are validated on load using Pydantic validation:

"},{"location":"configuration/#validation-examples","title":"Validation Examples","text":"

The max_iterations field has range validation:

The llm_provider field has literal validation:

"},{"location":"configuration/#error-handling","title":"Error Handling","text":"

Configuration errors raise ConfigurationError from src/utils/exceptions.py:

```22:25:src/utils/exceptions.py class ConfigurationError(DeepCriticalError): \"\"\"Raised when configuration is invalid.\"\"\"

pass\n

```

"},{"location":"configuration/#error-handling-example","title":"Error Handling Example","text":"

python from src.utils.config import settings from src.utils.exceptions import ConfigurationError try: api_key = settings.get_api_key() except ConfigurationError as e: print(f\"Configuration error: {e}\")

"},{"location":"configuration/#common-configuration-errors","title":"Common Configuration Errors","text":"
  1. Missing API Key: When get_api_key() is called but the required API key is not set
  2. Invalid Provider: When llm_provider is set to an unsupported value
  3. Out of Range: When numeric values exceed their min/max constraints
  4. Invalid Literal: When enum fields receive unsupported values
"},{"location":"configuration/#configuration-best-practices","title":"Configuration Best Practices","text":"
  1. Use .env File: Store sensitive keys in .env file (add to .gitignore)
  2. Check Availability: Use properties like has_openai_key before accessing API keys
  3. Handle Errors: Always catch ConfigurationError when calling get_api_key()
  4. Validate Early: Configuration is validated on import, so errors surface immediately
  5. Use Defaults: Leverage sensible defaults for optional configuration
"},{"location":"configuration/#future-enhancements","title":"Future Enhancements","text":"

The following configurations are planned for future phases:

  1. Additional LLM Providers: DeepSeek, OpenRouter, Gemini, Perplexity, Azure OpenAI, Local models
  2. Model Selection: Reasoning/main/fast model configuration
  3. Service Integration: Additional service integrations and configurations
"},{"location":"contributing/","title":"Contributing to The DETERMINATOR","text":"

Thank you for your interest in contributing to The DETERMINATOR! This guide will help you get started.

Note on Project Names: \"The DETERMINATOR\" is the product name, \"DeepCritical\" is the organization/project name, and \"determinator\" is the Python package name.

"},{"location":"contributing/#git-workflow","title":"Git Workflow","text":""},{"location":"contributing/#repository-information","title":"Repository Information","text":""},{"location":"contributing/#dual-repository-setup","title":"Dual Repository Setup","text":"

This project uses a dual repository setup:

"},{"location":"contributing/#remote-configuration","title":"Remote Configuration","text":"

When cloning, set up remotes as follows:

# Clone from GitHub\ngit clone https://github.com/DeepCritical/GradioDemo.git\ncd GradioDemo\n\n# Add HuggingFace remote (optional, for deployment)\ngit remote add huggingface-upstream https://huggingface.co/spaces/DataQuests/DeepCritical\n

Important: Never push directly to main or dev on HuggingFace. Always work through GitHub PRs. GitHub is the source of truth; HuggingFace is for deployment/demo only.

"},{"location":"contributing/#package-manager","title":"Package Manager","text":"

This project uses uv as the package manager. All commands should be prefixed with uv run to ensure they run in the correct environment.

"},{"location":"contributing/#installation","title":"Installation","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync all dependencies including dev extras\nuv sync --all-extras\n\n# Install pre-commit hooks\nuv run pre-commit install\n
"},{"location":"contributing/#development-commands","title":"Development Commands","text":"
# Installation\nuv sync --all-extras              # Install all dependencies including dev\nuv run pre-commit install          # Install pre-commit hooks\n\n# Code Quality Checks (run all before committing)\nuv run ruff check src tests       # Lint with ruff\nuv run ruff format src tests      # Format with ruff\nuv run mypy src                   # Type checking\nuv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire  # Tests with coverage\n\n# Testing Commands\nuv run pytest tests/unit/ -v -m \"not openai\" -p no:logfire              # Run unit tests (excludes OpenAI tests)\nuv run pytest tests/ -v -m \"huggingface\" -p no:logfire                 # Run HuggingFace tests\nuv run pytest tests/ -v -p no:logfire                                  # Run all tests\nuv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire  # Tests with terminal coverage\nuv run pytest --cov=src --cov-report=html -p no:logfire                # Generate HTML coverage report (opens htmlcov/index.html)\n\n# Documentation Commands\nuv run mkdocs build                # Build documentation\nuv run mkdocs serve                # Serve documentation locally (http://127.0.0.1:8000)\n
"},{"location":"contributing/#test-markers","title":"Test Markers","text":"

The project uses pytest markers to categorize tests. See Testing Guidelines for details:

Note: The -p no:logfire flag disables the logfire plugin to avoid conflicts during testing.

"},{"location":"contributing/#getting-started","title":"Getting Started","text":"

  1. Fork the repository on GitHub: DeepCritical/GradioDemo

  2. Clone your fork:

git clone https://github.com/yourusername/GradioDemo.git\ncd GradioDemo\n
  1. Install dependencies:
uv sync --all-extras\nuv run pre-commit install\n
  1. Create a feature branch:
git checkout -b yourname-feature-name\n

  1. Make your changes following the guidelines below

  2. Run checks:

uv run ruff check src tests\nuv run mypy src\nuv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire\n
  1. Commit and push:
git commit -m \"Description of changes\"\ngit push origin yourname-feature-name\n
  1. Create a pull request on GitHub
"},{"location":"contributing/#development-guidelines","title":"Development Guidelines","text":""},{"location":"contributing/#code-style","title":"Code Style","text":""},{"location":"contributing/#error-handling","title":"Error Handling","text":""},{"location":"contributing/#testing","title":"Testing","text":""},{"location":"contributing/#implementation-patterns","title":"Implementation Patterns","text":""},{"location":"contributing/#prompt-engineering","title":"Prompt Engineering","text":""},{"location":"contributing/#code-quality","title":"Code Quality","text":""},{"location":"contributing/#mcp-integration","title":"MCP Integration","text":""},{"location":"contributing/#mcp-tools","title":"MCP Tools","text":""},{"location":"contributing/#gradio-mcp-server","title":"Gradio MCP Server","text":""},{"location":"contributing/#common-pitfalls","title":"Common Pitfalls","text":"
  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Hallucinated citations: Always validate references
  4. Global mutable state: Use ContextVar or pass via parameters
  5. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
  6. Rate limiting: Always implement for external APIs
  7. Error chaining: Always use from e when raising exceptions
"},{"location":"contributing/#key-principles","title":"Key Principles","text":"
  1. Type Safety First: All code must pass mypy --strict
  2. Async Everything: All I/O must be async
  3. Test-Driven: Write tests before implementation
  4. No Hallucinations: Validate all citations
  5. Graceful Degradation: Support free tier (HF Inference) when no API keys
  6. Lazy Loading: Don't require optional dependencies at import time
  7. Structured Logging: Use structlog, never print()
  8. Error Chaining: Always preserve exception context
"},{"location":"contributing/#pull-request-process","title":"Pull Request Process","text":"
  1. Ensure all checks pass: uv run ruff check src tests && uv run mypy src && uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire
  2. Update documentation if needed
  3. Add tests for new features
  4. Update CHANGELOG if applicable
  5. Request review from maintainers
  6. Address review feedback
  7. Wait for approval before merging
"},{"location":"contributing/#project-structure","title":"Project Structure","text":""},{"location":"contributing/#questions","title":"Questions?","text":"

Thank you for contributing to The DETERMINATOR!

"},{"location":"contributing/code-quality/","title":"Code Quality & Documentation","text":"

This document outlines code quality standards and documentation requirements for The DETERMINATOR.

"},{"location":"contributing/code-quality/#linting","title":"Linting","text":""},{"location":"contributing/code-quality/#type-checking","title":"Type Checking","text":""},{"location":"contributing/code-quality/#pre-commit","title":"Pre-commit","text":"

Pre-commit hooks run automatically on commit to ensure code quality. Configuration is in .pre-commit-config.yaml.

"},{"location":"contributing/code-quality/#installation","title":"Installation","text":"
# Install dependencies (includes pre-commit package)\nuv sync --all-extras\n\n# Set up git hooks (must be run separately)\nuv run pre-commit install\n

Note: uv sync --all-extras installs the pre-commit package, but you must run uv run pre-commit install separately to set up the git hooks.

"},{"location":"contributing/code-quality/#pre-commit-hooks","title":"Pre-commit Hooks","text":"

The following hooks run automatically on commit:

  1. ruff: Lints code and fixes issues automatically
  2. Runs on: src/ (excludes tests/, reference_repos/)

  3. Auto-fixes: Yes

  4. ruff-format: Formats code with ruff

  5. Runs on: src/ (excludes tests/, reference_repos/)

  6. Auto-fixes: Yes

  7. mypy: Type checking

  8. Runs on: src/ (excludes folder/)

  9. Additional dependencies: pydantic, pydantic-settings, tenacity, pydantic-ai

  10. pytest-unit: Runs unit tests (excludes OpenAI and embedding_provider tests)

  11. Runs: tests/unit/ with -m \"not openai and not embedding_provider\"

  12. Always runs: Yes (not just on changed files)

  13. pytest-local-embeddings: Runs local embedding tests

  14. Runs: tests/ with -m \"local_embeddings\"
  15. Always runs: Yes
"},{"location":"contributing/code-quality/#manual-pre-commit-run","title":"Manual Pre-commit Run","text":"

To run pre-commit hooks manually (without committing):

uv run pre-commit run --all-files\n
"},{"location":"contributing/code-quality/#troubleshooting","title":"Troubleshooting","text":""},{"location":"contributing/code-quality/#documentation","title":"Documentation","text":""},{"location":"contributing/code-quality/#building-documentation","title":"Building Documentation","text":"

Documentation is built using MkDocs. Source files are in docs/, and the configuration is in mkdocs.yml.

# Build documentation\nuv run mkdocs build\n\n# Serve documentation locally (http://127.0.0.1:8000)\nuv run mkdocs serve\n

The documentation site is published at: https://deepcritical.github.io/GradioDemo/

"},{"location":"contributing/code-quality/#docstrings","title":"Docstrings","text":"

Example:

"},{"location":"contributing/code-quality/#code-comments","title":"Code Comments","text":""},{"location":"contributing/code-quality/#see-also","title":"See Also","text":""},{"location":"contributing/code-style/","title":"Code Style & Conventions","text":"

This document outlines the code style and conventions for The DETERMINATOR.

"},{"location":"contributing/code-style/#package-manager","title":"Package Manager","text":"

This project uses uv as the package manager. All commands should be prefixed with uv run to ensure they run in the correct environment.

"},{"location":"contributing/code-style/#installation","title":"Installation","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync all dependencies including dev extras\nuv sync --all-extras\n
"},{"location":"contributing/code-style/#running-commands","title":"Running Commands","text":"

All development commands should use uv run prefix:

# Instead of: pytest tests/\nuv run pytest tests/\n\n# Instead of: ruff check src\nuv run ruff check src\n\n# Instead of: mypy src\nuv run mypy src\n

This ensures commands run in the correct virtual environment managed by uv.

"},{"location":"contributing/code-style/#type-safety","title":"Type Safety","text":""},{"location":"contributing/code-style/#pydantic-models","title":"Pydantic Models","text":""},{"location":"contributing/code-style/#async-patterns","title":"Async Patterns","text":" 
loop = asyncio.get_running_loop()\nresult = await loop.run_in_executor(None, cpu_bound_function, args)\n
"},{"location":"contributing/code-style/#common-pitfalls","title":"Common Pitfalls","text":"
  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Global mutable state: Use ContextVar or pass via parameters
  4. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
"},{"location":"contributing/code-style/#see-also","title":"See Also","text":""},{"location":"contributing/error-handling/","title":"Error Handling & Logging","text":"

This document outlines error handling and logging conventions for The DETERMINATOR.

"},{"location":"contributing/error-handling/#exception-hierarchy","title":"Exception Hierarchy","text":"

Use custom exception hierarchy (src/utils/exceptions.py):

"},{"location":"contributing/error-handling/#error-handling-rules","title":"Error Handling Rules","text":" 
logger.error(\"Operation failed\", error=str(e), context=value)\n
"},{"location":"contributing/error-handling/#logging","title":"Logging","text":""},{"location":"contributing/error-handling/#logging-examples","title":"Logging Examples","text":"
logger.info(\"Starting search\", query=query, tools=[t.name for t in tools])\nlogger.warning(\"Search tool failed\", tool=tool.name, error=str(result))\nlogger.error(\"Assessment failed\", error=str(e))\n
"},{"location":"contributing/error-handling/#error-chaining","title":"Error Chaining","text":"

Always preserve exception context:

try:\n    result = await api_call()\nexcept httpx.HTTPError as e:\n    raise SearchError(f\"API call failed: {e}\") from e\n
"},{"location":"contributing/error-handling/#see-also","title":"See Also","text":""},{"location":"contributing/implementation-patterns/","title":"Implementation Patterns","text":"

This document outlines common implementation patterns used in The DETERMINATOR.

"},{"location":"contributing/implementation-patterns/#search-tools","title":"Search Tools","text":"

All tools implement SearchTool protocol (src/tools/base.py):

Example pattern:

class MySearchTool:\n    @property\n    def name(self) -> str:\n        return \"mytool\"\n    \n    @retry(stop=stop_after_attempt(3), wait=wait_exponential(...))\n    async def search(self, query: str, max_results: int = 10) -> list[Evidence]:\n        # Implementation\n        return evidence_list\n
"},{"location":"contributing/implementation-patterns/#judge-handlers","title":"Judge Handlers","text":""},{"location":"contributing/implementation-patterns/#agent-factory-pattern","title":"Agent Factory Pattern","text":""},{"location":"contributing/implementation-patterns/#state-management","title":"State Management","text":""},{"location":"contributing/implementation-patterns/#singleton-pattern","title":"Singleton Pattern","text":"

Use @lru_cache(maxsize=1) for singletons:

"},{"location":"contributing/implementation-patterns/#see-also","title":"See Also","text":""},{"location":"contributing/prompt-engineering/","title":"Prompt Engineering & Citation Validation","text":"

This document outlines prompt engineering guidelines and citation validation rules.

"},{"location":"contributing/prompt-engineering/#judge-prompts","title":"Judge Prompts","text":""},{"location":"contributing/prompt-engineering/#hypothesis-prompts","title":"Hypothesis Prompts","text":""},{"location":"contributing/prompt-engineering/#report-prompts","title":"Report Prompts","text":""},{"location":"contributing/prompt-engineering/#citation-validation","title":"Citation Validation","text":""},{"location":"contributing/prompt-engineering/#citation-validation-rules","title":"Citation Validation Rules","text":"
  1. Every reference URL must EXACTLY match a provided evidence URL
  2. Do NOT invent, fabricate, or hallucinate any references
  3. Do NOT modify paper titles, authors, dates, or URLs
  4. If unsure about a citation, OMIT it rather than guess
  5. Copy URLs exactly as provided - do not create similar-looking URLs
"},{"location":"contributing/prompt-engineering/#evidence-selection","title":"Evidence Selection","text":""},{"location":"contributing/prompt-engineering/#see-also","title":"See Also","text":""},{"location":"contributing/testing/","title":"Testing Requirements","text":"

This document outlines testing requirements and guidelines for The DETERMINATOR.

"},{"location":"contributing/testing/#test-structure","title":"Test Structure","text":""},{"location":"contributing/testing/#test-markers","title":"Test Markers","text":"

The project uses pytest markers to categorize tests. These markers are defined in pyproject.toml:

"},{"location":"contributing/testing/#running-tests-by-marker","title":"Running Tests by Marker","text":"
# Run only unit tests (excludes OpenAI tests by default)\nuv run pytest tests/unit/ -v -m \"not openai\" -p no:logfire\n\n# Run HuggingFace tests\nuv run pytest tests/ -v -m \"huggingface\" -p no:logfire\n\n# Run all tests\nuv run pytest tests/ -v -p no:logfire\n\n# Run only local embedding tests\nuv run pytest tests/ -v -m \"local_embeddings\" -p no:logfire\n\n# Exclude slow tests\nuv run pytest tests/ -v -m \"not slow\" -p no:logfire\n

Note: The -p no:logfire flag disables the logfire plugin to avoid conflicts during testing.

"},{"location":"contributing/testing/#mocking","title":"Mocking","text":""},{"location":"contributing/testing/#tdd-workflow","title":"TDD Workflow","text":"
  1. Write failing test in tests/unit/
  2. Implement in src/
  3. Ensure test passes
  4. Run checks: uv run ruff check src tests && uv run mypy src && uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire
"},{"location":"contributing/testing/#test-command-examples","title":"Test Command Examples","text":"
# Run unit tests (default, excludes OpenAI tests)\nuv run pytest tests/unit/ -v -m \"not openai\" -p no:logfire\n\n# Run HuggingFace tests\nuv run pytest tests/ -v -m \"huggingface\" -p no:logfire\n\n# Run all tests\nuv run pytest tests/ -v -p no:logfire\n
"},{"location":"contributing/testing/#test-examples","title":"Test Examples","text":"
@pytest.mark.unit\nasync def test_pubmed_search(mock_httpx_client):\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=5)\n    assert len(results) > 0\n    assert all(isinstance(r, Evidence) for r in results)\n\n@pytest.mark.integration\nasync def test_real_pubmed_search():\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=3)\n    assert len(results) <= 3\n
"},{"location":"contributing/testing/#test-coverage","title":"Test Coverage","text":""},{"location":"contributing/testing/#terminal-coverage-report","title":"Terminal Coverage Report","text":"
uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire\n

This shows coverage with missing lines highlighted in the terminal output.

"},{"location":"contributing/testing/#html-coverage-report","title":"HTML Coverage Report","text":"
uv run pytest --cov=src --cov-report=html -p no:logfire\n

This generates an HTML coverage report in htmlcov/index.html. Open this file in your browser to see detailed coverage information.

"},{"location":"contributing/testing/#coverage-goals","title":"Coverage Goals","text":""},{"location":"contributing/testing/#see-also","title":"See Also","text":""},{"location":"getting-started/examples/","title":"Examples","text":"

This page provides examples of using The DETERMINATOR for various research tasks.

"},{"location":"getting-started/examples/#basic-research-query","title":"Basic Research Query","text":""},{"location":"getting-started/examples/#example-1-drug-information","title":"Example 1: Drug Information","text":"

Query: 

What are the latest treatments for Alzheimer's disease?\n

What The DETERMINATOR Does: 1. Searches PubMed for recent papers 2. Searches ClinicalTrials.gov for active trials 3. Evaluates evidence quality 4. Synthesizes findings into a comprehensive report

"},{"location":"getting-started/examples/#example-2-clinical-trial-search","title":"Example 2: Clinical Trial Search","text":"

Query: 

What clinical trials are investigating metformin for cancer prevention?\n

What The DETERMINATOR Does:

  1. Searches ClinicalTrials.gov for relevant trials
  2. Searches PubMed for supporting literature
  3. Provides trial details and status
  4. Summarizes findings
"},{"location":"getting-started/examples/#advanced-research-queries","title":"Advanced Research Queries","text":""},{"location":"getting-started/examples/#example-3-comprehensive-review","title":"Example 3: Comprehensive Review","text":"

Query:

Review the evidence for using metformin as an anti-aging intervention, \nincluding clinical trials, mechanisms of action, and safety profile.\n

What The DETERMINATOR Does: 1. Uses deep research mode (multi-section) 2. Searches multiple sources in parallel 3. Generates sections on: - Clinical trials - Mechanisms of action - Safety profile 4. Synthesizes comprehensive report

"},{"location":"getting-started/examples/#example-4-hypothesis-testing","title":"Example 4: Hypothesis Testing","text":"

Query: 

Test the hypothesis that regular exercise reduces Alzheimer's disease risk.\n

What The DETERMINATOR Does: 1. Generates testable hypotheses 2. Searches for supporting/contradicting evidence 3. Performs statistical analysis (if Modal configured) 4. Provides verdict: SUPPORTED, REFUTED, or INCONCLUSIVE

"},{"location":"getting-started/examples/#mcp-tool-examples","title":"MCP Tool Examples","text":""},{"location":"getting-started/examples/#using-search_pubmed","title":"Using search_pubmed","text":"
Search PubMed for \"CRISPR gene editing cancer therapy\"\n
"},{"location":"getting-started/examples/#using-search_clinical_trials","title":"Using search_clinical_trials","text":"
Find active clinical trials for \"diabetes type 2 treatment\"\n
"},{"location":"getting-started/examples/#using-search_all","title":"Using search_all","text":"
Search all sources for \"COVID-19 vaccine side effects\"\n
"},{"location":"getting-started/examples/#using-analyze_hypothesis","title":"Using analyze_hypothesis","text":"
Analyze whether vitamin D supplementation reduces COVID-19 severity\n
"},{"location":"getting-started/examples/#code-examples","title":"Code Examples","text":""},{"location":"getting-started/examples/#python-api-usage","title":"Python API Usage","text":"
from src.orchestrator_factory import create_orchestrator\nfrom src.tools.search_handler import SearchHandler\nfrom src.agent_factory.judges import create_judge_handler\n\n# Create orchestrator\nsearch_handler = SearchHandler()\njudge_handler = create_judge_handler()\n
# Run research query\nquery = \"What are the latest treatments for Alzheimer's disease?\"\nasync for event in orchestrator.run(query):\n    print(f\"Event: {event.type} - {event.data}\")\n
"},{"location":"getting-started/examples/#gradio-ui-integration","title":"Gradio UI Integration","text":"
import gradio as gr\nfrom src.app import create_research_interface\n\n# Create interface\ninterface = create_research_interface()\n\n# Launch\ninterface.launch(server_name=\"0.0.0.0\", server_port=7860)\n
"},{"location":"getting-started/examples/#research-patterns","title":"Research Patterns","text":""},{"location":"getting-started/examples/#iterative-research","title":"Iterative Research","text":"

Single-loop research with search-judge-synthesize cycles:

from src.orchestrator.research_flow import IterativeResearchFlow\n
async for event in flow.run(query):\n    # Handle events\n    pass\n
"},{"location":"getting-started/examples/#deep-research","title":"Deep Research","text":"

Multi-section parallel research:

from src.orchestrator.research_flow import DeepResearchFlow\n
async for event in flow.run(query):\n    # Handle events\n    pass\n
"},{"location":"getting-started/examples/#configuration-examples","title":"Configuration Examples","text":""},{"location":"getting-started/examples/#basic-configuration","title":"Basic Configuration","text":"
# .env file\nLLM_PROVIDER=openai\nOPENAI_API_KEY=your_key_here\nMAX_ITERATIONS=10\n
"},{"location":"getting-started/examples/#advanced-configuration","title":"Advanced Configuration","text":"
# .env file\nLLM_PROVIDER=anthropic\nANTHROPIC_API_KEY=your_key_here\nEMBEDDING_PROVIDER=local\nWEB_SEARCH_PROVIDER=duckduckgo\nMAX_ITERATIONS=20\nDEFAULT_TOKEN_LIMIT=200000\nUSE_GRAPH_EXECUTION=true\n
"},{"location":"getting-started/examples/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/installation/","title":"Installation","text":"

This guide will help you install and set up DeepCritical on your system.

"},{"location":"getting-started/installation/#prerequisites","title":"Prerequisites","text":""},{"location":"getting-started/installation/#installation-steps","title":"Installation Steps","text":""},{"location":"getting-started/installation/#1-install-uv-recommended","title":"1. Install uv (Recommended)","text":"

uv is a fast Python package installer and resolver. Install it using the standalone installer (recommended):

Unix/macOS/Linux: 

curl -LsSf https://astral.sh/uv/install.sh | sh\n

Windows (PowerShell): 

powershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n

Alternative methods: 

# Using pipx (recommended if you have pipx installed)\npipx install uv\n\n# Or using pip\npip install uv\n

After installation, restart your terminal or add ~/.cargo/bin to your PATH.

"},{"location":"getting-started/installation/#2-clone-the-repository","title":"2. Clone the Repository","text":"
git clone https://github.com/DeepCritical/GradioDemo.git\ncd GradioDemo\n
"},{"location":"getting-started/installation/#3-install-dependencies","title":"3. Install Dependencies","text":"

Using uv (recommended):

uv sync\n

Using pip:

pip install -e .\n
"},{"location":"getting-started/installation/#4-install-optional-dependencies","title":"4. Install Optional Dependencies","text":"

For embeddings support (local sentence-transformers):

uv sync --extra embeddings\n

For Modal sandbox execution:

uv sync --extra modal\n

For Magentic orchestration:

uv sync --extra magentic\n

Install all extras:

uv sync --all-extras\n
"},{"location":"getting-started/installation/#5-configure-environment-variables","title":"5. Configure Environment Variables","text":"

Create a .env file in the project root:

# Required: At least one LLM provider\nLLM_PROVIDER=openai  # or \"anthropic\" or \"huggingface\"\nOPENAI_API_KEY=your_openai_api_key_here\n\n# Optional: Other services\nNCBI_API_KEY=your_ncbi_api_key_here  # For higher PubMed rate limits\nMODAL_TOKEN_ID=your_modal_token_id\nMODAL_TOKEN_SECRET=your_modal_token_secret\n

See the Configuration Guide for all available options.

"},{"location":"getting-started/installation/#6-verify-installation","title":"6. Verify Installation","text":"

Run the application:

uv run gradio run src/app.py\n

Open your browser to http://localhost:7860 to verify the installation.

"},{"location":"getting-started/installation/#development-setup","title":"Development Setup","text":"

For development, install dev dependencies:

uv sync --all-extras --dev\n

Install pre-commit hooks:

uv run pre-commit install\n
"},{"location":"getting-started/installation/#troubleshooting","title":"Troubleshooting","text":""},{"location":"getting-started/installation/#common-issues","title":"Common Issues","text":"

Import Errors: - Ensure you've installed all required dependencies - Check that Python 3.11+ is being used

API Key Errors: - Verify your .env file is in the project root - Check that API keys are correctly formatted - Ensure at least one LLM provider is configured

Module Not Found: - Run uv sync or pip install -e . again - Check that you're in the correct virtual environment

Port Already in Use: - Change the port in src/app.py or use environment variable - Kill the process using port 7860

"},{"location":"getting-started/installation/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/mcp-integration/","title":"MCP Integration","text":"

The DETERMINATOR exposes a Model Context Protocol (MCP) server, allowing you to use its search tools directly from Claude Desktop or other MCP clients.

"},{"location":"getting-started/mcp-integration/#what-is-mcp","title":"What is MCP?","text":"

The Model Context Protocol (MCP) is a standard for connecting AI assistants to external tools and data sources. The DETERMINATOR implements an MCP server that exposes its search capabilities as MCP tools.

"},{"location":"getting-started/mcp-integration/#mcp-server-url","title":"MCP Server URL","text":"

When running locally:

http://localhost:7860/gradio_api/mcp/\n
"},{"location":"getting-started/mcp-integration/#claude-desktop-configuration","title":"Claude Desktop Configuration","text":""},{"location":"getting-started/mcp-integration/#1-locate-configuration-file","title":"1. Locate Configuration File","text":"

macOS: 

~/Library/Application Support/Claude/claude_desktop_config.json\n

Windows: 

%APPDATA%\\Claude\\claude_desktop_config.json\n

Linux: 

~/.config/Claude/claude_desktop_config.json\n

"},{"location":"getting-started/mcp-integration/#2-add-the-determinator-server","title":"2. Add The DETERMINATOR Server","text":"

Edit claude_desktop_config.json and add:

{\n  \"mcpServers\": {\n    \"determinator\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#3-restart-claude-desktop","title":"3. Restart Claude Desktop","text":"

Close and restart Claude Desktop for changes to take effect.

"},{"location":"getting-started/mcp-integration/#4-verify-connection","title":"4. Verify Connection","text":"

In Claude Desktop, you should see The DETERMINATOR tools available: - search_pubmed - search_clinical_trials - search_biorxiv - search_all - analyze_hypothesis

"},{"location":"getting-started/mcp-integration/#available-tools","title":"Available Tools","text":""},{"location":"getting-started/mcp-integration/#search_pubmed","title":"search_pubmed","text":"

Search peer-reviewed biomedical literature from PubMed.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search PubMed for \"metformin diabetes\"\n

"},{"location":"getting-started/mcp-integration/#search_clinical_trials","title":"search_clinical_trials","text":"

Search ClinicalTrials.gov for interventional studies.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search clinical trials for \"Alzheimer's disease treatment\"\n

"},{"location":"getting-started/mcp-integration/#search_biorxiv","title":"search_biorxiv","text":"

Search bioRxiv/medRxiv preprints via Europe PMC.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search bioRxiv for \"CRISPR gene editing\"\n

"},{"location":"getting-started/mcp-integration/#search_all","title":"search_all","text":"

Search all sources simultaneously (PubMed, ClinicalTrials.gov, Europe PMC).

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results per source (default: 10)

Example: 

Search all sources for \"COVID-19 vaccine efficacy\"\n

"},{"location":"getting-started/mcp-integration/#analyze_hypothesis","title":"analyze_hypothesis","text":"

Perform secure statistical analysis using Modal sandboxes.

Parameters: - hypothesis (string): Hypothesis to analyze - data (string, optional): Data description or code

Example: 

Analyze the hypothesis that metformin reduces cancer risk\n

"},{"location":"getting-started/mcp-integration/#using-tools-in-claude-desktop","title":"Using Tools in Claude Desktop","text":"

Once configured, you can ask Claude to use DeepCritical tools:

Use DeepCritical to search PubMed for recent papers on Alzheimer's disease treatments.\n

Claude will automatically: 1. Call the appropriate DeepCritical tool 2. Retrieve results 3. Use the results in its response

"},{"location":"getting-started/mcp-integration/#troubleshooting","title":"Troubleshooting","text":""},{"location":"getting-started/mcp-integration/#connection-issues","title":"Connection Issues","text":"

Server Not Found: - Ensure DeepCritical is running (uv run gradio run src/app.py) - Verify the URL in claude_desktop_config.json is correct - Check that port 7860 is not blocked by firewall

Tools Not Appearing: - Restart Claude Desktop after configuration changes - Check Claude Desktop logs for errors - Verify MCP server is accessible at the configured URL

"},{"location":"getting-started/mcp-integration/#authentication","title":"Authentication","text":"

If DeepCritical requires authentication: - Configure API keys in DeepCritical settings - Use HuggingFace OAuth login - Ensure API keys are valid

"},{"location":"getting-started/mcp-integration/#advanced-configuration","title":"Advanced Configuration","text":""},{"location":"getting-started/mcp-integration/#custom-port","title":"Custom Port","text":"

If running on a different port, update the URL:

{\n  \"mcpServers\": {\n    \"deepcritical\": {\n      \"url\": \"http://localhost:8080/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#multiple-instances","title":"Multiple Instances","text":"

You can configure multiple DeepCritical instances:

{\n  \"mcpServers\": {\n    \"deepcritical-local\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    },\n    \"deepcritical-remote\": {\n      \"url\": \"https://your-server.com/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/quick-start/","title":"Single Command Deploy","text":"

Deploy with docker instandly with a single command :

docker run -it -p 7860:7860 --platform=linux/amd64 \\\n    -e DB_KEY=\"YOUR_VALUE_HERE\" \\\n    -e SERP_API=\"YOUR_VALUE_HERE\" \\\n    -e INFERENCE_API=\"YOUR_VALUE_HERE\" \\\n    -e MODAL_TOKEN_ID=\"YOUR_VALUE_HERE\" \\\n    -e MODAL_TOKEN_SECRET=\"YOUR_VALUE_HERE\" \\\n    -e NCBI_API_KEY=\"YOUR_VALUE_HERE\" \\\n    -e SERPER_API_KEY=\"YOUR_VALUE_HERE\" \\\n    -e CHROMA_DB_PATH=\"./chroma_db\" \\\n    -e CHROMA_DB_HOST=\"localhost\" \\\n    -e CHROMA_DB_PORT=\"8000\" \\\n    -e RAG_COLLECTION_NAME=\"deepcritical_evidence\" \\\n    -e RAG_SIMILARITY_TOP_K=\"5\" \\\n    -e RAG_AUTO_INGEST=\"true\" \\\n    -e USE_GRAPH_EXECUTION=\"false\" \\\n    -e DEFAULT_TOKEN_LIMIT=\"100000\" \\\n    -e DEFAULT_TIME_LIMIT_MINUTES=\"10\" \\\n    -e DEFAULT_ITERATIONS_LIMIT=\"10\" \\\n    -e WEB_SEARCH_PROVIDER=\"duckduckgo\" \\\n    -e MAX_ITERATIONS=\"10\" \\\n    -e SEARCH_TIMEOUT=\"30\" \\\n    -e LOG_LEVEL=\"DEBUG\" \\\n    -e EMBEDDING_PROVIDER=\"local\" \\\n    -e OPENAI_EMBEDDING_MODEL=\"text-embedding-3-small\" \\\n    -e LOCAL_EMBEDDING_MODEL=\"BAAI/bge-small-en-v1.5\" \\\n    -e HUGGINGFACE_EMBEDDING_MODEL=\"sentence-transformers/all-MiniLM-L6-v2\" \\\n    -e HF_FALLBACK_MODELS=\"Qwen/Qwen3-Next-80B-A3B-Thinking,Qwen/Qwen3-Next-80B-A3B-Instruct,meta-llama/Llama-3.3-70B-Instruct,meta-llama/Llama-3.1-8B-Instruct,HuggingFaceH4/zephyr-7b-beta,Qwen/Qwen2-7B-Instruct\" \\\n    -e HUGGINGFACE_MODEL=\"Qwen/Qwen3-Next-80B-A3B-Thinking\" \\\n    registry.hf.space/dataquests-deepcritical:latest python src/app.py\n   ```\n\n## Quick start guide\n\nGet up and running with The DETERMINATOR in minutes.\n\n## Start the Application\n\n```bash\ngradio src/app.py\n

Open your browser to http://localhost:7860.

"},{"location":"getting-started/quick-start/#first-research-query","title":"First Research Query","text":"
  1. Enter a Research Question

Type your research question in the chat interface, for example: - \"What are the latest treatments for Alzheimer's disease?\" - \"Review the evidence for metformin in cancer prevention\" - \"What clinical trials are investigating COVID-19 vaccines?\"

  1. Submit the Query

Click \"Submit\" or press Enter. The system will: - Generate observations about your query - Identify knowledge gaps - Search multiple sources (PubMed, ClinicalTrials.gov, Europe PMC) - Evaluate evidence quality - Synthesize findings into a report

  1. Review Results

Watch the real-time progress in the chat interface: - Search operations and results - Evidence evaluation - Report generation - Final research report with citations

"},{"location":"getting-started/quick-start/#authentication","title":"Authentication","text":""},{"location":"getting-started/quick-start/#huggingface-oauth-recommended","title":"HuggingFace OAuth (Recommended)","text":"
  1. Click \"Sign in with HuggingFace\" at the top of the app
  2. Authorize the application
  3. Your HuggingFace API token will be automatically used
  4. No need to manually enter API keys
"},{"location":"getting-started/quick-start/#manual-api-key","title":"Manual API Key","text":"
  1. Open the Settings accordion
  2. Enter your API key:
  3. OpenAI API key
  4. Anthropic API key
  5. HuggingFace API key
  6. Click \"Save Settings\"
  7. Manual keys take priority over OAuth tokens
"},{"location":"getting-started/quick-start/#understanding-the-interface","title":"Understanding the Interface","text":""},{"location":"getting-started/quick-start/#chat-interface","title":"Chat Interface","text":""},{"location":"getting-started/quick-start/#status-indicators","title":"Status Indicators","text":""},{"location":"getting-started/quick-start/#settings","title":"Settings","text":""},{"location":"getting-started/quick-start/#example-queries","title":"Example Queries","text":""},{"location":"getting-started/quick-start/#simple-query","title":"Simple Query","text":"
What are the side effects of metformin?\n
"},{"location":"getting-started/quick-start/#complex-query","title":"Complex Query","text":"
Review the evidence for using metformin as an anti-aging intervention, \nincluding clinical trials, mechanisms of action, and safety profile.\n
"},{"location":"getting-started/quick-start/#clinical-trial-query","title":"Clinical Trial Query","text":"
What are the active clinical trials investigating Alzheimer's disease treatments?\n
"},{"location":"getting-started/quick-start/#next-steps","title":"Next Steps","text":""},{"location":"overview/architecture/","title":"Architecture Overview","text":"

The DETERMINATOR is a powerful generalist deep research agent system that uses iterative search-and-judge loops to comprehensively investigate any research question. It stops at nothing until finding precise answers, only stopping at configured limits (budget, time, iterations). The system automatically determines if medical knowledge sources are needed and adapts its search strategy accordingly. It supports multiple orchestration patterns, graph-based execution, parallel research workflows, and long-running task management with real-time streaming.

"},{"location":"overview/architecture/#core-architecture","title":"Core Architecture","text":""},{"location":"overview/architecture/#orchestration-patterns","title":"Orchestration Patterns","text":"
  1. Graph Orchestrator (src/orchestrator/graph_orchestrator.py):
  2. Graph-based execution using Pydantic AI agents as nodes
  3. Supports both iterative and deep research patterns
  4. Node types: Agent, State, Decision, Parallel
  5. Edge types: Sequential, Conditional, Parallel
  6. Conditional routing based on knowledge gaps, budget, and iterations
  7. Parallel execution for concurrent research loops
  8. Event streaming via AsyncGenerator[AgentEvent] for real-time UI updates

  9. Fallback to agent chains when graph execution is disabled

  10. Deep Research Flow (src/orchestrator/research_flow.py):

  11. Pattern: Planner \u2192 Parallel Iterative Loops (one per section) \u2192 Synthesis
  12. Uses PlannerAgent to break query into report sections
  13. Runs IterativeResearchFlow instances in parallel per section via WorkflowManager
  14. Synthesizes results using LongWriterAgent or ProofreaderAgent
  15. Supports both graph execution (use_graph=True) and agent chains (use_graph=False)
  16. Budget tracking per section and globally

  17. State synchronization across parallel loops

  18. Iterative Research Flow (src/orchestrator/research_flow.py):

  19. Pattern: Generate observations \u2192 Evaluate gaps \u2192 Select tools \u2192 Execute \u2192 Judge \u2192 Continue/Complete
  20. Uses KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent, WriterAgent
  21. JudgeHandler assesses evidence sufficiency
  22. Iterates until research complete or constraints met (iterations, time, tokens)

  23. Supports graph execution and agent chains

  24. Magentic Orchestrator (src/orchestrator_magentic.py):

  25. Multi-agent coordination using agent-framework-core
  26. ChatAgent pattern with internal LLMs per agent
  27. Uses MagenticBuilder with participants: searcher, hypothesizer, judge, reporter
  28. Manager orchestrates agents via OpenAIChatClient
  29. Requires OpenAI API key (function calling support)
  30. Event-driven: converts Magentic events to AgentEvent for UI streaming

  31. Supports long-running workflows with max rounds and stall/reset handling

  32. Hierarchical Orchestrator (src/orchestrator_hierarchical.py):

  33. Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge
  34. Adapts Magentic ChatAgent to SubIterationTeam protocol
  35. Event-driven via asyncio.Queue for coordination

  36. Supports sub-iteration patterns for complex research tasks

  37. Legacy Simple Mode (src/legacy_orchestrator.py):

  38. Linear search-judge-synthesize loop
  39. Uses SearchHandlerProtocol and JudgeHandlerProtocol
  40. Generator-based design yielding AgentEvent objects
  41. Backward compatibility for simple use cases
"},{"location":"overview/architecture/#long-running-task-support","title":"Long-Running Task Support","text":"

The system is designed for long-running research tasks with comprehensive state management and streaming:

  1. Event Streaming:
  2. All orchestrators yield AgentEvent objects via AsyncGenerator
  3. Real-time UI updates through Gradio chat interface
  4. Event types: started, searching, search_complete, judging, judge_complete, looping, synthesizing, hypothesizing, complete, error

  5. Metadata includes iteration numbers, tool names, result counts, durations

  6. Budget Tracking (src/middleware/budget_tracker.py):

  7. Per-loop and global budget management
  8. Tracks: tokens, time (seconds), iterations
  9. Budget enforcement at decision nodes
  10. Token estimation (~4 chars per token)
  11. Early termination when budgets exceeded

  12. Budget summaries for monitoring

  13. Workflow Manager (src/middleware/workflow_manager.py):

  14. Coordinates parallel research loops
  15. Tracks loop status: pending, running, completed, failed, cancelled
  16. Synchronizes evidence between loops and global state
  17. Handles errors per loop (doesn't fail all if one fails)
  18. Supports loop cancellation and timeout handling

  19. Evidence deduplication across parallel loops

  20. State Management (src/middleware/state_machine.py):

  21. Thread-safe isolation using ContextVar for concurrent requests
  22. WorkflowState tracks: evidence, conversation history, embedding service
  23. Evidence deduplication by URL
  24. Semantic search via embedding service
  25. State persistence across long-running workflows

  26. Supports both iterative and deep research patterns

  27. Gradio UI (src/app.py):

  28. Real-time streaming of research progress
  29. Accordion-based UI for pending/done operations
  30. OAuth integration (HuggingFace)
  31. Multiple backend support (API keys, free tier)
  32. Handles long-running tasks with progress indicators
  33. Event accumulation for pending operations
"},{"location":"overview/architecture/#graph-architecture","title":"Graph Architecture","text":"

The graph orchestrator (src/orchestrator/graph_orchestrator.py) implements a flexible graph-based execution model:

Node Types:

Edge Types:

Graph Patterns:

Execution Flow:

  1. Graph construction from nodes and edges
  2. Graph validation (no cycles, all nodes reachable)
  3. Graph execution from entry node
  4. Node execution based on type
  5. Edge evaluation for next node(s)
  6. Parallel execution via asyncio.gather()
  7. State updates at state nodes
  8. Event streaming for UI
"},{"location":"overview/architecture/#key-components","title":"Key Components","text":""},{"location":"overview/architecture/#research-team--parallel-execution","title":"Research Team & Parallel Execution","text":"

The system supports complex research workflows through:

  1. WorkflowManager: Coordinates multiple parallel research loops
  2. Creates and tracks ResearchLoop instances
  3. Runs loops in parallel via asyncio.gather()
  4. Synchronizes evidence to global state

  5. Handles loop failures gracefully

  6. Deep Research Pattern: Breaks complex queries into sections

  7. Planner creates report outline with sections
  8. Each section runs as independent iterative research loop
  9. Loops execute in parallel
  10. Evidence shared across loops via global state

  11. Final synthesis combines all section results

  12. State Synchronization: Thread-safe evidence sharing

  13. Evidence deduplication by URL
  14. Global state accessible to all loops
  15. Semantic search across all collected evidence
  16. Conversation history tracking per iteration
"},{"location":"overview/architecture/#configuration--modes","title":"Configuration & Modes","text":"

Note: The UI provides separate controls for orchestrator mode and graph research mode. When using graph-based orchestrators (iterative/deep/auto), the graph research mode determines the specific pattern used within the graph execution.

"},{"location":"overview/features/","title":"Features","text":"

The DETERMINATOR provides a comprehensive set of features for AI-assisted research:

"},{"location":"overview/features/#core-features","title":"Core Features","text":""},{"location":"overview/features/#multi-source-search","title":"Multi-Source Search","text":""},{"location":"overview/features/#mcp-integration","title":"MCP Integration","text":""},{"location":"overview/features/#authentication","title":"Authentication","text":""},{"location":"overview/features/#secure-code-execution","title":"Secure Code Execution","text":""},{"location":"overview/features/#semantic-search--rag","title":"Semantic Search & RAG","text":""},{"location":"overview/features/#orchestration-patterns","title":"Orchestration Patterns","text":"

Orchestrator Modes: - simple: Legacy linear search-judge loop - advanced (or magentic): Multi-agent coordination (requires OpenAI API key) - iterative: Knowledge-gap-driven research with single loop - deep: Parallel section-based research with planning - auto: Intelligent mode detection based on query complexity

Graph Research Modes (used within graph orchestrator): - iterative: Single research loop pattern - deep: Multi-section parallel research pattern - auto: Auto-detect pattern based on query complexity

Execution Modes: - use_graph=True: Graph-based execution with parallel and conditional routing - use_graph=False: Agent chains with sequential execution (backward compatible)

"},{"location":"overview/features/#real-time-streaming","title":"Real-Time Streaming","text":""},{"location":"overview/features/#budget-management","title":"Budget Management","text":""},{"location":"overview/features/#state-management","title":"State Management","text":""},{"location":"overview/features/#multimodal-input--output","title":"Multimodal Input & Output","text":""},{"location":"overview/features/#advanced-features","title":"Advanced Features","text":""},{"location":"overview/features/#agent-system","title":"Agent System","text":""},{"location":"overview/features/#search-tools","title":"Search Tools","text":""},{"location":"overview/features/#error-handling","title":"Error Handling","text":""},{"location":"overview/features/#configuration","title":"Configuration","text":""},{"location":"overview/features/#testing","title":"Testing","text":""},{"location":"overview/features/#ui-features","title":"UI Features","text":""},{"location":"overview/features/#gradio-interface","title":"Gradio Interface","text":""},{"location":"overview/features/#mcp-server","title":"MCP Server","text":""},{"location":"overview/features/#development-features","title":"Development Features","text":""},{"location":"overview/features/#code-quality","title":"Code Quality","text":""},{"location":"overview/features/#documentation","title":"Documentation","text":""},{"location":"overview/quick-start/","title":"Quick Start","text":"

Get started with DeepCritical in minutes.

"},{"location":"overview/quick-start/#installation","title":"Installation","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync dependencies\nuv sync\n
"},{"location":"overview/quick-start/#run-the-ui","title":"Run the UI","text":"
# Start the Gradio app\nuv run gradio run src/app.py\n

Open your browser to http://localhost:7860.

"},{"location":"overview/quick-start/#basic-usage","title":"Basic Usage","text":""},{"location":"overview/quick-start/#1-authentication-required","title":"1. Authentication (REQUIRED)","text":"

Authentication is mandatory - you must authenticate before using the application. The app will display an error message if you try to use it without authentication.

HuggingFace OAuth Login (Recommended): - Click the \"Sign in with HuggingFace\" button at the top of the app - Your HuggingFace API token will be automatically used for AI inference - No need to manually enter API keys when logged in

Manual API Key (Alternative): - Set environment variable HF_TOKEN or HUGGINGFACE_API_KEY before starting the app - The app will automatically use these tokens if OAuth login is not available - Supports HuggingFace API keys only (OpenAI/Anthropic keys are not used in the current implementation)

"},{"location":"overview/quick-start/#2-start-a-research-query","title":"2. Start a Research Query","text":"
  1. Enter your research question in the chat interface
  2. Text Input: Type your question directly
  3. Image Input: Click the \ud83d\udcf7 icon to upload images (OCR will extract text)
  4. Audio Input: Click the \ud83c\udfa4 icon to record or upload audio (STT will transcribe to text)
  5. Click \"Submit\" or press Enter
  6. Watch the real-time progress as the system:
  7. Generates observations
  8. Identifies knowledge gaps
  9. Searches multiple sources
  10. Evaluates evidence
  11. Synthesizes findings
  12. Review the final research report
  13. Audio Output: If enabled, the final response will include audio synthesis (TTS)

Multimodal Features: - Configure image/audio input and output in the sidebar settings - Image OCR and audio STT/TTS can be enabled/disabled independently - TTS voice and speed can be customized in the Audio Output settings

"},{"location":"overview/quick-start/#3-mcp-integration-optional","title":"3. MCP Integration (Optional)","text":"

Connect DeepCritical to Claude Desktop:

  1. Add to your claude_desktop_config.json: 

    {\n  \"mcpServers\": {\n    \"deepcritical\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    }\n  }\n}\n

  2. Restart Claude Desktop

  3. Use DeepCritical tools directly from Claude Desktop
"},{"location":"overview/quick-start/#available-tools","title":"Available Tools","text":"

Note: The application automatically uses all available search tools (Neo4j, PubMed, ClinicalTrials.gov, Europe PMC, Web search, RAG) based on query analysis. Neo4j knowledge graph search is included by default for biomedical queries.

"},{"location":"overview/quick-start/#next-steps","title":"Next Steps","text":""}]} \ No newline at end of file +{"config":{"lang":["en"],"separator":"[\\s\\-,:!=\\[\\]()\"`/]+|\\.(?!\\d)|&[lg]t;|&","pipeline":["stopWordFilter"],"fields":{"title":{"boost":1000.0},"text":{"boost":1.0},"tags":{"boost":1000000.0}}},"docs":[{"location":"","title":"The DETERMINATOR","text":"

Generalist Deep Research Agent - Stops at Nothing Until Finding Precise Answers

The DETERMINATOR is a powerful generalist deep research agent system that uses iterative search-and-judge loops to comprehensively investigate any research question. It stops at nothing until finding precise answers, only stopping at configured limits (budget, time, iterations).

Key Features: - Generalist: Handles queries from any domain (medical, technical, business, scientific, etc.) - Automatic Source Selection: Automatically determines if medical knowledge sources (PubMed, ClinicalTrials.gov) are needed - Multi-Source Search: Web search, PubMed, ClinicalTrials.gov, Europe PMC, RAG - Iterative Refinement: Continues searching and refining until precise answers are found - Evidence Synthesis: Comprehensive reports with proper citations

Important: The DETERMINATOR is a research tool that synthesizes evidence. It cannot provide medical advice or answer medical questions directly.

"},{"location":"#features","title":"Features","text":""},{"location":"#quick-start","title":"Quick Start","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync dependencies\nuv sync\n\n# Start the Gradio app\nuv run gradio run src/app.py\n

Open your browser to http://localhost:7860.

For detailed installation and setup instructions, see the Getting Started Guide.

"},{"location":"#architecture","title":"Architecture","text":"

The DETERMINATOR uses a Vertical Slice Architecture:

  1. Search Slice: Retrieving evidence from multiple sources (web, PubMed, ClinicalTrials.gov, Europe PMC, RAG) based on query analysis
  2. Judge Slice: Evaluating evidence quality using LLMs
  3. Orchestrator Slice: Managing the research loop and UI

The system supports three main research patterns:

Learn more about the Architecture.

"},{"location":"#documentation","title":"Documentation","text":""},{"location":"#links","title":"Links","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/","title":"MkDocs & Material UI Improvement Assessment","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#current-configuration-analysis","title":"Current Configuration Analysis","text":"

Your current mkdocs.yml already includes many excellent features: - \u2705 Material theme with light/dark mode toggle - \u2705 Navigation tabs, sections, expand, and top navigation - \u2705 Search with suggestions and highlighting - \u2705 Code annotation and copy buttons - \u2705 Mermaid diagram support - \u2705 Code include plugin - \u2705 Minification for performance - \u2705 Comprehensive markdown extensions

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#recommended-improvements","title":"Recommended Improvements","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#1-versioning--multi-version-documentation--high-priority","title":"1. Versioning & Multi-Version Documentation \u2b50 High Priority","text":"

If you plan to maintain multiple versions or branches:

plugins:\n  - search\n  - mermaid2\n  - codeinclude\n  - minify:\n      minify_html: true\n      minify_js: true\n      minify_css: true\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n  # Optional: For versioning\n  # - versioning:\n  #     version: ['dev', 'main']\n

Benefits: Shows when pages were last updated, helps users understand document freshness.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#2-git-integration--revision-information--high-priority","title":"2. Git Integration & Revision Information \u2b50 High Priority","text":"

Add revision dates and authors to pages:

plugins:\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n      fallback_to_build_date: true\n  - git-committers:\n      repository: DeepCritical/GradioDemo\n      branch: dev\n

Benefits: Users see when content was last updated, builds trust in documentation freshness.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#3-enhanced-navigation-features--high-priority","title":"3. Enhanced Navigation Features \u2b50 High Priority","text":"

Add breadcrumbs and improve navigation:

theme:\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.indexes  # Add index pages\n    - navigation.instant  # Instant page loads\n    - navigation.tracking  # Track scroll position\n    - navigation.smooth  # Smooth scrolling\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link  # Link to specific tabs\n    - content.tooltips  # Tooltips for abbreviations\n

Benefits: Better UX, easier navigation, professional feel.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#4-content-tabs-for-code-examples--high-priority","title":"4. Content Tabs for Code Examples \u2b50 High Priority","text":"

Perfect for showing multiple code examples (Python, TypeScript, etc.):

markdown_extensions:\n  - pymdownx.tabbed:\n      alternate_style: true\n      combine_header_slug: true  # Add this\n

Usage in docs: 

=== \"Python\"\n    ```python\n    def example():\n        pass\n    ```\n\n=== \"TypeScript\"\n    ```typescript\n    function example() {}\n    ```\n

Benefits: Clean way to show multiple implementations without cluttering pages.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#5-enhanced-admonitions--medium-priority","title":"5. Enhanced Admonitions \u2b50 Medium Priority","text":"

Add more admonition types and better styling:

markdown_extensions:\n  - admonition\n  - pymdownx.details\n  - pymdownx.superfences:\n      custom_fences:\n        - name: mermaid\n          class: mermaid\n          format: !!python/name:pymdownx.superfences.fence_code_format\n        # Add custom admonition fences\n        - name: danger\n          class: danger\n          format: !!python/name:pymdownx.superfences.fence_code_format\n

Usage: 

!!! danger \"Important\"\n    This is a critical warning.\n

Benefits: Better visual hierarchy for warnings, tips, and important information.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#6-math-formula-support--medium-priority-if-needed","title":"6. Math Formula Support \u2b50 Medium Priority (if needed)","text":"

If your documentation includes mathematical formulas:

markdown_extensions:\n  - pymdownx.arithmatex:\n      generic: true\n  - pymdownx.superfences:\n      custom_fences:\n        - name: math\n          class: arithmetic\n          format: !!python/name:pymdownx.superfences.fence_code_format\n\nextra_javascript:\n  - javascripts/mathjax.js\n  - https://polyfill.io/v3/polyfill.min.js?features=es6\n  - https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js\n

Benefits: Essential for scientific/technical documentation with formulas.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#7-better-code-highlighting--medium-priority","title":"7. Better Code Highlighting \u2b50 Medium Priority","text":"

Add more language support and better themes:

markdown_extensions:\n  - pymdownx.highlight:\n      anchor_linenums: true\n      line_spans: __span\n      pygments_lang_class: true\n      use_pygments: true\n      noclasses: false  # Use CSS classes instead of inline styles\n

Benefits: Better syntax highlighting, more language support.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#8-social-links-enhancement--low-priority","title":"8. Social Links Enhancement \u2b50 Low Priority","text":"

Add more social platforms and better icons:

extra:\n  social:\n    - icon: fontawesome/brands/github\n      link: https://github.com/DeepCritical/GradioDemo\n      name: GitHub\n    - icon: fontawesome/brands/twitter\n      link: https://twitter.com/yourhandle\n      name: Twitter\n    - icon: material/web\n      link: https://huggingface.co/spaces/DataQuests/DeepCritical\n      name: HuggingFace Space\n    - icon: fontawesome/brands/discord\n      link: https://discord.gg/yourserver\n      name: Discord\n

Benefits: Better community engagement, more ways to connect.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#9-analytics-integration--medium-priority","title":"9. Analytics Integration \u2b50 Medium Priority","text":"

Add privacy-respecting analytics:

extra:\n  analytics:\n    provider: google\n    property: G-XXXXXXXXXX\n  # Or use privacy-focused alternative:\n  # analytics:\n  #   provider: plausible\n  #   domain: yourdomain.com\n

Benefits: Understand how users interact with your documentation.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#10-custom-cssjs-for-branding--low-priority","title":"10. Custom CSS/JS for Branding \u2b50 Low Priority","text":"

Add custom styling:

extra_css:\n  - stylesheets/extra.css\n\nextra_javascript:\n  - javascripts/extra.js\n

Benefits: Customize appearance, add interactive features.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#11-better-table-of-contents--medium-priority","title":"11. Better Table of Contents \u2b50 Medium Priority","text":"

Enhance TOC with more options:

markdown_extensions:\n  - toc:\n      permalink: true\n      permalink_title: \"Anchor link to this section\"\n      baselevel: 1\n      toc_depth: 3\n      slugify: !!python/object/apply:pymdownx.slugs.slugify\n        kwds:\n          case: lower\n

Benefits: Better navigation within long pages, SEO-friendly anchor links.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#12-image-optimization--medium-priority","title":"12. Image Optimization \u2b50 Medium Priority","text":"

Add image handling plugin:

plugins:\n  - search\n  - mermaid2\n  - codeinclude\n  - minify:\n      minify_html: true\n      minify_js: true\n      minify_css: true\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n  # Optional: Image optimization\n  # - awesome-pages  # For better page organization\n

Benefits: Faster page loads, better mobile experience.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#13-keyboard-shortcuts--low-priority","title":"13. Keyboard Shortcuts \u2b50 Low Priority","text":"

Enable keyboard navigation:

theme:\n  keyboard_shortcuts:\n    search: true\n    previous: true\n    next: true\n

Benefits: Power users can navigate faster.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#14-print-styles--low-priority","title":"14. Print Styles \u2b50 Low Priority","text":"

Better printing experience:

theme:\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.indexes\n    - navigation.instant\n    - navigation.tracking\n    - navigation.smooth\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link\n    - content.tooltips\n    - content.action.edit  # Edit button\n    - content.action.view  # View source\n

Benefits: Users can print documentation cleanly.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#15-better-search-configuration--medium-priority","title":"15. Better Search Configuration \u2b50 Medium Priority","text":"

Enhance search capabilities:

plugins:\n  - search:\n      lang:\n        - en\n      separator: '[\\s\\-,:!=\\[\\]()\"`/]+|\\.(?!\\d)|&[lg]t;|&amp;'\n      prebuild_index: true  # For faster search\n      indexing: full  # Full-text indexing\n

Benefits: Faster, more accurate search results.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#16-api-documentation-enhancements--high-priority-for-your-api-docs","title":"16. API Documentation Enhancements \u2b50 High Priority (for your API docs)","text":"

Since you have extensive API documentation, consider:

markdown_extensions:\n  - pymdownx.superfences:\n      custom_fences:\n        - name: mermaid\n          class: mermaid\n          format: !!python/name:pymdownx.superfences.fence_code_format\n      preserve_tabs: true\n  # Add API-specific features\n  - attr_list\n  - md_in_html\n  - pymdownx.caret\n  - pymdownx.tilde\n

Benefits: Better formatting for API endpoints, parameters, responses.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#17-blognews-section--low-priority-if-needed","title":"17. Blog/News Section \u2b50 Low Priority (if needed)","text":"

If you want to add a blog:

plugins:\n  - blog:\n      blog_dir: blog\n      blog_description: \"News and updates\"\n      post_date_format: full\n      post_url_format: '{slug}'\n      archive: true\n

Benefits: Keep users updated with changelog, announcements.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#18-tags-and-categories--low-priority","title":"18. Tags and Categories \u2b50 Low Priority","text":"

Organize content with tags:

markdown_extensions:\n  - meta\n

Then in frontmatter: 

---\ntags:\n  - api\n  - agents\n  - getting-started\n---\n

Benefits: Better content organization, related content discovery.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#19-better-mobile-experience--high-priority","title":"19. Better Mobile Experience \u2b50 High Priority","text":"

Ensure mobile optimization:

theme:\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.instant  # Helps on mobile\n    - navigation.tracking\n    - navigation.smooth\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link\n    - content.tooltips\n    - toc.integrate  # Better mobile TOC\n

Benefits: Better experience for mobile users (growing segment).

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#20-feedback-mechanism--medium-priority","title":"20. Feedback Mechanism \u2b50 Medium Priority","text":"

Add feedback buttons:

extra:\n  feedback:\n    title: \"Was this page helpful?\"\n    ratings:\n      - icon: material/thumb-up-outline\n        name: \"This page was helpful\"\n      - icon: material/thumb-down-outline\n        name: \"This page could be improved\"\n

Benefits: Understand what content needs improvement.

"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#priority-recommendations","title":"Priority Recommendations","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#immediate-high-impact-easy-implementation","title":"Immediate (High Impact, Easy Implementation)","text":"
  1. \u2705 Git revision dates - Shows content freshness
  2. \u2705 Enhanced navigation features - Better UX
  3. \u2705 Content tabs - Perfect for code examples
  4. \u2705 Better search configuration - Faster search
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#short-term-high-impact-medium-effort","title":"Short-term (High Impact, Medium Effort)","text":"
  1. \u2705 API documentation enhancements - Better API docs
  2. \u2705 Enhanced admonitions - Better visual hierarchy
  3. \u2705 Mobile optimization - Better mobile experience
  4. \u2705 Analytics - Understand user behavior
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#long-term-nice-to-have","title":"Long-term (Nice to Have)","text":"
  1. \u26a0\ufe0f Versioning - If you need multiple versions
  2. \u26a0\ufe0f Math formulas - If you have mathematical content
  3. \u26a0\ufe0f Blog section - If you want to publish updates
  4. \u26a0\ufe0f Custom CSS/JS - For advanced customization
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#implementation-example","title":"Implementation Example","text":"

Here's an enhanced mkdocs.yml with the high-priority improvements:

site_name: The DETERMINATOR\nsite_description: Generalist Deep Research Agent that Stops at Nothing\nsite_author: The DETERMINATOR Team\nsite_url: https://deepcritical.github.io/GradioDemo/\n\nrepo_name: DeepCritical/GradioDemo\nrepo_url: https://github.com/DeepCritical/GradioDemo\nedit_uri: edit/dev/docs/\n\nstrict: false\n\ntheme:\n  name: material\n  palette:\n    - scheme: default\n      primary: orange\n      accent: red\n      toggle:\n        icon: material/brightness-7\n        name: Switch to dark mode\n    - scheme: slate\n      primary: orange\n      accent: red\n      toggle:\n        icon: material/brightness-4\n        name: Switch to light mode\n  features:\n    - navigation.tabs\n    - navigation.sections\n    - navigation.expand\n    - navigation.top\n    - navigation.indexes\n    - navigation.instant\n    - navigation.tracking\n    - navigation.smooth\n    - search.suggest\n    - search.highlight\n    - content.code.annotate\n    - content.code.copy\n    - content.tabs.link\n    - content.tooltips\n    - toc.integrate\n  icon:\n    repo: fontawesome/brands/github\n  language: en\n\nplugins:\n  - search:\n      lang:\n        - en\n      separator: '[\\s\\-,:!=\\[\\]()\"`/]+|\\.(?!\\d)|&[lg]t;|&amp;'\n      prebuild_index: true\n      indexing: full\n  - mermaid2\n  - codeinclude\n  - git-revision-date-localized:\n      enable_creation_date: true\n      type: timeago\n      fallback_to_build_date: true\n  - minify:\n      minify_html: true\n      minify_js: true\n      minify_css: true\n\nmarkdown_extensions:\n  - dev.docs_plugins:\n      base_path: \".\"\n  - pymdownx.highlight:\n      anchor_linenums: true\n      line_spans: __span\n      pygments_lang_class: true\n      use_pygments: true\n      noclasses: false\n  - pymdownx.inlinehilite\n  - pymdownx.superfences:\n      custom_fences:\n        - name: mermaid\n          class: mermaid\n          format: !!python/name:pymdownx.superfences.fence_code_format\n      preserve_tabs: true\n  - pymdownx.tabbed:\n      alternate_style: true\n      combine_header_slug: true\n  - pymdownx.tasklist:\n      custom_checkbox: true\n  - pymdownx.emoji:\n      emoji_generator: !!python/name:pymdownx.emoji.to_svg\n      emoji_index: !!python/name:pymdownx.emoji.twemoji\n  - pymdownx.snippets\n  - admonition\n  - pymdownx.details\n  - attr_list\n  - md_in_html\n  - tables\n  - meta\n  - toc:\n      permalink: true\n      permalink_title: \"Anchor link to this section\"\n      baselevel: 1\n      toc_depth: 3\n      slugify: !!python/object/apply:pymdownx.slugs.slugify\n        kwds:\n          case: lower\n\nnav:\n  - Home: index.md\n  - Overview:\n    - overview/architecture.md\n    - overview/features.md\n  - Getting Started:\n    - getting-started/installation.md\n    - getting-started/quick-start.md\n    - getting-started/mcp-integration.md\n    - getting-started/examples.md\n  - Configuration:\n    - configuration/index.md\n  - Architecture:\n    - \"Graph Orchestration\": architecture/graph_orchestration.md\n    - \"Workflow Diagrams\": architecture/workflow-diagrams.md\n    - \"Agents\": architecture/agents.md\n    - \"Orchestrators\": architecture/orchestrators.md\n    - \"Tools\": architecture/tools.md\n    - \"Middleware\": architecture/middleware.md\n    - \"Services\": architecture/services.md\n  - API Reference:\n    - api/agents.md\n    - api/tools.md\n    - api/orchestrators.md\n    - api/services.md\n    - api/models.md\n  - Contributing:\n    - contributing/index.md\n    - contributing/code-quality.md\n    - contributing/code-style.md\n    - contributing/error-handling.md\n    - contributing/implementation-patterns.md\n    - contributing/prompt-engineering.md\n    - contributing/testing.md\n  - License: LICENSE.md\n  - Team: team.md\n\nextra:\n  social:\n    - icon: fontawesome/brands/github\n      link: https://github.com/DeepCritical/GradioDemo\n      name: GitHub\n    - icon: material/web\n      link: https://huggingface.co/spaces/DataQuests/DeepCritical\n      name: HuggingFace Space\n  version:\n    provider: mike\n  generator:\n    enabled: false\n\ncopyright: Copyright &copy; 2024 DeepCritical Team\n
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#additional-documentation-improvements","title":"Additional Documentation Improvements","text":""},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#content-structure","title":"Content Structure","text":"
  1. Add a changelog page - Keep users informed of updates
  2. Add a FAQ section - Address common questions
  3. Add a glossary - Define technical terms
  4. Add a troubleshooting guide - Help users solve common issues
  5. Add video tutorials - Embed videos for complex topics
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#visual-enhancements","title":"Visual Enhancements","text":"
  1. Add diagrams - Use more Mermaid diagrams for complex flows
  2. Add screenshots - Visual guides for UI features
  3. Add code examples - More practical examples
  4. Add comparison tables - Compare different approaches/options
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#seo--discoverability","title":"SEO & Discoverability","text":"
  1. Add meta descriptions - Better search engine results
  2. Add Open Graph tags - Better social media sharing
  3. Add sitemap - Help search engines index your docs
  4. Add robots.txt - Control search engine crawling
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#next-steps","title":"Next Steps","text":"
  1. Review this assessment
  2. Prioritize features based on your needs
  3. Test changes in a branch
  4. Gather user feedback
  5. Iterate and improve
"},{"location":"MKDOCS_IMPROVEMENTS_ASSESSMENT/#resources","title":"Resources","text":""},{"location":"team/","title":"Team","text":"

DeepCritical is developed by a team of researchers and developers working on AI-assisted research.

"},{"location":"team/#team-members","title":"Team Members","text":""},{"location":"team/#zj","title":"ZJ","text":""},{"location":"team/#mario-aderman","title":"Mario Aderman","text":""},{"location":"team/#joseph-pollack","title":"Joseph Pollack","text":""},{"location":"team/#virat-chauran","title":"Virat Chauran","text":""},{"location":"team/#anna-bossler","title":"Anna Bossler","text":""},{"location":"team/#about","title":"About","text":"

The DeepCritical team met online in the Alzheimer's Critical Literature Review Group in the Hugging Science initiative. We're building the agent framework we want to use for AI-assisted research to turn the vast amounts of clinical data into cures.

"},{"location":"team/#contributing","title":"Contributing","text":"

We welcome contributions! See the Contributing Guide for details.

"},{"location":"team/#links","title":"Links","text":""},{"location":"api/agents/","title":"Agents API Reference","text":"

This page documents the API for DeepCritical agents.

"},{"location":"api/agents/#knowledgegapagent","title":"KnowledgeGapAgent","text":"

Module: src.agents.knowledge_gap

Purpose: Evaluates research state and identifies knowledge gaps.

"},{"location":"api/agents/#methods","title":"Methods","text":""},{"location":"api/agents/#evaluate","title":"evaluate","text":"

Evaluates research completeness and identifies outstanding knowledge gaps.

Parameters: - query: Research query string - background_context: Background context for the query (default: \"\") - conversation_history: History of actions, findings, and thoughts as string (default: \"\") - iteration: Current iteration number (default: 0) - time_elapsed_minutes: Elapsed time in minutes (default: 0.0) - max_time_minutes: Maximum time limit in minutes (default: 10)

Returns: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

"},{"location":"api/agents/#toolselectoragent","title":"ToolSelectorAgent","text":"

Module: src.agents.tool_selector

Purpose: Selects appropriate tools for addressing knowledge gaps.

"},{"location":"api/agents/#methods_1","title":"Methods","text":""},{"location":"api/agents/#select_tools","title":"select_tools","text":"

Selects tools for addressing a knowledge gap.

Parameters: - gap: The knowledge gap to address - query: Research query string - background_context: Optional background context (default: \"\") - conversation_history: History of actions, findings, and thoughts as string (default: \"\")

Returns: AgentSelectionPlan with list of AgentTask objects.

"},{"location":"api/agents/#writeragent","title":"WriterAgent","text":"

Module: src.agents.writer

Purpose: Generates final reports from research findings.

"},{"location":"api/agents/#methods_2","title":"Methods","text":""},{"location":"api/agents/#write_report","title":"write_report","text":"

Generates a markdown report from research findings.

Parameters: - query: Research query string - findings: Research findings to include in report - output_length: Optional description of desired output length (default: \"\") - output_instructions: Optional additional instructions for report generation (default: \"\")

Returns: Markdown string with numbered citations.

"},{"location":"api/agents/#longwriteragent","title":"LongWriterAgent","text":"

Module: src.agents.long_writer

Purpose: Long-form report generation with section-by-section writing.

"},{"location":"api/agents/#methods_3","title":"Methods","text":""},{"location":"api/agents/#write_next_section","title":"write_next_section","text":"

Writes the next section of a long-form report.

Parameters: - original_query: The original research query - report_draft: Current report draft as string (all sections written so far) - next_section_title: Title of the section to write - next_section_draft: Draft content for the next section

Returns: LongWriterOutput with formatted section and references.

"},{"location":"api/agents/#write_report_1","title":"write_report","text":"

Generates final report from draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Complete report draft

Returns: Final markdown report string.

"},{"location":"api/agents/#proofreaderagent","title":"ProofreaderAgent","text":"

Module: src.agents.proofreader

Purpose: Proofreads and polishes report drafts.

"},{"location":"api/agents/#methods_4","title":"Methods","text":""},{"location":"api/agents/#proofread","title":"proofread","text":"

Proofreads and polishes a report draft.

Parameters: - query: Research query string - report_title: Title of the report - report_draft: Report draft to proofread

Returns: Polished markdown string.

"},{"location":"api/agents/#thinkingagent","title":"ThinkingAgent","text":"

Module: src.agents.thinking

Purpose: Generates observations from conversation history.

"},{"location":"api/agents/#methods_5","title":"Methods","text":""},{"location":"api/agents/#generate_observations","title":"generate_observations","text":"

Generates observations from conversation history.

Parameters: - query: Research query string - background_context: Optional background context (default: \"\") - conversation_history: History of actions, findings, and thoughts as string (default: \"\") - iteration: Current iteration number (default: 1)

Returns: Observation string.

"},{"location":"api/agents/#inputparseragent","title":"InputParserAgent","text":"

Module: src.agents.input_parser

Purpose: Parses and improves user queries, detects research mode.

"},{"location":"api/agents/#methods_6","title":"Methods","text":""},{"location":"api/agents/#parse","title":"parse","text":"

Parses and improves a user query.

Parameters: - query: Original query string

Returns: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: \"iterative\" or \"deep\" - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"api/agents/#factory-functions","title":"Factory Functions","text":"

All agents have factory functions in src.agent_factory.agents:

Parameters: - model: Optional Pydantic AI model. If None, uses get_model() from settings. - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars)

Returns: Agent instance.

"},{"location":"api/agents/#see-also","title":"See Also","text":""},{"location":"api/models/","title":"Models API Reference","text":"

This page documents the Pydantic models used throughout DeepCritical.

"},{"location":"api/models/#evidence","title":"Evidence","text":"

Module: src.utils.models

Purpose: Represents evidence from search results.

Fields: - citation: Citation information (title, URL, date, authors) - content: Evidence text content - relevance: Relevance score (0.0-1.0) - metadata: Additional metadata dictionary

"},{"location":"api/models/#citation","title":"Citation","text":"

Module: src.utils.models

Purpose: Citation information for evidence.

Fields: - source: Source name (e.g., \"pubmed\", \"clinicaltrials\", \"europepmc\", \"web\", \"rag\") - title: Article/trial title - url: Source URL - date: Publication date (YYYY-MM-DD or \"Unknown\") - authors: List of authors (optional)

"},{"location":"api/models/#knowledgegapoutput","title":"KnowledgeGapOutput","text":"

Module: src.utils.models

Purpose: Output from knowledge gap evaluation.

Fields: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

"},{"location":"api/models/#agentselectionplan","title":"AgentSelectionPlan","text":"

Module: src.utils.models

Purpose: Plan for tool/agent selection.

Fields: - tasks: List of agent tasks to execute

"},{"location":"api/models/#agenttask","title":"AgentTask","text":"

Module: src.utils.models

Purpose: Individual agent task.

Fields: - gap: The knowledge gap being addressed (optional) - agent: Name of agent to use - query: The specific query for the agent - entity_website: The website of the entity being researched, if known (optional)

"},{"location":"api/models/#reportdraft","title":"ReportDraft","text":"

Module: src.utils.models

Purpose: Draft structure for long-form reports.

Fields: - sections: List of report sections

"},{"location":"api/models/#reportsection","title":"ReportSection","text":"

Module: src.utils.models

Purpose: Individual section in a report draft.

Fields: - section_title: The title of the section - section_content: The content of the section

"},{"location":"api/models/#parsedquery","title":"ParsedQuery","text":"

Module: src.utils.models

Purpose: Parsed and improved query.

Fields: - original_query: Original query string - improved_query: Refined query string - research_mode: Research mode (\"iterative\" or \"deep\") - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"api/models/#conversation","title":"Conversation","text":"

Module: src.utils.models

Purpose: Conversation history with iterations.

Fields: - history: List of iteration data

"},{"location":"api/models/#iterationdata","title":"IterationData","text":"

Module: src.utils.models

Purpose: Data for a single iteration.

Fields: - gap: The gap addressed in the iteration - tool_calls: The tool calls made - findings: The findings collected from tool calls - thought: The thinking done to reflect on the success of the iteration and next steps

"},{"location":"api/models/#agentevent","title":"AgentEvent","text":"

Module: src.utils.models

Purpose: Event emitted during research execution.

Fields: - type: Event type (e.g., \"started\", \"search_complete\", \"complete\") - iteration: Iteration number (optional) - data: Event data dictionary

"},{"location":"api/models/#budgetstatus","title":"BudgetStatus","text":"

Module: src.utils.models

Purpose: Current budget status.

Fields: - tokens_used: Total tokens used - tokens_limit: Token budget limit - time_elapsed_seconds: Time elapsed in seconds - time_limit_seconds: Time budget limit (default: 600.0 seconds / 10 minutes) - iterations: Number of iterations completed - iterations_limit: Maximum iterations (default: 10) - iteration_tokens: Tokens used per iteration (iteration number -> token count)

"},{"location":"api/models/#see-also","title":"See Also","text":""},{"location":"api/orchestrators/","title":"Orchestrators API Reference","text":"

This page documents the API for DeepCritical orchestrators.

"},{"location":"api/orchestrators/#iterativeresearchflow","title":"IterativeResearchFlow","text":"

Module: src.orchestrator.research_flow

Purpose: Single-loop research with search-judge-synthesize cycles.

"},{"location":"api/orchestrators/#methods","title":"Methods","text":""},{"location":"api/orchestrators/#run","title":"run","text":"

Runs iterative research flow.

Parameters: - query: Research query string - background_context: Background context (default: \"\") - output_length: Optional description of desired output length (default: \"\") - output_instructions: Optional additional instructions for report generation (default: \"\")

Returns: Final report string.

Note: max_iterations, max_time_minutes, and token_budget are constructor parameters, not run() parameters.

"},{"location":"api/orchestrators/#deepresearchflow","title":"DeepResearchFlow","text":"

Module: src.orchestrator.research_flow

Purpose: Multi-section parallel research with planning and synthesis.

"},{"location":"api/orchestrators/#methods_1","title":"Methods","text":""},{"location":"api/orchestrators/#run_1","title":"run","text":"

Runs deep research flow.

Parameters: - query: Research query string

Returns: Final report string.

Note: max_iterations_per_section, max_time_minutes, and token_budget are constructor parameters, not run() parameters.

"},{"location":"api/orchestrators/#graphorchestrator","title":"GraphOrchestrator","text":"

Module: src.orchestrator.graph_orchestrator

Purpose: Graph-based execution using Pydantic AI agents as nodes.

"},{"location":"api/orchestrators/#methods_2","title":"Methods","text":""},{"location":"api/orchestrators/#run_2","title":"run","text":"

Runs graph-based research orchestration.

Parameters: - query: Research query string

Yields: AgentEvent objects during graph execution.

Note: research_mode and use_graph are constructor parameters, not run() parameters.

"},{"location":"api/orchestrators/#orchestrator-factory","title":"Orchestrator Factory","text":"

Module: src.orchestrator_factory

Purpose: Factory for creating orchestrators.

"},{"location":"api/orchestrators/#functions","title":"Functions","text":""},{"location":"api/orchestrators/#create_orchestrator","title":"create_orchestrator","text":"

Creates an orchestrator instance.

Parameters: - search_handler: Search handler protocol implementation (optional, required for simple mode) - judge_handler: Judge handler protocol implementation (optional, required for simple mode) - config: Configuration object (optional) - mode: Orchestrator mode (\"simple\", \"advanced\", \"magentic\", \"iterative\", \"deep\", \"auto\", or None for auto-detect) - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars)

Returns: Orchestrator instance.

Raises: - ValueError: If requirements not met

Modes: - \"simple\": Legacy orchestrator - \"advanced\" or \"magentic\": Magentic orchestrator (requires OpenAI API key) - None: Auto-detect based on API key availability

"},{"location":"api/orchestrators/#magenticorchestrator","title":"MagenticOrchestrator","text":"

Module: src.orchestrator_magentic

Purpose: Multi-agent coordination using Microsoft Agent Framework.

"},{"location":"api/orchestrators/#methods_3","title":"Methods","text":""},{"location":"api/orchestrators/#run_3","title":"run","text":"

Runs Magentic orchestration.

Parameters: - query: Research query string

Yields: AgentEvent objects converted from Magentic events.

Note: max_rounds and max_stalls are constructor parameters, not run() parameters.

Requirements: - agent-framework-core package - OpenAI API key

"},{"location":"api/orchestrators/#see-also","title":"See Also","text":""},{"location":"api/services/","title":"Services API Reference","text":"

This page documents the API for DeepCritical services.

"},{"location":"api/services/#embeddingservice","title":"EmbeddingService","text":"

Module: src.services.embeddings

Purpose: Local sentence-transformers for semantic search and deduplication.

"},{"location":"api/services/#methods","title":"Methods","text":""},{"location":"api/services/#embed","title":"embed","text":"

Generates embedding for a text string.

Parameters: - text: Text to embed

Returns: Embedding vector as list of floats.

"},{"location":"api/services/#embed_batch","title":"embed_batch","text":"
async def embed_batch(self, texts: list[str]) -> list[list[float]]\n

Generates embeddings for multiple texts.

Parameters: - texts: List of texts to embed

Returns: List of embedding vectors.

"},{"location":"api/services/#similarity","title":"similarity","text":"
async def similarity(self, text1: str, text2: str) -> float\n

Calculates similarity between two texts.

Parameters: - text1: First text - text2: Second text

Returns: Similarity score (0.0-1.0).

"},{"location":"api/services/#find_duplicates","title":"find_duplicates","text":"
async def find_duplicates(\n    self,\n    texts: list[str],\n    threshold: float = 0.85\n) -> list[tuple[int, int]]\n

Finds duplicate texts based on similarity threshold.

Parameters: - texts: List of texts to check - threshold: Similarity threshold (default: 0.85)

Returns: List of (index1, index2) tuples for duplicate pairs.

"},{"location":"api/services/#add_evidence","title":"add_evidence","text":"
async def add_evidence(\n    self,\n    evidence_id: str,\n    content: str,\n    metadata: dict[str, Any]\n) -> None\n

Adds evidence to vector store for semantic search.

Parameters: - evidence_id: Unique identifier for the evidence - content: Evidence text content - metadata: Additional metadata dictionary

"},{"location":"api/services/#search_similar","title":"search_similar","text":"
async def search_similar(\n    self,\n    query: str,\n    n_results: int = 5\n) -> list[dict[str, Any]]\n

Finds semantically similar evidence.

Parameters: - query: Search query string - n_results: Number of results to return (default: 5)

Returns: List of dictionaries with id, content, metadata, and distance keys.

"},{"location":"api/services/#deduplicate","title":"deduplicate","text":"
async def deduplicate(\n    self,\n    new_evidence: list[Evidence],\n    threshold: float = 0.9\n) -> list[Evidence]\n

Removes semantically duplicate evidence.

Parameters: - new_evidence: List of evidence items to deduplicate - threshold: Similarity threshold (default: 0.9, where 0.9 = 90% similar is duplicate)

Returns: List of unique evidence items (not already in vector store).

"},{"location":"api/services/#factory-function","title":"Factory Function","text":""},{"location":"api/services/#get_embedding_service","title":"get_embedding_service","text":"
@lru_cache(maxsize=1)\ndef get_embedding_service() -> EmbeddingService\n

Returns singleton EmbeddingService instance.

"},{"location":"api/services/#llamaindexragservice","title":"LlamaIndexRAGService","text":"

Module: src.services.rag

Purpose: Retrieval-Augmented Generation using LlamaIndex.

"},{"location":"api/services/#methods_1","title":"Methods","text":""},{"location":"api/services/#ingest_evidence","title":"ingest_evidence","text":"

Ingests evidence into RAG service.

Parameters: - evidence_list: List of Evidence objects to ingest

Note: Supports multiple embedding providers (OpenAI, local sentence-transformers, Hugging Face).

"},{"location":"api/services/#retrieve","title":"retrieve","text":"
def retrieve(\n    self,\n    query: str,\n    top_k: int | None = None\n) -> list[dict[str, Any]]\n

Retrieves relevant documents for a query.

Parameters: - query: Search query string - top_k: Number of top results to return (defaults to similarity_top_k from constructor)

Returns: List of dictionaries with text, score, and metadata keys.

"},{"location":"api/services/#query","title":"query","text":"
def query(\n    self,\n    query_str: str,\n    top_k: int | None = None\n) -> str\n

Queries RAG service and returns synthesized response.

Parameters: - query_str: Query string - top_k: Number of results to use (defaults to similarity_top_k from constructor)

Returns: Synthesized response string.

Raises: - ConfigurationError: If no LLM API key is available for query synthesis

"},{"location":"api/services/#ingest_documents","title":"ingest_documents","text":"
def ingest_documents(self, documents: list[Any]) -> None\n

Ingests raw LlamaIndex Documents.

Parameters: - documents: List of LlamaIndex Document objects

"},{"location":"api/services/#clear_collection","title":"clear_collection","text":"
def clear_collection(self) -> None\n

Clears all documents from the collection.

"},{"location":"api/services/#factory-function_1","title":"Factory Function","text":""},{"location":"api/services/#get_rag_service","title":"get_rag_service","text":"
def get_rag_service(\n    collection_name: str = \"deepcritical_evidence\",\n    oauth_token: str | None = None,\n    **kwargs: Any\n) -> LlamaIndexRAGService\n

Get or create a RAG service instance.

Parameters: - collection_name: Name of the ChromaDB collection (default: \"deepcritical_evidence\") - oauth_token: Optional OAuth token from HuggingFace login (takes priority over env vars) - **kwargs: Additional arguments for LlamaIndexRAGService (e.g., use_openai_embeddings=False)

Returns: Configured LlamaIndexRAGService instance.

Note: By default, uses local embeddings (sentence-transformers) which require no API keys.

"},{"location":"api/services/#statisticalanalyzer","title":"StatisticalAnalyzer","text":"

Module: src.services.statistical_analyzer

Purpose: Secure execution of AI-generated statistical code.

"},{"location":"api/services/#methods_2","title":"Methods","text":""},{"location":"api/services/#analyze","title":"analyze","text":"
async def analyze(\n    self,\n    query: str,\n    evidence: list[Evidence],\n    hypothesis: dict[str, Any] | None = None\n) -> AnalysisResult\n

Analyzes a research question using statistical methods.

Parameters: - query: The research question - evidence: List of Evidence objects to analyze - hypothesis: Optional hypothesis dict with drug, target, pathway, effect, confidence keys

Returns: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - confidence: Confidence in verdict (0.0-1.0) - statistical_evidence: Summary of statistical findings - code_generated: Python code that was executed - execution_output: Output from code execution - key_takeaways: Key takeaways from analysis - limitations: List of limitations

Note: Requires Modal credentials for sandbox execution.

"},{"location":"api/services/#see-also","title":"See Also","text":""},{"location":"api/tools/","title":"Tools API Reference","text":"

This page documents the API for DeepCritical search tools.

"},{"location":"api/tools/#searchtool-protocol","title":"SearchTool Protocol","text":"

All tools implement the SearchTool protocol:

class SearchTool(Protocol):\n    @property\n    def name(self) -> str: ...\n    \n    async def search(\n        self, \n        query: str, \n        max_results: int = 10\n    ) -> list[Evidence]: ...\n
"},{"location":"api/tools/#pubmedtool","title":"PubMedTool","text":"

Module: src.tools.pubmed

Purpose: Search peer-reviewed biomedical literature from PubMed.

"},{"location":"api/tools/#properties","title":"Properties","text":""},{"location":"api/tools/#name","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"pubmed\"

"},{"location":"api/tools/#methods","title":"Methods","text":""},{"location":"api/tools/#search","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches PubMed for articles.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with PubMed articles.

Raises: - SearchError: If search fails (timeout, HTTP error, XML parsing error) - RateLimitError: If rate limit is exceeded (429 status code)

Note: Uses NCBI E-utilities (ESearch \u2192 EFetch). Rate limit: 0.34s between requests. Handles single vs. multiple articles.

"},{"location":"api/tools/#clinicaltrialstool","title":"ClinicalTrialsTool","text":"

Module: src.tools.clinicaltrials

Purpose: Search ClinicalTrials.gov for interventional studies.

"},{"location":"api/tools/#properties_1","title":"Properties","text":""},{"location":"api/tools/#name_1","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"clinicaltrials\"

"},{"location":"api/tools/#methods_1","title":"Methods","text":""},{"location":"api/tools/#search_1","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches ClinicalTrials.gov for trials.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with clinical trials.

Note: Only returns interventional studies with status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION. Uses requests library (NOT httpx - WAF blocks httpx). Runs in thread pool for async compatibility.

Raises: - SearchError: If search fails (HTTP error, request exception)

"},{"location":"api/tools/#europepmctool","title":"EuropePMCTool","text":"

Module: src.tools.europepmc

Purpose: Search Europe PMC for preprints and peer-reviewed articles.

"},{"location":"api/tools/#properties_2","title":"Properties","text":""},{"location":"api/tools/#name_2","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"europepmc\"

"},{"location":"api/tools/#methods_2","title":"Methods","text":""},{"location":"api/tools/#search_2","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches Europe PMC for articles and preprints.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects with articles/preprints.

Note: Includes both preprints (marked with [PREPRINT - Not peer-reviewed]) and peer-reviewed articles. Handles preprint markers. Builds URLs from DOI or PMID.

Raises: - SearchError: If search fails (HTTP error, connection error)

"},{"location":"api/tools/#ragtool","title":"RAGTool","text":"

Module: src.tools.rag_tool

Purpose: Semantic search within collected evidence.

"},{"location":"api/tools/#initialization","title":"Initialization","text":"
def __init__(\n    self,\n    rag_service: LlamaIndexRAGService | None = None,\n    oauth_token: str | None = None\n) -> None\n

Parameters: - rag_service: Optional RAG service instance. If None, will be lazy-initialized. - oauth_token: Optional OAuth token from HuggingFace login (for RAG LLM)

"},{"location":"api/tools/#properties_3","title":"Properties","text":""},{"location":"api/tools/#name_3","title":"name","text":"
@property\ndef name(self) -> str\n

Returns tool name: \"rag\"

"},{"location":"api/tools/#methods_3","title":"Methods","text":""},{"location":"api/tools/#search_3","title":"search","text":"
async def search(\n    self,\n    query: str,\n    max_results: int = 10\n) -> list[Evidence]\n

Searches collected evidence using semantic similarity.

Parameters: - query: Search query string - max_results: Maximum number of results to return (default: 10)

Returns: List of Evidence objects from collected evidence.

Raises: - ConfigurationError: If RAG service is unavailable

Note: Requires evidence to be ingested into RAG service first. Wraps LlamaIndexRAGService. Returns Evidence from RAG results.

"},{"location":"api/tools/#searchhandler","title":"SearchHandler","text":"

Module: src.tools.search_handler

Purpose: Orchestrates parallel searches across multiple tools.

"},{"location":"api/tools/#initialization_1","title":"Initialization","text":"
def __init__(\n    self,\n    tools: list[SearchTool],\n    timeout: float = 30.0,\n    include_rag: bool = False,\n    auto_ingest_to_rag: bool = True,\n    oauth_token: str | None = None\n) -> None\n

Parameters: - tools: List of search tools to use - timeout: Timeout for each search in seconds (default: 30.0) - include_rag: Whether to include RAG tool in searches (default: False) - auto_ingest_to_rag: Whether to automatically ingest results into RAG (default: True) - oauth_token: Optional OAuth token from HuggingFace login (for RAG LLM)

"},{"location":"api/tools/#methods_4","title":"Methods","text":""},{"location":"api/tools/#execute","title":"execute","text":"

Searches multiple tools in parallel.

Parameters: - query: Search query string - max_results_per_tool: Maximum results per tool (default: 10)

Returns: SearchResult with: - query: The search query - evidence: Aggregated list of evidence - sources_searched: List of source names searched - total_found: Total number of results - errors: List of error messages from failed tools

Raises: - SearchError: If search times out

Note: Uses asyncio.gather() for parallel execution. Handles tool failures gracefully (returns errors in SearchResult.errors). Automatically ingests evidence into RAG if enabled.

"},{"location":"api/tools/#see-also","title":"See Also","text":""},{"location":"architecture/agents/","title":"Agents Architecture","text":"

DeepCritical uses Pydantic AI agents for all AI-powered operations. All agents follow a consistent pattern and use structured output types.

"},{"location":"architecture/agents/#agent-pattern","title":"Agent Pattern","text":""},{"location":"architecture/agents/#pydantic-ai-agents","title":"Pydantic AI Agents","text":"

Pydantic AI agents use the Agent class with the following structure:

Note: Factory functions accept an optional oauth_token parameter for HuggingFace authentication, which takes priority over environment variables.

"},{"location":"architecture/agents/#model-initialization","title":"Model Initialization","text":"

Agents use get_model() from src/agent_factory/judges.py if no model is provided. This supports:

The model selection is based on the configured LLM_PROVIDER in settings.

"},{"location":"architecture/agents/#error-handling","title":"Error Handling","text":"

Agents return fallback values on failure rather than raising exceptions:

All errors are logged with context using structlog.

"},{"location":"architecture/agents/#input-validation","title":"Input Validation","text":"

All agents validate inputs:

"},{"location":"architecture/agents/#output-types","title":"Output Types","text":"

Agents use structured output types from src/utils/models.py:

For text output (writer agents), agents return str directly.

"},{"location":"architecture/agents/#agent-types","title":"Agent Types","text":""},{"location":"architecture/agents/#knowledge-gap-agent","title":"Knowledge Gap Agent","text":"

File: src/agents/knowledge_gap.py

Purpose: Evaluates research state and identifies knowledge gaps.

Output: KnowledgeGapOutput with: - research_complete: Boolean indicating if research is complete - outstanding_gaps: List of remaining knowledge gaps

Methods: - async def evaluate(query, background_context, conversation_history, iteration, time_elapsed_minutes, max_time_minutes) -> KnowledgeGapOutput

"},{"location":"architecture/agents/#tool-selector-agent","title":"Tool Selector Agent","text":"

File: src/agents/tool_selector.py

Purpose: Selects appropriate tools for addressing knowledge gaps.

Output: AgentSelectionPlan with list of AgentTask objects.

Available Agents: - WebSearchAgent: General web search for fresh information - SiteCrawlerAgent: Research specific entities/companies - RAGAgent: Semantic search within collected evidence

"},{"location":"architecture/agents/#writer-agent","title":"Writer Agent","text":"

File: src/agents/writer.py

Purpose: Generates final reports from research findings.

Output: Markdown string with numbered citations.

Methods: - async def write_report(query, findings, output_length, output_instructions) -> str

Features: - Validates inputs - Truncates very long findings (max 50000 chars) with warning - Retry logic for transient failures (3 retries) - Citation validation before returning

"},{"location":"architecture/agents/#long-writer-agent","title":"Long Writer Agent","text":"

File: src/agents/long_writer.py

Purpose: Long-form report generation with section-by-section writing.

Input/Output: Uses ReportDraft models.

Methods: - async def write_next_section(query, draft, section_title, section_content) -> LongWriterOutput - async def write_report(query, report_title, report_draft) -> str

Features: - Writes sections iteratively - Aggregates references across sections - Reformats section headings and references - Deduplicates and renumbers references

"},{"location":"architecture/agents/#proofreader-agent","title":"Proofreader Agent","text":"

File: src/agents/proofreader.py

Purpose: Proofreads and polishes report drafts.

Input: ReportDraft Output: Polished markdown string

Methods: - async def proofread(query, report_title, report_draft) -> str

Features: - Removes duplicate content across sections - Adds executive summary if multiple sections - Preserves all references and citations - Improves flow and readability

"},{"location":"architecture/agents/#thinking-agent","title":"Thinking Agent","text":"

File: src/agents/thinking.py

Purpose: Generates observations from conversation history.

Output: Observation string

Methods: - async def generate_observations(query, background_context, conversation_history) -> str

"},{"location":"architecture/agents/#input-parser-agent","title":"Input Parser Agent","text":"

File: src/agents/input_parser.py

Purpose: Parses and improves user queries, detects research mode.

Output: ParsedQuery with: - original_query: Original query string - improved_query: Refined query string - research_mode: \"iterative\" or \"deep\" - key_entities: List of key entities - research_questions: List of research questions

"},{"location":"architecture/agents/#magentic-agents","title":"Magentic Agents","text":"

The following agents use the BaseAgent pattern from agent-framework and are used exclusively with MagenticOrchestrator:

"},{"location":"architecture/agents/#hypothesis-agent","title":"Hypothesis Agent","text":"

File: src/agents/hypothesis_agent.py

Purpose: Generates mechanistic hypotheses based on evidence.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Uses internal Pydantic AI Agent with HypothesisAssessment output type - Accesses shared evidence_store for evidence - Uses embedding service for diverse evidence selection (MMR algorithm) - Stores hypotheses in shared context

"},{"location":"architecture/agents/#search-agent","title":"Search Agent","text":"

File: src/agents/search_agent.py

Purpose: Wraps SearchHandler as an agent for Magentic orchestrator.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Executes searches via SearchHandlerProtocol - Deduplicates evidence using embedding service - Searches for semantically related evidence - Updates shared evidence store

"},{"location":"architecture/agents/#analysis-agent","title":"Analysis Agent","text":"

File: src/agents/analysis_agent.py

Purpose: Performs statistical analysis using Modal sandbox.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Wraps StatisticalAnalyzer service - Analyzes evidence and hypotheses - Returns verdict (SUPPORTED/REFUTED/INCONCLUSIVE) - Stores analysis results in shared context

"},{"location":"architecture/agents/#report-agent-magentic","title":"Report Agent (Magentic)","text":"

File: src/agents/report_agent.py

Purpose: Generates structured scientific reports from evidence and hypotheses.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse

Features: - Uses internal Pydantic AI Agent with ResearchReport output type - Accesses shared evidence store and hypotheses - Validates citations before returning - Formats report as markdown

"},{"location":"architecture/agents/#judge-agent","title":"Judge Agent","text":"

File: src/agents/judge_agent.py

Purpose: Evaluates evidence quality and determines if sufficient for synthesis.

Pattern: BaseAgent from agent-framework

Methods: - async def run(messages, thread, **kwargs) -> AgentRunResponse - async def run_stream(messages, thread, **kwargs) -> AsyncIterable[AgentRunResponseUpdate]

Features: - Wraps JudgeHandlerProtocol - Accesses shared evidence store - Returns JudgeAssessment with sufficient flag, confidence, and recommendation

"},{"location":"architecture/agents/#agent-patterns","title":"Agent Patterns","text":"

DeepCritical uses two distinct agent patterns:

"},{"location":"architecture/agents/#1-pydantic-ai-agents-traditional-pattern","title":"1. Pydantic AI Agents (Traditional Pattern)","text":"

These agents use the Pydantic AI Agent class directly and are used in iterative and deep research flows:

"},{"location":"architecture/agents/#2-magentic-agents-agent-framework-pattern","title":"2. Magentic Agents (Agent-Framework Pattern)","text":"

These agents use the BaseAgent class from agent-framework and are used in Magentic orchestrator:

Note: Magentic agents are used exclusively with the MagenticOrchestrator and follow the agent-framework protocol for multi-agent coordination.

"},{"location":"architecture/agents/#factory-functions","title":"Factory Functions","text":"

All agents have factory functions in src/agent_factory/agents.py:

Factory functions: - Use get_model() if no model provided - Accept oauth_token parameter for HuggingFace authentication - Raise ConfigurationError if creation fails - Log agent creation

"},{"location":"architecture/agents/#see-also","title":"See Also","text":""},{"location":"architecture/graph_orchestration/","title":"Graph Orchestration Architecture","text":""},{"location":"architecture/graph_orchestration/#overview","title":"Overview","text":"

DeepCritical implements a graph-based orchestration system for research workflows using Pydantic AI agents as nodes. This enables better parallel execution, conditional routing, and state management compared to simple agent chains.

"},{"location":"architecture/graph_orchestration/#graph-patterns","title":"Graph Patterns","text":""},{"location":"architecture/graph_orchestration/#iterative-research-graph","title":"Iterative Research Graph","text":"

The iterative research graph follows this pattern:

[Input] \u2192 [Thinking] \u2192 [Knowledge Gap] \u2192 [Decision: Complete?]\n                                              \u2193 No          \u2193 Yes\n                                    [Tool Selector]    [Writer]\n                                              \u2193\n                                    [Execute Tools] \u2192 [Loop Back]\n

Node IDs: thinking \u2192 knowledge_gap \u2192 continue_decision \u2192 tool_selector/writer \u2192 execute_tools \u2192 (loop back to thinking)

Special Node Handling: - execute_tools: State node that uses search_handler to execute searches and add evidence to workflow state - continue_decision: Decision node that routes based on research_complete flag from KnowledgeGapOutput

"},{"location":"architecture/graph_orchestration/#deep-research-graph","title":"Deep Research Graph","text":"

The deep research graph follows this pattern:

[Input] \u2192 [Planner] \u2192 [Store Plan] \u2192 [Parallel Loops] \u2192 [Collect Drafts] \u2192 [Synthesizer]\n                                        \u2193         \u2193         \u2193\n                                     [Loop1]  [Loop2]  [Loop3]\n

Node IDs: planner \u2192 store_plan \u2192 parallel_loops \u2192 collect_drafts \u2192 synthesizer

Special Node Handling: - planner: Agent node that creates ReportPlan with report outline - store_plan: State node that stores ReportPlan in context for parallel loops - parallel_loops: Parallel node that executes IterativeResearchFlow instances for each section - collect_drafts: State node that collects section drafts from parallel loops - synthesizer: Agent node that calls LongWriterAgent.write_report() directly with ReportDraft

"},{"location":"architecture/graph_orchestration/#deep-research","title":"Deep Research","text":"
\nsequenceDiagram\n    actor User\n    participant GraphOrchestrator\n    participant InputParser\n    participant GraphBuilder\n    participant GraphExecutor\n    participant Agent\n    participant BudgetTracker\n    participant WorkflowState\n\n    User->>GraphOrchestrator: run(query)\n    GraphOrchestrator->>InputParser: detect_research_mode(query)\n    InputParser-->>GraphOrchestrator: mode (iterative/deep)\n    GraphOrchestrator->>GraphBuilder: build_graph(mode)\n    GraphBuilder-->>GraphOrchestrator: ResearchGraph\n    GraphOrchestrator->>WorkflowState: init_workflow_state()\n    GraphOrchestrator->>BudgetTracker: create_budget()\n    GraphOrchestrator->>GraphExecutor: _execute_graph(graph)\n    \n    loop For each node in graph\n        GraphExecutor->>Agent: execute_node(agent_node)\n        Agent->>Agent: process_input\n        Agent-->>GraphExecutor: result\n        GraphExecutor->>WorkflowState: update_state(result)\n        GraphExecutor->>BudgetTracker: add_tokens(used)\n        GraphExecutor->>BudgetTracker: check_budget()\n        alt Budget exceeded\n            GraphExecutor->>GraphOrchestrator: emit(error_event)\n        else Continue\n            GraphExecutor->>GraphOrchestrator: emit(progress_event)\n        end\n    end\n    \n    GraphOrchestrator->>User: AsyncGenerator[AgentEvent]\n
"},{"location":"architecture/graph_orchestration/#iterative-research","title":"Iterative Research","text":"
sequenceDiagram\n    participant IterativeFlow\n    participant ThinkingAgent\n    participant KnowledgeGapAgent\n    participant ToolSelector\n    participant ToolExecutor\n    participant JudgeHandler\n    participant WriterAgent\n\n    IterativeFlow->>IterativeFlow: run(query)\n    \n    loop Until complete or max_iterations\n        IterativeFlow->>ThinkingAgent: generate_observations()\n        ThinkingAgent-->>IterativeFlow: observations\n        \n        IterativeFlow->>KnowledgeGapAgent: evaluate_gaps()\n        KnowledgeGapAgent-->>IterativeFlow: KnowledgeGapOutput\n        \n        alt Research complete\n            IterativeFlow->>WriterAgent: create_final_report()\n            WriterAgent-->>IterativeFlow: final_report\n        else Gaps remain\n            IterativeFlow->>ToolSelector: select_agents(gap)\n            ToolSelector-->>IterativeFlow: AgentSelectionPlan\n            \n            IterativeFlow->>ToolExecutor: execute_tool_tasks()\n            ToolExecutor-->>IterativeFlow: ToolAgentOutput[]\n            \n            IterativeFlow->>JudgeHandler: assess_evidence()\n            JudgeHandler-->>IterativeFlow: should_continue\n        end\n    end
"},{"location":"architecture/graph_orchestration/#graph-structure","title":"Graph Structure","text":""},{"location":"architecture/graph_orchestration/#nodes","title":"Nodes","text":"

Graph nodes represent different stages in the research workflow:

  1. Agent Nodes: Execute Pydantic AI agents
  2. Input: Prompt/query
  3. Output: Structured or unstructured response

  4. Examples: KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent

  5. State Nodes: Update or read workflow state

  6. Input: Current state
  7. Output: Updated state

  8. Examples: Update evidence, update conversation history

  9. Decision Nodes: Make routing decisions based on conditions

  10. Input: Current state/results
  11. Output: Next node ID

  12. Examples: Continue research vs. complete research

  13. Parallel Nodes: Execute multiple nodes concurrently

  14. Input: List of node IDs
  15. Output: Aggregated results
  16. Examples: Parallel iterative research loops
"},{"location":"architecture/graph_orchestration/#edges","title":"Edges","text":"

Edges define transitions between nodes:

  1. Sequential Edges: Always traversed (no condition)
  2. From: Source node
  3. To: Target node

  4. Condition: None (always True)

  5. Conditional Edges: Traversed based on condition

  6. From: Source node
  7. To: Target node
  8. Condition: Callable that returns bool

  9. Example: If research complete \u2192 go to writer, else \u2192 continue loop

  10. Parallel Edges: Used for parallel execution branches

  11. From: Parallel node
  12. To: Multiple target nodes
  13. Execution: All targets run concurrently
"},{"location":"architecture/graph_orchestration/#state-management","title":"State Management","text":"

State is managed via WorkflowState using ContextVar for thread-safe isolation:

State transitions occur at state nodes, which update the global workflow state.

"},{"location":"architecture/graph_orchestration/#execution-flow","title":"Execution Flow","text":"
  1. Graph Construction: Build graph from nodes and edges using create_iterative_graph() or create_deep_graph()
  2. Graph Validation: Ensure graph is valid (no cycles, all nodes reachable) via ResearchGraph.validate_structure()
  3. Graph Execution: Traverse graph from entry node using GraphOrchestrator._execute_graph()
  4. Node Execution: Execute each node based on type:
  5. Agent Nodes: Call agent.run() with transformed input
  6. State Nodes: Update workflow state via state_updater function
  7. Decision Nodes: Evaluate decision_function to get next node ID
  8. Parallel Nodes: Execute all parallel nodes concurrently via asyncio.gather()
  9. Edge Evaluation: Determine next node(s) based on edges and conditions
  10. Parallel Execution: Use asyncio.gather() for parallel nodes
  11. State Updates: Update state at state nodes via GraphExecutionContext.update_state()
  12. Event Streaming: Yield AgentEvent objects during execution for UI
"},{"location":"architecture/graph_orchestration/#graphexecutioncontext","title":"GraphExecutionContext","text":"

The GraphExecutionContext class manages execution state during graph traversal:

Methods: - set_node_result(node_id, result): Store result from node execution - get_node_result(node_id): Retrieve stored result - has_visited(node_id): Check if node was visited - mark_visited(node_id): Mark node as visited - update_state(updater, data): Update workflow state

"},{"location":"architecture/graph_orchestration/#conditional-routing","title":"Conditional Routing","text":"

Decision nodes evaluate conditions and return next node IDs:

"},{"location":"architecture/graph_orchestration/#parallel-execution","title":"Parallel Execution","text":"

Parallel nodes execute multiple nodes concurrently:

"},{"location":"architecture/graph_orchestration/#budget-enforcement","title":"Budget Enforcement","text":"

Budget constraints are enforced at decision nodes:

If any budget is exceeded, execution routes to exit node.

"},{"location":"architecture/graph_orchestration/#error-handling","title":"Error Handling","text":"

Errors are handled at multiple levels:

  1. Node Level: Catch errors in individual node execution
  2. Graph Level: Handle errors during graph traversal
  3. State Level: Rollback state changes on error

Errors are logged and yield error events for UI.

"},{"location":"architecture/graph_orchestration/#backward-compatibility","title":"Backward Compatibility","text":"

Graph execution is optional via feature flag:

This allows gradual migration and fallback if needed.

"},{"location":"architecture/graph_orchestration/#see-also","title":"See Also","text":""},{"location":"architecture/middleware/","title":"Middleware Architecture","text":"

DeepCritical uses middleware for state management, budget tracking, and workflow coordination.

"},{"location":"architecture/middleware/#state-management","title":"State Management","text":""},{"location":"architecture/middleware/#workflowstate","title":"WorkflowState","text":"

File: src/middleware/state_machine.py

Purpose: Thread-safe state management for research workflows

Implementation: Uses ContextVar for thread-safe isolation

State Components: - evidence: list[Evidence]: Collected evidence from searches - conversation: Conversation: Iteration history (gaps, tool calls, findings, thoughts) - embedding_service: Any: Embedding service for semantic search

Methods: - add_evidence(new_evidence: list[Evidence]) -> int: Adds evidence with URL-based deduplication. Returns the number of new items added (excluding duplicates). - async search_related(query: str, n_results: int = 5) -> list[Evidence]: Semantic search for related evidence using embedding service

Initialization:

Access:

"},{"location":"architecture/middleware/#workflow-manager","title":"Workflow Manager","text":"

File: src/middleware/workflow_manager.py

Purpose: Coordinates parallel research loops

Methods: - async add_loop(loop_id: str, query: str) -> ResearchLoop: Add a new research loop to manage - async run_loops_parallel(loop_configs: list[dict], loop_func: Callable, judge_handler: Any | None = None, budget_tracker: Any | None = None) -> list[Any]: Run multiple research loops in parallel. Takes configuration dicts and a loop function. - async update_loop_status(loop_id: str, status: LoopStatus, error: str | None = None): Update loop status - async sync_loop_evidence_to_state(loop_id: str): Synchronize evidence from a specific loop to global state

Features: - Uses asyncio.gather() for parallel execution - Handles errors per loop (doesn't fail all if one fails) - Tracks loop status: pending, running, completed, failed, cancelled - Evidence deduplication across parallel loops

Usage: 

from src.middleware.workflow_manager import WorkflowManager\n\nmanager = WorkflowManager()\nawait manager.add_loop(\"loop1\", \"Research query 1\")\nawait manager.add_loop(\"loop2\", \"Research query 2\")\n\nasync def run_research(config: dict) -> str:\n    loop_id = config[\"loop_id\"]\n    query = config[\"query\"]\n    # ... research logic ...\n    return \"report\"\n\nresults = await manager.run_loops_parallel(\n    loop_configs=[\n        {\"loop_id\": \"loop1\", \"query\": \"Research query 1\"},\n        {\"loop_id\": \"loop2\", \"query\": \"Research query 2\"},\n    ],\n    loop_func=run_research,\n)\n

"},{"location":"architecture/middleware/#budget-tracker","title":"Budget Tracker","text":"

File: src/middleware/budget_tracker.py

Purpose: Tracks and enforces resource limits

Budget Components: - Tokens: LLM token usage - Time: Elapsed time in seconds - Iterations: Number of iterations

Methods: - create_budget(loop_id: str, tokens_limit: int = 100000, time_limit_seconds: float = 600.0, iterations_limit: int = 10) -> BudgetStatus: Create a budget for a specific loop - add_tokens(loop_id: str, tokens: int): Add token usage to a loop's budget - start_timer(loop_id: str): Start time tracking for a loop - update_timer(loop_id: str): Update elapsed time for a loop - increment_iteration(loop_id: str): Increment iteration count for a loop - check_budget(loop_id: str) -> tuple[bool, str]: Check if a loop's budget has been exceeded. Returns (exceeded: bool, reason: str) - can_continue(loop_id: str) -> bool: Check if a loop can continue based on budget

Token Estimation: - estimate_tokens(text: str) -> int: ~4 chars per token - estimate_llm_call_tokens(prompt: str, response: str) -> int: Estimate LLM call tokens

Usage: 

from src.middleware.budget_tracker import BudgetTracker\n\ntracker = BudgetTracker()\nbudget = tracker.create_budget(\n    loop_id=\"research_loop\",\n    tokens_limit=100000,\n    time_limit_seconds=600,\n    iterations_limit=10\n)\ntracker.start_timer(\"research_loop\")\n# ... research operations ...\ntracker.add_tokens(\"research_loop\", 5000)\ntracker.update_timer(\"research_loop\")\nexceeded, reason = tracker.check_budget(\"research_loop\")\nif exceeded:\n    # Budget exceeded, stop research\n    pass\nif not tracker.can_continue(\"research_loop\"):\n    # Budget exceeded, stop research\n    pass\n

"},{"location":"architecture/middleware/#models","title":"Models","text":"

All middleware models are defined in src/utils/models.py:

"},{"location":"architecture/middleware/#thread-safety","title":"Thread Safety","text":"

All middleware components use ContextVar for thread-safe isolation:

"},{"location":"architecture/middleware/#see-also","title":"See Also","text":""},{"location":"architecture/orchestrators/","title":"Orchestrators Architecture","text":"

DeepCritical supports multiple orchestration patterns for research workflows.

"},{"location":"architecture/orchestrators/#research-flows","title":"Research Flows","text":""},{"location":"architecture/orchestrators/#iterativeresearchflow","title":"IterativeResearchFlow","text":"

File: src/orchestrator/research_flow.py

Pattern: Generate observations \u2192 Evaluate gaps \u2192 Select tools \u2192 Execute \u2192 Judge \u2192 Continue/Complete

Agents Used: - KnowledgeGapAgent: Evaluates research completeness - ToolSelectorAgent: Selects tools for addressing gaps - ThinkingAgent: Generates observations - WriterAgent: Creates final report - JudgeHandler: Assesses evidence sufficiency

Features: - Tracks iterations, time, budget - Supports graph execution (use_graph=True) and agent chains (use_graph=False) - Iterates until research complete or constraints met

Usage:

"},{"location":"architecture/orchestrators/#deepresearchflow","title":"DeepResearchFlow","text":"

File: src/orchestrator/research_flow.py

Pattern: Planner \u2192 Parallel iterative loops per section \u2192 Synthesizer

Agents Used: - PlannerAgent: Breaks query into report sections - IterativeResearchFlow: Per-section research (parallel) - LongWriterAgent or ProofreaderAgent: Final synthesis

Features: - Uses WorkflowManager for parallel execution - Budget tracking per section and globally - State synchronization across parallel loops - Supports graph execution and agent chains

Usage:

"},{"location":"architecture/orchestrators/#graph-orchestrator","title":"Graph Orchestrator","text":"

File: src/orchestrator/graph_orchestrator.py

Purpose: Graph-based execution using Pydantic AI agents as nodes

Features: - Uses graph execution (use_graph=True) or agent chains (use_graph=False) as fallback - Routes based on research mode (iterative/deep/auto) - Streams AgentEvent objects for UI - Uses GraphExecutionContext to manage execution state

Node Types: - Agent Nodes: Execute Pydantic AI agents - State Nodes: Update or read workflow state - Decision Nodes: Make routing decisions - Parallel Nodes: Execute multiple nodes concurrently

Edge Types: - Sequential Edges: Always traversed - Conditional Edges: Traversed based on condition - Parallel Edges: Used for parallel execution branches

Special Node Handling:

The GraphOrchestrator has special handling for certain nodes:

GraphExecutionContext:

The orchestrator uses GraphExecutionContext to manage execution state: - Tracks current node, visited nodes, and node results - Manages workflow state and budget tracker - Provides methods to store and retrieve node execution results

"},{"location":"architecture/orchestrators/#orchestrator-factory","title":"Orchestrator Factory","text":"

File: src/orchestrator_factory.py

Purpose: Factory for creating orchestrators

Modes: - Simple: Legacy orchestrator (backward compatible) - Advanced: Magentic orchestrator (requires OpenAI API key) - Auto-detect: Chooses based on API key availability

Usage:

"},{"location":"architecture/orchestrators/#magentic-orchestrator","title":"Magentic Orchestrator","text":"

File: src/orchestrator_magentic.py

Purpose: Multi-agent coordination using Microsoft Agent Framework

Features: - Uses agent-framework-core - ChatAgent pattern with internal LLMs per agent - MagenticBuilder with participants: - searcher: SearchAgent (wraps SearchHandler) - hypothesizer: HypothesisAgent (generates hypotheses) - judge: JudgeAgent (evaluates evidence) - reporter: ReportAgent (generates final report) - Manager orchestrates agents via chat client (OpenAI or HuggingFace) - Event-driven: converts Magentic events to AgentEvent for UI streaming via _process_event() method - Supports max rounds, stall detection, and reset handling

Event Processing:

The orchestrator processes Magentic events and converts them to AgentEvent: - MagenticOrchestratorMessageEvent \u2192 AgentEvent with type based on message content - MagenticAgentMessageEvent \u2192 AgentEvent with type based on agent name - MagenticAgentDeltaEvent \u2192 AgentEvent for streaming updates - MagenticFinalResultEvent \u2192 AgentEvent with type \"complete\"

Requirements: - agent-framework-core package - OpenAI API key or HuggingFace authentication

"},{"location":"architecture/orchestrators/#hierarchical-orchestrator","title":"Hierarchical Orchestrator","text":"

File: src/orchestrator_hierarchical.py

Purpose: Hierarchical orchestrator using middleware and sub-teams

Features: - Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge - Adapts Magentic ChatAgent to SubIterationTeam protocol - Event-driven via asyncio.Queue for coordination - Supports sub-iteration patterns for complex research tasks

"},{"location":"architecture/orchestrators/#legacy-simple-mode","title":"Legacy Simple Mode","text":"

File: src/legacy_orchestrator.py

Purpose: Linear search-judge-synthesize loop

Features: - Uses SearchHandlerProtocol and JudgeHandlerProtocol - Generator-based design yielding AgentEvent objects - Backward compatibility for simple use cases

"},{"location":"architecture/orchestrators/#state-initialization","title":"State Initialization","text":"

All orchestrators must initialize workflow state:

"},{"location":"architecture/orchestrators/#event-streaming","title":"Event Streaming","text":"

All orchestrators yield AgentEvent objects:

Event Types: - started: Research started - searching: Search in progress - search_complete: Search completed - judging: Evidence evaluation in progress - judge_complete: Evidence evaluation completed - looping: Iteration in progress - hypothesizing: Generating hypotheses - analyzing: Statistical analysis in progress - analysis_complete: Statistical analysis completed - synthesizing: Synthesizing results - complete: Research completed - error: Error occurred - streaming: Streaming update (delta events)

Event Structure:

"},{"location":"architecture/orchestrators/#see-also","title":"See Also","text":""},{"location":"architecture/services/","title":"Services Architecture","text":"

DeepCritical provides several services for embeddings, RAG, and statistical analysis.

"},{"location":"architecture/services/#embedding-service","title":"Embedding Service","text":"

File: src/services/embeddings.py

Purpose: Local sentence-transformers for semantic search and deduplication

Features: - No API Key Required: Uses local sentence-transformers models - Async-Safe: All operations use run_in_executor() to avoid blocking the event loop - ChromaDB Storage: In-memory vector storage for embeddings - Deduplication: 0.9 similarity threshold by default (90% similarity = duplicate, configurable)

Model: Configurable via settings.local_embedding_model (default: all-MiniLM-L6-v2)

Methods: - async def embed(text: str) -> list[float]: Generate embeddings (async-safe via run_in_executor()) - async def embed_batch(texts: list[str]) -> list[list[float]]: Batch embedding (more efficient) - async def add_evidence(evidence_id: str, content: str, metadata: dict[str, Any]) -> None: Add evidence to vector store - async def search_similar(query: str, n_results: int = 5) -> list[dict[str, Any]]: Find semantically similar evidence - async def deduplicate(new_evidence: list[Evidence], threshold: float = 0.9) -> list[Evidence]: Remove semantically duplicate evidence

Usage: 

from src.services.embeddings import get_embedding_service\n\nservice = get_embedding_service()\nembedding = await service.embed(\"text to embed\")\n

"},{"location":"architecture/services/#llamaindex-rag-service","title":"LlamaIndex RAG Service","text":"

File: src/services/llamaindex_rag.py

Purpose: Retrieval-Augmented Generation using LlamaIndex

Features: - Multiple Embedding Providers: OpenAI embeddings (requires OPENAI_API_KEY) or local sentence-transformers (no API key) - Multiple LLM Providers: HuggingFace LLM (preferred) or OpenAI LLM (fallback) for query synthesis - ChromaDB Storage: Vector database for document storage (supports in-memory mode) - Metadata Preservation: Preserves source, title, URL, date, authors - Lazy Initialization: Graceful fallback if dependencies not available

Initialization Parameters: - use_openai_embeddings: bool | None: Force OpenAI embeddings (None = auto-detect) - use_in_memory: bool: Use in-memory ChromaDB client (useful for tests) - oauth_token: str | None: Optional OAuth token from HuggingFace login (takes priority over env vars)

Methods: - async def ingest_evidence(evidence: list[Evidence]) -> None: Ingest evidence into RAG - async def retrieve(query: str, top_k: int = 5) -> list[Document]: Retrieve relevant documents - async def query(query: str, top_k: int = 5) -> str: Query with RAG

Usage: 

from src.services.llamaindex_rag import get_rag_service\n\nservice = get_rag_service(\n    use_openai_embeddings=False,  # Use local embeddings\n    use_in_memory=True,  # Use in-memory ChromaDB\n    oauth_token=token  # Optional HuggingFace token\n)\nif service:\n    documents = await service.retrieve(\"query\", top_k=5)\n

"},{"location":"architecture/services/#statistical-analyzer","title":"Statistical Analyzer","text":"

File: src/services/statistical_analyzer.py

Purpose: Secure execution of AI-generated statistical code

Features: - Modal Sandbox: Secure, isolated execution environment - Code Generation: Generates Python code via LLM - Library Pinning: Version-pinned libraries in SANDBOX_LIBRARIES - Network Isolation: block_network=True by default

Libraries Available: - pandas, numpy, scipy - matplotlib, scikit-learn - statsmodels

Output: AnalysisResult with: - verdict: SUPPORTED, REFUTED, or INCONCLUSIVE - code: Generated analysis code - output: Execution output - error: Error message if execution failed

Usage: 

from src.services.statistical_analyzer import StatisticalAnalyzer\n\nanalyzer = StatisticalAnalyzer()\nresult = await analyzer.analyze(\n    hypothesis=\"Metformin reduces cancer risk\",\n    evidence=evidence_list\n)\n

"},{"location":"architecture/services/#singleton-pattern","title":"Singleton Pattern","text":"

Services use singleton patterns for lazy initialization:

EmbeddingService: Uses a global variable pattern:

LlamaIndexRAGService: Direct instantiation (no caching):

This ensures: - Single instance per process - Lazy initialization - No dependencies required at import time

"},{"location":"architecture/services/#service-availability","title":"Service Availability","text":"

Services check availability before use:

from src.utils.config import settings\n\nif settings.modal_available:\n    # Use Modal sandbox\n    pass\n\nif settings.has_openai_key:\n    # Use OpenAI embeddings for RAG\n    pass\n
"},{"location":"architecture/services/#see-also","title":"See Also","text":""},{"location":"architecture/tools/","title":"Tools Architecture","text":"

DeepCritical implements a protocol-based search tool system for retrieving evidence from multiple sources.

"},{"location":"architecture/tools/#searchtool-protocol","title":"SearchTool Protocol","text":"

All tools implement the SearchTool protocol from src/tools/base.py:

"},{"location":"architecture/tools/#rate-limiting","title":"Rate Limiting","text":"

All tools use the @retry decorator from tenacity:

Tools with API rate limits implement _rate_limit() method and use shared rate limiters from src/tools/rate_limiter.py.

"},{"location":"architecture/tools/#error-handling","title":"Error Handling","text":"

Tools raise custom exceptions:

Tools handle HTTP errors (429, 500, timeout) and return empty lists on non-critical errors (with warning logs).

"},{"location":"architecture/tools/#query-preprocessing","title":"Query Preprocessing","text":"

Tools use preprocess_query() from src/tools/query_utils.py to:

"},{"location":"architecture/tools/#evidence-conversion","title":"Evidence Conversion","text":"

All tools convert API responses to Evidence objects with:

Missing fields are handled gracefully with defaults.

"},{"location":"architecture/tools/#tool-implementations","title":"Tool Implementations","text":""},{"location":"architecture/tools/#pubmed-tool","title":"PubMed Tool","text":"

File: src/tools/pubmed.py

API: NCBI E-utilities (ESearch \u2192 EFetch)

Rate Limiting: - 0.34s between requests (3 req/sec without API key) - 0.1s between requests (10 req/sec with NCBI API key)

Features: - XML parsing with xmltodict - Handles single vs. multiple articles - Query preprocessing - Evidence conversion with metadata extraction

"},{"location":"architecture/tools/#clinicaltrials-tool","title":"ClinicalTrials Tool","text":"

File: src/tools/clinicaltrials.py

API: ClinicalTrials.gov API v2

Important: Uses requests library (NOT httpx) because WAF blocks httpx TLS fingerprint.

Execution: Runs in thread pool: await asyncio.to_thread(requests.get, ...)

Filtering: - Only interventional studies - Status: COMPLETED, ACTIVE_NOT_RECRUITING, RECRUITING, ENROLLING_BY_INVITATION

Features: - Parses nested JSON structure - Extracts trial metadata - Evidence conversion

"},{"location":"architecture/tools/#europe-pmc-tool","title":"Europe PMC Tool","text":"

File: src/tools/europepmc.py

API: Europe PMC REST API

Features: - Handles preprint markers: [PREPRINT - Not peer-reviewed] - Builds URLs from DOI or PMID - Checks pubTypeList for preprint detection - Includes both preprints and peer-reviewed articles

"},{"location":"architecture/tools/#rag-tool","title":"RAG Tool","text":"

File: src/tools/rag_tool.py

Purpose: Semantic search within collected evidence

Implementation: Wraps LlamaIndexRAGService

Features: - Returns Evidence from RAG results - Handles evidence ingestion - Semantic similarity search - Metadata preservation

"},{"location":"architecture/tools/#search-handler","title":"Search Handler","text":"

File: src/tools/search_handler.py

Purpose: Orchestrates parallel searches across multiple tools

Initialization Parameters: - tools: list[SearchTool]: List of search tools to use - timeout: float = 30.0: Timeout for each search in seconds - include_rag: bool = False: Whether to include RAG tool in searches - auto_ingest_to_rag: bool = True: Whether to automatically ingest results into RAG - oauth_token: str | None = None: Optional OAuth token from HuggingFace login (for RAG LLM)

Methods: - async def execute(query: str, max_results_per_tool: int = 10) -> SearchResult: Execute search across all tools in parallel

Features: - Uses asyncio.gather() with return_exceptions=True for parallel execution - Aggregates results into SearchResult with evidence and metadata - Handles tool failures gracefully (continues with other tools) - Deduplicates results by URL - Automatically ingests results into RAG if auto_ingest_to_rag=True - Can add RAG tool dynamically via add_rag_tool() method

"},{"location":"architecture/tools/#tool-registration","title":"Tool Registration","text":"

Tools are registered in the search handler:

from src.tools.pubmed import PubMedTool\nfrom src.tools.clinicaltrials import ClinicalTrialsTool\nfrom src.tools.europepmc import EuropePMCTool\nfrom src.tools.search_handler import SearchHandler\n\nsearch_handler = SearchHandler(\n    tools=[\n        PubMedTool(),\n        ClinicalTrialsTool(),\n        EuropePMCTool(),\n    ],\n    include_rag=True,  # Include RAG tool for semantic search\n    auto_ingest_to_rag=True,  # Automatically ingest results into RAG\n    oauth_token=token  # Optional HuggingFace token for RAG LLM\n)\n\n# Execute search\nresult = await search_handler.execute(\"query\", max_results_per_tool=10)\n
"},{"location":"architecture/tools/#see-also","title":"See Also","text":""},{"location":"architecture/workflow-diagrams/","title":"DeepCritical Workflow - Simplified Magentic Architecture","text":"

Architecture Pattern: Microsoft Magentic Orchestration Design Philosophy: Simple, dynamic, manager-driven coordination Key Innovation: Intelligent manager replaces rigid sequential phases

"},{"location":"architecture/workflow-diagrams/#1-high-level-magentic-workflow","title":"1. High-Level Magentic Workflow","text":"
flowchart TD\n    Start([User Query]) --> Manager[Magentic Manager<br/>Plan \u2022 Select \u2022 Assess \u2022 Adapt]\n\n    Manager -->|Plans| Task1[Task Decomposition]\n    Task1 --> Manager\n\n    Manager -->|Selects & Executes| HypAgent[Hypothesis Agent]\n    Manager -->|Selects & Executes| SearchAgent[Search Agent]\n    Manager -->|Selects & Executes| AnalysisAgent[Analysis Agent]\n    Manager -->|Selects & Executes| ReportAgent[Report Agent]\n\n    HypAgent -->|Results| Manager\n    SearchAgent -->|Results| Manager\n    AnalysisAgent -->|Results| Manager\n    ReportAgent -->|Results| Manager\n\n    Manager -->|Assesses Quality| Decision{Good Enough?}\n    Decision -->|No - Refine| Manager\n    Decision -->|No - Different Agent| Manager\n    Decision -->|No - Stalled| Replan[Reset Plan]\n    Replan --> Manager\n\n    Decision -->|Yes| Synthesis[Synthesize Final Result]\n    Synthesis --> Output([Research Report])\n\n    style Start fill:#e1f5e1\n    style Manager fill:#ffe6e6\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style Decision fill:#ffd6d6\n    style Synthesis fill:#d4edda\n    style Output fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#2-magentic-manager-the-6-phase-cycle","title":"2. Magentic Manager: The 6-Phase Cycle","text":"
flowchart LR\n    P1[1. Planning<br/>Analyze task<br/>Create strategy] --> P2[2. Agent Selection<br/>Pick best agent<br/>for subtask]\n    P2 --> P3[3. Execution<br/>Run selected<br/>agent with tools]\n    P3 --> P4[4. Assessment<br/>Evaluate quality<br/>Check progress]\n    P4 --> Decision{Quality OK?<br/>Progress made?}\n    Decision -->|Yes| P6[6. Synthesis<br/>Combine results<br/>Generate report]\n    Decision -->|No| P5[5. Iteration<br/>Adjust plan<br/>Try again]\n    P5 --> P2\n    P6 --> Done([Complete])\n\n    style P1 fill:#fff4e6\n    style P2 fill:#ffe6e6\n    style P3 fill:#e6f3ff\n    style P4 fill:#ffd6d6\n    style P5 fill:#fff3cd\n    style P6 fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#3-simplified-agent-architecture","title":"3. Simplified Agent Architecture","text":"
graph TB\n    subgraph \"Orchestration Layer\"\n        Manager[Magentic Manager<br/>\u2022 Plans workflow<br/>\u2022 Selects agents<br/>\u2022 Assesses quality<br/>\u2022 Adapts strategy]\n        SharedContext[(Shared Context<br/>\u2022 Hypotheses<br/>\u2022 Search Results<br/>\u2022 Analysis<br/>\u2022 Progress)]\n        Manager <--> SharedContext\n    end\n\n    subgraph \"Specialist Agents\"\n        HypAgent[Hypothesis Agent<br/>\u2022 Domain understanding<br/>\u2022 Hypothesis generation<br/>\u2022 Testability refinement]\n        SearchAgent[Search Agent<br/>\u2022 Multi-source search<br/>\u2022 RAG retrieval<br/>\u2022 Result ranking]\n        AnalysisAgent[Analysis Agent<br/>\u2022 Evidence extraction<br/>\u2022 Statistical analysis<br/>\u2022 Code execution]\n        ReportAgent[Report Agent<br/>\u2022 Report assembly<br/>\u2022 Visualization<br/>\u2022 Citation formatting]\n    end\n\n    subgraph \"MCP Tools\"\n        WebSearch[Web Search<br/>PubMed \u2022 arXiv \u2022 bioRxiv]\n        CodeExec[Code Execution<br/>Sandboxed Python]\n        RAG[RAG Retrieval<br/>Vector DB \u2022 Embeddings]\n        Viz[Visualization<br/>Charts \u2022 Graphs]\n    end\n\n    Manager -->|Selects & Directs| HypAgent\n    Manager -->|Selects & Directs| SearchAgent\n    Manager -->|Selects & Directs| AnalysisAgent\n    Manager -->|Selects & Directs| ReportAgent\n\n    HypAgent --> SharedContext\n    SearchAgent --> SharedContext\n    AnalysisAgent --> SharedContext\n    ReportAgent --> SharedContext\n\n    SearchAgent --> WebSearch\n    SearchAgent --> RAG\n    AnalysisAgent --> CodeExec\n    ReportAgent --> CodeExec\n    ReportAgent --> Viz\n\n    style Manager fill:#ffe6e6\n    style SharedContext fill:#ffe6f0\n    style HypAgent fill:#fff4e6\n    style SearchAgent fill:#fff4e6\n    style AnalysisAgent fill:#fff4e6\n    style ReportAgent fill:#fff4e6\n    style WebSearch fill:#e6f3ff\n    style CodeExec fill:#e6f3ff\n    style RAG fill:#e6f3ff\n    style Viz fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#4-dynamic-workflow-example","title":"4. Dynamic Workflow Example","text":"
sequenceDiagram\n    participant User\n    participant Manager\n    participant HypAgent\n    participant SearchAgent\n    participant AnalysisAgent\n    participant ReportAgent\n\n    User->>Manager: \"Research protein folding in Alzheimer's\"\n\n    Note over Manager: PLAN: Generate hypotheses \u2192 Search \u2192 Analyze \u2192 Report\n\n    Manager->>HypAgent: Generate 3 hypotheses\n    HypAgent-->>Manager: Returns 3 hypotheses\n    Note over Manager: ASSESS: Good quality, proceed\n\n    Manager->>SearchAgent: Search literature for hypothesis 1\n    SearchAgent-->>Manager: Returns 15 papers\n    Note over Manager: ASSESS: Good results, continue\n\n    Manager->>SearchAgent: Search for hypothesis 2\n    SearchAgent-->>Manager: Only 2 papers found\n    Note over Manager: ASSESS: Insufficient, refine search\n\n    Manager->>SearchAgent: Refined query for hypothesis 2\n    SearchAgent-->>Manager: Returns 12 papers\n    Note over Manager: ASSESS: Better, proceed\n\n    Manager->>AnalysisAgent: Analyze evidence for all hypotheses\n    AnalysisAgent-->>Manager: Returns analysis with code\n    Note over Manager: ASSESS: Complete, generate report\n\n    Manager->>ReportAgent: Create comprehensive report\n    ReportAgent-->>Manager: Returns formatted report\n    Note over Manager: SYNTHESIZE: Combine all results\n\n    Manager->>User: Final Research Report
"},{"location":"architecture/workflow-diagrams/#5-manager-decision-logic","title":"5. Manager Decision Logic","text":"
flowchart TD\n    Start([Manager Receives Task]) --> Plan[Create Initial Plan]\n\n    Plan --> Select[Select Agent for Next Subtask]\n    Select --> Execute[Execute Agent]\n    Execute --> Collect[Collect Results]\n\n    Collect --> Assess[Assess Quality & Progress]\n\n    Assess --> Q1{Quality Sufficient?}\n    Q1 -->|No| Q2{Same Agent Can Fix?}\n    Q2 -->|Yes| Feedback[Provide Specific Feedback]\n    Feedback --> Execute\n    Q2 -->|No| Different[Try Different Agent]\n    Different --> Select\n\n    Q1 -->|Yes| Q3{Task Complete?}\n    Q3 -->|No| Q4{Making Progress?}\n    Q4 -->|Yes| Select\n    Q4 -->|No - Stalled| Replan[Reset Plan & Approach]\n    Replan --> Plan\n\n    Q3 -->|Yes| Synth[Synthesize Final Result]\n    Synth --> Done([Return Report])\n\n    style Start fill:#e1f5e1\n    style Plan fill:#fff4e6\n    style Select fill:#ffe6e6\n    style Execute fill:#e6f3ff\n    style Assess fill:#ffd6d6\n    style Q1 fill:#ffe6e6\n    style Q2 fill:#ffe6e6\n    style Q3 fill:#ffe6e6\n    style Q4 fill:#ffe6e6\n    style Synth fill:#d4edda\n    style Done fill:#e1f5e1
"},{"location":"architecture/workflow-diagrams/#6-hypothesis-agent-workflow","title":"6. Hypothesis Agent Workflow","text":"
flowchart LR\n    Input[Research Query] --> Domain[Identify Domain<br/>& Key Concepts]\n    Domain --> Context[Retrieve Background<br/>Knowledge]\n    Context --> Generate[Generate 3-5<br/>Initial Hypotheses]\n    Generate --> Refine[Refine for<br/>Testability]\n    Refine --> Rank[Rank by<br/>Quality Score]\n    Rank --> Output[Return Top<br/>Hypotheses]\n\n    Output --> Struct[Hypothesis Structure:<br/>\u2022 Statement<br/>\u2022 Rationale<br/>\u2022 Testability Score<br/>\u2022 Data Requirements<br/>\u2022 Expected Outcomes]\n\n    style Input fill:#e1f5e1\n    style Output fill:#fff4e6\n    style Struct fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#7-search-agent-workflow","title":"7. Search Agent Workflow","text":"
flowchart TD\n    Input[Hypotheses] --> Strategy[Formulate Search<br/>Strategy per Hypothesis]\n\n    Strategy --> Multi[Multi-Source Search]\n\n    Multi --> PubMed[PubMed Search<br/>via MCP]\n    Multi --> ArXiv[arXiv Search<br/>via MCP]\n    Multi --> BioRxiv[bioRxiv Search<br/>via MCP]\n\n    PubMed --> Aggregate[Aggregate Results]\n    ArXiv --> Aggregate\n    BioRxiv --> Aggregate\n\n    Aggregate --> Filter[Filter & Rank<br/>by Relevance]\n    Filter --> Dedup[Deduplicate<br/>Cross-Reference]\n    Dedup --> Embed[Embed Documents<br/>via MCP]\n    Embed --> Vector[(Vector DB)]\n    Vector --> RAGRetrieval[RAG Retrieval<br/>Top-K per Hypothesis]\n    RAGRetrieval --> Output[Return Contextualized<br/>Search Results]\n\n    style Input fill:#fff4e6\n    style Multi fill:#ffe6e6\n    style Vector fill:#ffe6f0\n    style Output fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#8-analysis-agent-workflow","title":"8. Analysis Agent Workflow","text":"
flowchart TD\n    Input1[Hypotheses] --> Extract\n    Input2[Search Results] --> Extract[Extract Evidence<br/>per Hypothesis]\n\n    Extract --> Methods[Determine Analysis<br/>Methods Needed]\n\n    Methods --> Branch{Requires<br/>Computation?}\n    Branch -->|Yes| GenCode[Generate Python<br/>Analysis Code]\n    Branch -->|No| Qual[Qualitative<br/>Synthesis]\n\n    GenCode --> Execute[Execute Code<br/>via MCP Sandbox]\n    Execute --> Interpret1[Interpret<br/>Results]\n    Qual --> Interpret2[Interpret<br/>Findings]\n\n    Interpret1 --> Synthesize[Synthesize Evidence<br/>Across Sources]\n    Interpret2 --> Synthesize\n\n    Synthesize --> Verdict[Determine Verdict<br/>per Hypothesis]\n    Verdict --> Support[\u2022 Supported<br/>\u2022 Refuted<br/>\u2022 Inconclusive]\n    Support --> Gaps[Identify Knowledge<br/>Gaps & Limitations]\n    Gaps --> Output[Return Analysis<br/>Report]\n\n    style Input1 fill:#fff4e6\n    style Input2 fill:#e6f3ff\n    style Execute fill:#ffe6e6\n    style Output fill:#e6ffe6
"},{"location":"architecture/workflow-diagrams/#9-report-agent-workflow","title":"9. Report Agent Workflow","text":"
flowchart TD\n    Input1[Query] --> Assemble\n    Input2[Hypotheses] --> Assemble\n    Input3[Search Results] --> Assemble\n    Input4[Analysis] --> Assemble[Assemble Report<br/>Sections]\n\n    Assemble --> Exec[Executive Summary]\n    Assemble --> Intro[Introduction]\n    Assemble --> Methods[Methods]\n    Assemble --> Results[Results per<br/>Hypothesis]\n    Assemble --> Discussion[Discussion]\n    Assemble --> Future[Future Directions]\n    Assemble --> Refs[References]\n\n    Results --> VizCheck{Needs<br/>Visualization?}\n    VizCheck -->|Yes| GenViz[Generate Viz Code]\n    GenViz --> ExecViz[Execute via MCP<br/>Create Charts]\n    ExecViz --> Combine\n    VizCheck -->|No| Combine[Combine All<br/>Sections]\n\n    Exec --> Combine\n    Intro --> Combine\n    Methods --> Combine\n    Discussion --> Combine\n    Future --> Combine\n    Refs --> Combine\n\n    Combine --> Format[Format Output]\n    Format --> MD[Markdown]\n    Format --> PDF[PDF]\n    Format --> JSON[JSON]\n\n    MD --> Output[Return Final<br/>Report]\n    PDF --> Output\n    JSON --> Output\n\n    style Input1 fill:#e1f5e1\n    style Input2 fill:#fff4e6\n    style Input3 fill:#e6f3ff\n    style Input4 fill:#e6ffe6\n    style Output fill:#d4edda
"},{"location":"architecture/workflow-diagrams/#10-data-flow--event-streaming","title":"10. Data Flow & Event Streaming","text":"
flowchart TD\n    User[\ud83d\udc64 User] -->|Research Query| UI[Gradio UI]\n    UI -->|Submit| Manager[Magentic Manager]\n\n    Manager -->|Event: Planning| UI\n    Manager -->|Select Agent| HypAgent[Hypothesis Agent]\n    HypAgent -->|Event: Delta/Message| UI\n    HypAgent -->|Hypotheses| Context[(Shared Context)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| SearchAgent[Search Agent]\n    SearchAgent -->|MCP Request| WebSearch[Web Search Tool]\n    WebSearch -->|Results| SearchAgent\n    SearchAgent -->|Event: Delta/Message| UI\n    SearchAgent -->|Documents| Context\n    SearchAgent -->|Embeddings| VectorDB[(Vector DB)]\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| AnalysisAgent[Analysis Agent]\n    AnalysisAgent -->|MCP Request| CodeExec[Code Execution Tool]\n    CodeExec -->|Results| AnalysisAgent\n    AnalysisAgent -->|Event: Delta/Message| UI\n    AnalysisAgent -->|Analysis| Context\n\n    Context -->|Retrieved by| Manager\n    Manager -->|Select Agent| ReportAgent[Report Agent]\n    ReportAgent -->|MCP Request| CodeExec\n    ReportAgent -->|Event: Delta/Message| UI\n    ReportAgent -->|Report| Context\n\n    Manager -->|Event: Final Result| UI\n    UI -->|Display| User\n\n    style User fill:#e1f5e1\n    style UI fill:#e6f3ff\n    style Manager fill:#ffe6e6\n    style Context fill:#ffe6f0\n    style VectorDB fill:#ffe6f0\n    style WebSearch fill:#f0f0f0\n    style CodeExec fill:#f0f0f0
"},{"location":"architecture/workflow-diagrams/#11-mcp-tool-architecture","title":"11. MCP Tool Architecture","text":"
graph TB\n    subgraph \"Agent Layer\"\n        Manager[Magentic Manager]\n        HypAgent[Hypothesis Agent]\n        SearchAgent[Search Agent]\n        AnalysisAgent[Analysis Agent]\n        ReportAgent[Report Agent]\n    end\n\n    subgraph \"MCP Protocol Layer\"\n        Registry[MCP Tool Registry<br/>\u2022 Discovers tools<br/>\u2022 Routes requests<br/>\u2022 Manages connections]\n    end\n\n    subgraph \"MCP Servers\"\n        Server1[Web Search Server<br/>localhost:8001<br/>\u2022 PubMed<br/>\u2022 arXiv<br/>\u2022 bioRxiv]\n        Server2[Code Execution Server<br/>localhost:8002<br/>\u2022 Sandboxed Python<br/>\u2022 Package management]\n        Server3[RAG Server<br/>localhost:8003<br/>\u2022 Vector embeddings<br/>\u2022 Similarity search]\n        Server4[Visualization Server<br/>localhost:8004<br/>\u2022 Chart generation<br/>\u2022 Plot rendering]\n    end\n\n    subgraph \"External Services\"\n        PubMed[PubMed API]\n        ArXiv[arXiv API]\n        BioRxiv[bioRxiv API]\n        Modal[Modal Sandbox]\n        ChromaDB[(ChromaDB)]\n    end\n\n    SearchAgent -->|Request| Registry\n    AnalysisAgent -->|Request| Registry\n    ReportAgent -->|Request| Registry\n\n    Registry --> Server1\n    Registry --> Server2\n    Registry --> Server3\n    Registry --> Server4\n\n    Server1 --> PubMed\n    Server1 --> ArXiv\n    Server1 --> BioRxiv\n    Server2 --> Modal\n    Server3 --> ChromaDB\n\n    style Manager fill:#ffe6e6\n    style Registry fill:#fff4e6\n    style Server1 fill:#e6f3ff\n    style Server2 fill:#e6f3ff\n    style Server3 fill:#e6f3ff\n    style Server4 fill:#e6f3ff
"},{"location":"architecture/workflow-diagrams/#12-progress-tracking--stall-detection","title":"12. Progress Tracking & Stall Detection","text":"
stateDiagram-v2\n    [*] --> Initialization: User Query\n\n    Initialization --> Planning: Manager starts\n\n    Planning --> AgentExecution: Select agent\n\n    AgentExecution --> Assessment: Collect results\n\n    Assessment --> QualityCheck: Evaluate output\n\n    QualityCheck --> AgentExecution: Poor quality<br/>(retry < max_rounds)\n    QualityCheck --> Planning: Poor quality<br/>(try different agent)\n    QualityCheck --> NextAgent: Good quality<br/>(task incomplete)\n    QualityCheck --> Synthesis: Good quality<br/>(task complete)\n\n    NextAgent --> AgentExecution: Select next agent\n\n    state StallDetection <<choice>>\n    Assessment --> StallDetection: Check progress\n    StallDetection --> Planning: No progress<br/>(stall count < max)\n    StallDetection --> ErrorRecovery: No progress<br/>(max stalls reached)\n\n    ErrorRecovery --> PartialReport: Generate partial results\n    PartialReport --> [*]\n\n    Synthesis --> FinalReport: Combine all outputs\n    FinalReport --> [*]\n\n    note right of QualityCheck\n        Manager assesses:\n        \u2022 Output completeness\n        \u2022 Quality metrics\n        \u2022 Progress made\n    end note\n\n    note right of StallDetection\n        Stall = no new progress\n        after agent execution\n        Triggers plan reset\n    end note
"},{"location":"architecture/workflow-diagrams/#13-gradio-ui-integration","title":"13. Gradio UI Integration","text":"
graph TD\n    App[Gradio App<br/>DeepCritical Research Agent]\n\n    App --> Input[Input Section]\n    App --> Status[Status Section]\n    App --> Output[Output Section]\n\n    Input --> Query[Research Question<br/>Text Area]\n    Input --> Controls[Controls]\n    Controls --> MaxHyp[Max Hypotheses: 1-10]\n    Controls --> MaxRounds[Max Rounds: 5-20]\n    Controls --> Submit[Start Research Button]\n\n    Status --> Log[Real-time Event Log<br/>\u2022 Manager planning<br/>\u2022 Agent selection<br/>\u2022 Execution updates<br/>\u2022 Quality assessment]\n    Status --> Progress[Progress Tracker<br/>\u2022 Current agent<br/>\u2022 Round count<br/>\u2022 Stall count]\n\n    Output --> Tabs[Tabbed Results]\n    Tabs --> Tab1[Hypotheses Tab<br/>Generated hypotheses with scores]\n    Tabs --> Tab2[Search Results Tab<br/>Papers & sources found]\n    Tabs --> Tab3[Analysis Tab<br/>Evidence & verdicts]\n    Tabs --> Tab4[Report Tab<br/>Final research report]\n    Tab4 --> Download[Download Report<br/>MD / PDF / JSON]\n\n    Submit -.->|Triggers| Workflow[Magentic Workflow]\n    Workflow -.->|MagenticOrchestratorMessageEvent| Log\n    Workflow -.->|MagenticAgentDeltaEvent| Log\n    Workflow -.->|MagenticAgentMessageEvent| Log\n    Workflow -.->|MagenticFinalResultEvent| Tab4\n\n    style App fill:#e1f5e1\n    style Input fill:#fff4e6\n    style Status fill:#e6f3ff\n    style Output fill:#e6ffe6\n    style Workflow fill:#ffe6e6
"},{"location":"architecture/workflow-diagrams/#14-complete-system-context","title":"14. Complete System Context","text":"
graph LR\n    User[\ud83d\udc64 Researcher<br/>Asks research questions] -->|Submits query| DC[DeepCritical<br/>Magentic Workflow]\n\n    DC -->|Literature search| PubMed[PubMed API<br/>Medical papers]\n    DC -->|Preprint search| ArXiv[arXiv API<br/>Scientific preprints]\n    DC -->|Biology search| BioRxiv[bioRxiv API<br/>Biology preprints]\n    DC -->|Agent reasoning| Claude[Claude API<br/>Sonnet 4 / Opus]\n    DC -->|Code execution| Modal[Modal Sandbox<br/>Safe Python env]\n    DC -->|Vector storage| Chroma[ChromaDB<br/>Embeddings & RAG]\n\n    DC -->|Deployed on| HF[HuggingFace Spaces<br/>Gradio 6.0]\n\n    PubMed -->|Results| DC\n    ArXiv -->|Results| DC\n    BioRxiv -->|Results| DC\n    Claude -->|Responses| DC\n    Modal -->|Output| DC\n    Chroma -->|Context| DC\n\n    DC -->|Research report| User\n\n    style User fill:#e1f5e1\n    style DC fill:#ffe6e6\n    style PubMed fill:#e6f3ff\n    style ArXiv fill:#e6f3ff\n    style BioRxiv fill:#e6f3ff\n    style Claude fill:#ffd6d6\n    style Modal fill:#f0f0f0\n    style Chroma fill:#ffe6f0\n    style HF fill:#d4edda
"},{"location":"architecture/workflow-diagrams/#15-workflow-timeline-simplified","title":"15. Workflow Timeline (Simplified)","text":"
gantt\n    title DeepCritical Magentic Workflow - Typical Execution\n    dateFormat mm:ss\n    axisFormat %M:%S\n\n    section Manager Planning\n    Initial planning         :p1, 00:00, 10s\n\n    section Hypothesis Agent\n    Generate hypotheses      :h1, after p1, 30s\n    Manager assessment       :h2, after h1, 5s\n\n    section Search Agent\n    Search hypothesis 1      :s1, after h2, 20s\n    Search hypothesis 2      :s2, after s1, 20s\n    Search hypothesis 3      :s3, after s2, 20s\n    RAG processing          :s4, after s3, 15s\n    Manager assessment      :s5, after s4, 5s\n\n    section Analysis Agent\n    Evidence extraction     :a1, after s5, 15s\n    Code generation        :a2, after a1, 20s\n    Code execution         :a3, after a2, 25s\n    Synthesis              :a4, after a3, 20s\n    Manager assessment     :a5, after a4, 5s\n\n    section Report Agent\n    Report assembly        :r1, after a5, 30s\n    Visualization          :r2, after r1, 15s\n    Formatting             :r3, after r2, 10s\n\n    section Manager Synthesis\n    Final synthesis        :f1, after r3, 10s
"},{"location":"architecture/workflow-diagrams/#key-differences-from-original-design","title":"Key Differences from Original Design","text":"Aspect Original (Judge-in-Loop) New (Magentic) Control Flow Fixed sequential phases Dynamic agent selection Quality Control Separate Judge Agent Manager assessment built-in Retry Logic Phase-level with feedback Agent-level with adaptation Flexibility Rigid 4-phase pipeline Adaptive workflow Complexity 5 agents (including Judge) 4 agents (no Judge) Progress Tracking Manual state management Built-in round/stall detection Agent Coordination Sequential handoff Manager-driven dynamic selection Error Recovery Retry same phase Try different agent or replan"},{"location":"architecture/workflow-diagrams/#simplified-design-principles","title":"Simplified Design Principles","text":"
  1. Manager is Intelligent: LLM-powered manager handles planning, selection, and quality assessment
  2. No Separate Judge: Manager's assessment phase replaces dedicated Judge Agent
  3. Dynamic Workflow: Agents can be called multiple times in any order based on need
  4. Built-in Safety: max_round_count (15) and max_stall_count (3) prevent infinite loops
  5. Event-Driven UI: Real-time streaming updates to Gradio interface
  6. MCP-Powered Tools: All external capabilities via Model Context Protocol
  7. Shared Context: Centralized state accessible to all agents
  8. Progress Awareness: Manager tracks what's been done and what's needed
"},{"location":"architecture/workflow-diagrams/#legend","title":"Legend","text":""},{"location":"architecture/workflow-diagrams/#implementation-highlights","title":"Implementation Highlights","text":"

Simple 4-Agent Setup:

Manager handles quality assessment in its instructions: - Checks hypothesis quality (testable, novel, clear) - Validates search results (relevant, authoritative, recent) - Assesses analysis soundness (methodology, evidence, conclusions) - Ensures report completeness (all sections, proper citations)

No separate Judge Agent needed - manager does it all!

Document Version: 2.0 (Magentic Simplified) Last Updated: 2025-11-24 Architecture: Microsoft Magentic Orchestration Pattern Agents: 4 (Hypothesis, Search, Analysis, Report) + 1 Manager License: MIT

"},{"location":"architecture/workflow-diagrams/#see-also","title":"See Also","text":""},{"location":"configuration/","title":"Configuration Guide","text":""},{"location":"configuration/#overview","title":"Overview","text":"

DeepCritical uses Pydantic Settings for centralized configuration management. All settings are defined in the Settings class in src/utils/config.py and can be configured via environment variables or a .env file.

The configuration system provides:

"},{"location":"configuration/#quick-start","title":"Quick Start","text":"
  1. Create a .env file in the project root
  2. Set at least one LLM API key (OPENAI_API_KEY, ANTHROPIC_API_KEY, or HF_TOKEN)
  3. Optionally configure other services as needed
  4. The application will automatically load and validate your configuration
"},{"location":"configuration/#configuration-system-architecture","title":"Configuration System Architecture","text":""},{"location":"configuration/#settings-class","title":"Settings Class","text":"

The [Settings][settings-class] class extends BaseSettings from pydantic_settings and defines all application configuration:

View source

"},{"location":"configuration/#singleton-instance","title":"Singleton Instance","text":"

A global settings instance is available for import:

View source

"},{"location":"configuration/#usage-pattern","title":"Usage Pattern","text":"

Access configuration throughout the codebase:

from src.utils.config import settings\n\n# Check if API keys are available\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\n# Access configuration values\nmax_iterations = settings.max_iterations\nweb_search_provider = settings.web_search_provider\n
"},{"location":"configuration/#required-configuration","title":"Required Configuration","text":""},{"location":"configuration/#llm-provider","title":"LLM Provider","text":"

You must configure at least one LLM provider. The system supports:

"},{"location":"configuration/#openai-configuration","title":"OpenAI Configuration","text":"
LLM_PROVIDER=openai\nOPENAI_API_KEY=your_openai_api_key_here\nOPENAI_MODEL=gpt-5.1\n

The default model is defined in the Settings class:

"},{"location":"configuration/#anthropic-configuration","title":"Anthropic Configuration","text":"
LLM_PROVIDER=anthropic\nANTHROPIC_API_KEY=your_anthropic_api_key_here\nANTHROPIC_MODEL=claude-sonnet-4-5-20250929\n

The default model is defined in the Settings class:

"},{"location":"configuration/#huggingface-configuration","title":"HuggingFace Configuration","text":"

HuggingFace can work without an API key for public models, but an API key provides higher rate limits:

# Option 1: Using HF_TOKEN (preferred)\nHF_TOKEN=your_huggingface_token_here\n\n# Option 2: Using HUGGINGFACE_API_KEY (alternative)\nHUGGINGFACE_API_KEY=your_huggingface_api_key_here\n\n# Default model\nHUGGINGFACE_MODEL=meta-llama/Llama-3.1-8B-Instruct\n

The HuggingFace token can be set via either environment variable:

"},{"location":"configuration/#optional-configuration","title":"Optional Configuration","text":""},{"location":"configuration/#embedding-configuration","title":"Embedding Configuration","text":"

DeepCritical supports multiple embedding providers for semantic search and RAG:

# Embedding Provider: \"openai\", \"local\", or \"huggingface\"\nEMBEDDING_PROVIDER=local\n\n# OpenAI Embedding Model (used by LlamaIndex RAG)\nOPENAI_EMBEDDING_MODEL=text-embedding-3-small\n\n# Local Embedding Model (sentence-transformers, used by EmbeddingService)\nLOCAL_EMBEDDING_MODEL=all-MiniLM-L6-v2\n\n# HuggingFace Embedding Model\nHUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2\n

The embedding provider configuration:

Note: OpenAI embeddings require OPENAI_API_KEY. The local provider (default) uses sentence-transformers and requires no API key.

"},{"location":"configuration/#web-search-configuration","title":"Web Search Configuration","text":"

DeepCritical supports multiple web search providers:

# Web Search Provider: \"serper\", \"searchxng\", \"brave\", \"tavily\", or \"duckduckgo\"\n# Default: \"duckduckgo\" (no API key required)\nWEB_SEARCH_PROVIDER=duckduckgo\n\n# Serper API Key (for Google search via Serper)\nSERPER_API_KEY=your_serper_api_key_here\n\n# SearchXNG Host URL (for self-hosted search)\nSEARCHXNG_HOST=http://localhost:8080\n\n# Brave Search API Key\nBRAVE_API_KEY=your_brave_api_key_here\n\n# Tavily API Key\nTAVILY_API_KEY=your_tavily_api_key_here\n

The web search provider configuration:

Note: DuckDuckGo is the default and requires no API key, making it ideal for development and testing.

"},{"location":"configuration/#pubmed-configuration","title":"PubMed Configuration","text":"

PubMed search supports optional NCBI API key for higher rate limits:

# NCBI API Key (optional, for higher rate limits: 10 req/sec vs 3 req/sec)\nNCBI_API_KEY=your_ncbi_api_key_here\n

The PubMed tool uses this configuration:

"},{"location":"configuration/#agent-configuration","title":"Agent Configuration","text":"

Control agent behavior and research loop execution:

# Maximum iterations per research loop (1-50, default: 10)\nMAX_ITERATIONS=10\n\n# Search timeout in seconds\nSEARCH_TIMEOUT=30\n\n# Use graph-based execution for research flows\nUSE_GRAPH_EXECUTION=false\n

The agent configuration fields:

"},{"location":"configuration/#budget--rate-limiting-configuration","title":"Budget & Rate Limiting Configuration","text":"

Control resource limits for research loops:

# Default token budget per research loop (1000-1000000, default: 100000)\nDEFAULT_TOKEN_LIMIT=100000\n\n# Default time limit per research loop in minutes (1-120, default: 10)\nDEFAULT_TIME_LIMIT_MINUTES=10\n\n# Default iterations limit per research loop (1-50, default: 10)\nDEFAULT_ITERATIONS_LIMIT=10\n

The budget configuration with validation:

"},{"location":"configuration/#rag-service-configuration","title":"RAG Service Configuration","text":"

Configure the Retrieval-Augmented Generation service:

# ChromaDB collection name for RAG\nRAG_COLLECTION_NAME=deepcritical_evidence\n\n# Number of top results to retrieve from RAG (1-50, default: 5)\nRAG_SIMILARITY_TOP_K=5\n\n# Automatically ingest evidence into RAG\nRAG_AUTO_INGEST=true\n

The RAG configuration:

"},{"location":"configuration/#chromadb-configuration","title":"ChromaDB Configuration","text":"

Configure the vector database for embeddings and RAG:

# ChromaDB storage path\nCHROMA_DB_PATH=./chroma_db\n\n# Whether to persist ChromaDB to disk\nCHROMA_DB_PERSIST=true\n\n# ChromaDB server host (for remote ChromaDB, optional)\nCHROMA_DB_HOST=localhost\n\n# ChromaDB server port (for remote ChromaDB, optional)\nCHROMA_DB_PORT=8000\n

The ChromaDB configuration:

"},{"location":"configuration/#external-services","title":"External Services","text":""},{"location":"configuration/#modal-configuration","title":"Modal Configuration","text":"

Modal is used for secure sandbox execution of statistical analysis:

# Modal Token ID (for Modal sandbox execution)\nMODAL_TOKEN_ID=your_modal_token_id_here\n\n# Modal Token Secret\nMODAL_TOKEN_SECRET=your_modal_token_secret_here\n

The Modal configuration:

"},{"location":"configuration/#logging-configuration","title":"Logging Configuration","text":"

Configure structured logging:

# Log Level: \"DEBUG\", \"INFO\", \"WARNING\", or \"ERROR\"\nLOG_LEVEL=INFO\n

The logging configuration:

Logging is configured via the configure_logging() function:

"},{"location":"configuration/#configuration-properties","title":"Configuration Properties","text":"

The Settings class provides helpful properties for checking configuration state:

"},{"location":"configuration/#api-key-availability","title":"API Key Availability","text":"

Check which API keys are available:

Usage:

from src.utils.config import settings\n\n# Check API key availability\nif settings.has_openai_key:\n    # Use OpenAI\n    pass\n\nif settings.has_anthropic_key:\n    # Use Anthropic\n    pass\n\nif settings.has_huggingface_key:\n    # Use HuggingFace\n    pass\n\nif settings.has_any_llm_key:\n    # At least one LLM is available\n    pass\n
"},{"location":"configuration/#service-availability","title":"Service Availability","text":"

Check if external services are configured:

Usage:

from src.utils.config import settings\n\n# Check service availability\nif settings.modal_available:\n    # Use Modal sandbox\n    pass\n\nif settings.web_search_available:\n    # Web search is configured\n    pass\n
"},{"location":"configuration/#api-key-retrieval","title":"API Key Retrieval","text":"

Get the API key for the configured provider:

For OpenAI-specific operations (e.g., Magentic mode):

"},{"location":"configuration/#configuration-usage-in-codebase","title":"Configuration Usage in Codebase","text":"

The configuration system is used throughout the codebase:

"},{"location":"configuration/#llm-factory","title":"LLM Factory","text":"

The LLM factory uses settings to create appropriate models:

"},{"location":"configuration/#embedding-service","title":"Embedding Service","text":"

The embedding service uses local embedding model configuration:

"},{"location":"configuration/#orchestrator-factory","title":"Orchestrator Factory","text":"

The orchestrator factory uses settings to determine mode:

"},{"location":"configuration/#environment-variables-reference","title":"Environment Variables Reference","text":""},{"location":"configuration/#required-at-least-one-llm","title":"Required (at least one LLM)","text":""},{"location":"configuration/#llm-configuration-variables","title":"LLM Configuration Variables","text":""},{"location":"configuration/#embedding-configuration-variables","title":"Embedding Configuration Variables","text":""},{"location":"configuration/#web-search-configuration-variables","title":"Web Search Configuration Variables","text":""},{"location":"configuration/#pubmed-configuration-variables","title":"PubMed Configuration Variables","text":""},{"location":"configuration/#agent-configuration-variables","title":"Agent Configuration Variables","text":""},{"location":"configuration/#budget-configuration-variables","title":"Budget Configuration Variables","text":""},{"location":"configuration/#rag-configuration-variables","title":"RAG Configuration Variables","text":""},{"location":"configuration/#chromadb-configuration-variables","title":"ChromaDB Configuration Variables","text":""},{"location":"configuration/#external-services-variables","title":"External Services Variables","text":""},{"location":"configuration/#logging-configuration-variables","title":"Logging Configuration Variables","text":""},{"location":"configuration/#validation","title":"Validation","text":"

Settings are validated on load using Pydantic validation:

"},{"location":"configuration/#validation-examples","title":"Validation Examples","text":"

The max_iterations field has range validation:

The llm_provider field has literal validation:

"},{"location":"configuration/#error-handling","title":"Error Handling","text":"

Configuration errors raise ConfigurationError from src/utils/exceptions.py:

```22:25:src/utils/exceptions.py class ConfigurationError(DeepCriticalError): \"\"\"Raised when configuration is invalid.\"\"\"

pass\n

```

"},{"location":"configuration/#error-handling-example","title":"Error Handling Example","text":"

python from src.utils.config import settings from src.utils.exceptions import ConfigurationError try: api_key = settings.get_api_key() except ConfigurationError as e: print(f\"Configuration error: {e}\")

"},{"location":"configuration/#common-configuration-errors","title":"Common Configuration Errors","text":"
  1. Missing API Key: When get_api_key() is called but the required API key is not set
  2. Invalid Provider: When llm_provider is set to an unsupported value
  3. Out of Range: When numeric values exceed their min/max constraints
  4. Invalid Literal: When enum fields receive unsupported values
"},{"location":"configuration/#configuration-best-practices","title":"Configuration Best Practices","text":"
  1. Use .env File: Store sensitive keys in .env file (add to .gitignore)
  2. Check Availability: Use properties like has_openai_key before accessing API keys
  3. Handle Errors: Always catch ConfigurationError when calling get_api_key()
  4. Validate Early: Configuration is validated on import, so errors surface immediately
  5. Use Defaults: Leverage sensible defaults for optional configuration
"},{"location":"configuration/#future-enhancements","title":"Future Enhancements","text":"

The following configurations are planned for future phases:

  1. Additional LLM Providers: DeepSeek, OpenRouter, Gemini, Perplexity, Azure OpenAI, Local models
  2. Model Selection: Reasoning/main/fast model configuration
  3. Service Integration: Additional service integrations and configurations
"},{"location":"contributing/","title":"Contributing to The DETERMINATOR","text":"

Thank you for your interest in contributing to The DETERMINATOR! This guide will help you get started.

Note on Project Names: \"The DETERMINATOR\" is the product name, \"DeepCritical\" is the organization/project name, and \"determinator\" is the Python package name.

"},{"location":"contributing/#git-workflow","title":"Git Workflow","text":""},{"location":"contributing/#repository-information","title":"Repository Information","text":""},{"location":"contributing/#dual-repository-setup","title":"Dual Repository Setup","text":"

This project uses a dual repository setup:

"},{"location":"contributing/#remote-configuration","title":"Remote Configuration","text":"

When cloning, set up remotes as follows:

# Clone from GitHub\ngit clone https://github.com/DeepCritical/GradioDemo.git\ncd GradioDemo\n\n# Add HuggingFace remote (optional, for deployment)\ngit remote add huggingface-upstream https://huggingface.co/spaces/DataQuests/DeepCritical\n

Important: Never push directly to main or dev on HuggingFace. Always work through GitHub PRs. GitHub is the source of truth; HuggingFace is for deployment/demo only.

"},{"location":"contributing/#package-manager","title":"Package Manager","text":"

This project uses uv as the package manager. All commands should be prefixed with uv run to ensure they run in the correct environment.

"},{"location":"contributing/#installation","title":"Installation","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync all dependencies including dev extras\nuv sync --all-extras\n\n# Install pre-commit hooks\nuv run pre-commit install\n
"},{"location":"contributing/#development-commands","title":"Development Commands","text":"
# Installation\nuv sync --all-extras              # Install all dependencies including dev\nuv run pre-commit install          # Install pre-commit hooks\n\n# Code Quality Checks (run all before committing)\nuv run ruff check src tests       # Lint with ruff\nuv run ruff format src tests      # Format with ruff\nuv run mypy src                   # Type checking\nuv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire  # Tests with coverage\n\n# Testing Commands\nuv run pytest tests/unit/ -v -m \"not openai\" -p no:logfire              # Run unit tests (excludes OpenAI tests)\nuv run pytest tests/ -v -m \"huggingface\" -p no:logfire                 # Run HuggingFace tests\nuv run pytest tests/ -v -p no:logfire                                  # Run all tests\nuv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire  # Tests with terminal coverage\nuv run pytest --cov=src --cov-report=html -p no:logfire                # Generate HTML coverage report (opens htmlcov/index.html)\n\n# Documentation Commands\nuv run mkdocs build                # Build documentation\nuv run mkdocs serve                # Serve documentation locally (http://127.0.0.1:8000)\n
"},{"location":"contributing/#test-markers","title":"Test Markers","text":"

The project uses pytest markers to categorize tests. See Testing Guidelines for details:

Note: The -p no:logfire flag disables the logfire plugin to avoid conflicts during testing.

"},{"location":"contributing/#getting-started","title":"Getting Started","text":"

  1. Fork the repository on GitHub: DeepCritical/GradioDemo

  2. Clone your fork:

git clone https://github.com/yourusername/GradioDemo.git\ncd GradioDemo\n
  1. Install dependencies:
uv sync --all-extras\nuv run pre-commit install\n
  1. Create a feature branch:
git checkout -b yourname-feature-name\n

  1. Make your changes following the guidelines below

  2. Run checks:

uv run ruff check src tests\nuv run mypy src\nuv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire\n
  1. Commit and push:
git commit -m \"Description of changes\"\ngit push origin yourname-feature-name\n
  1. Create a pull request on GitHub
"},{"location":"contributing/#development-guidelines","title":"Development Guidelines","text":""},{"location":"contributing/#code-style","title":"Code Style","text":""},{"location":"contributing/#error-handling","title":"Error Handling","text":""},{"location":"contributing/#testing","title":"Testing","text":""},{"location":"contributing/#implementation-patterns","title":"Implementation Patterns","text":""},{"location":"contributing/#prompt-engineering","title":"Prompt Engineering","text":""},{"location":"contributing/#code-quality","title":"Code Quality","text":""},{"location":"contributing/#mcp-integration","title":"MCP Integration","text":""},{"location":"contributing/#mcp-tools","title":"MCP Tools","text":""},{"location":"contributing/#gradio-mcp-server","title":"Gradio MCP Server","text":""},{"location":"contributing/#common-pitfalls","title":"Common Pitfalls","text":"
  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Hallucinated citations: Always validate references
  4. Global mutable state: Use ContextVar or pass via parameters
  5. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
  6. Rate limiting: Always implement for external APIs
  7. Error chaining: Always use from e when raising exceptions
"},{"location":"contributing/#key-principles","title":"Key Principles","text":"
  1. Type Safety First: All code must pass mypy --strict
  2. Async Everything: All I/O must be async
  3. Test-Driven: Write tests before implementation
  4. No Hallucinations: Validate all citations
  5. Graceful Degradation: Support free tier (HF Inference) when no API keys
  6. Lazy Loading: Don't require optional dependencies at import time
  7. Structured Logging: Use structlog, never print()
  8. Error Chaining: Always preserve exception context
"},{"location":"contributing/#pull-request-process","title":"Pull Request Process","text":"
  1. Ensure all checks pass: uv run ruff check src tests && uv run mypy src && uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire
  2. Update documentation if needed
  3. Add tests for new features
  4. Update CHANGELOG if applicable
  5. Request review from maintainers
  6. Address review feedback
  7. Wait for approval before merging
"},{"location":"contributing/#project-structure","title":"Project Structure","text":""},{"location":"contributing/#questions","title":"Questions?","text":"

Thank you for contributing to The DETERMINATOR!

"},{"location":"contributing/code-quality/","title":"Code Quality & Documentation","text":"

This document outlines code quality standards and documentation requirements for The DETERMINATOR.

"},{"location":"contributing/code-quality/#linting","title":"Linting","text":""},{"location":"contributing/code-quality/#type-checking","title":"Type Checking","text":""},{"location":"contributing/code-quality/#pre-commit","title":"Pre-commit","text":"

Pre-commit hooks run automatically on commit to ensure code quality. Configuration is in .pre-commit-config.yaml.

"},{"location":"contributing/code-quality/#installation","title":"Installation","text":"
# Install dependencies (includes pre-commit package)\nuv sync --all-extras\n\n# Set up git hooks (must be run separately)\nuv run pre-commit install\n

Note: uv sync --all-extras installs the pre-commit package, but you must run uv run pre-commit install separately to set up the git hooks.

"},{"location":"contributing/code-quality/#pre-commit-hooks","title":"Pre-commit Hooks","text":"

The following hooks run automatically on commit:

  1. ruff: Lints code and fixes issues automatically
  2. Runs on: src/ (excludes tests/, reference_repos/)

  3. Auto-fixes: Yes

  4. ruff-format: Formats code with ruff

  5. Runs on: src/ (excludes tests/, reference_repos/)

  6. Auto-fixes: Yes

  7. mypy: Type checking

  8. Runs on: src/ (excludes folder/)

  9. Additional dependencies: pydantic, pydantic-settings, tenacity, pydantic-ai

  10. pytest-unit: Runs unit tests (excludes OpenAI and embedding_provider tests)

  11. Runs: tests/unit/ with -m \"not openai and not embedding_provider\"

  12. Always runs: Yes (not just on changed files)

  13. pytest-local-embeddings: Runs local embedding tests

  14. Runs: tests/ with -m \"local_embeddings\"
  15. Always runs: Yes
"},{"location":"contributing/code-quality/#manual-pre-commit-run","title":"Manual Pre-commit Run","text":"

To run pre-commit hooks manually (without committing):

uv run pre-commit run --all-files\n
"},{"location":"contributing/code-quality/#troubleshooting","title":"Troubleshooting","text":""},{"location":"contributing/code-quality/#documentation","title":"Documentation","text":""},{"location":"contributing/code-quality/#building-documentation","title":"Building Documentation","text":"

Documentation is built using MkDocs. Source files are in docs/, and the configuration is in mkdocs.yml.

# Build documentation\nuv run mkdocs build\n\n# Serve documentation locally (http://127.0.0.1:8000)\nuv run mkdocs serve\n

The documentation site is published at: https://deepcritical.github.io/GradioDemo/

"},{"location":"contributing/code-quality/#docstrings","title":"Docstrings","text":"

Example:

"},{"location":"contributing/code-quality/#code-comments","title":"Code Comments","text":""},{"location":"contributing/code-quality/#see-also","title":"See Also","text":""},{"location":"contributing/code-style/","title":"Code Style & Conventions","text":"

This document outlines the code style and conventions for The DETERMINATOR.

"},{"location":"contributing/code-style/#package-manager","title":"Package Manager","text":"

This project uses uv as the package manager. All commands should be prefixed with uv run to ensure they run in the correct environment.

"},{"location":"contributing/code-style/#installation","title":"Installation","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync all dependencies including dev extras\nuv sync --all-extras\n
"},{"location":"contributing/code-style/#running-commands","title":"Running Commands","text":"

All development commands should use uv run prefix:

# Instead of: pytest tests/\nuv run pytest tests/\n\n# Instead of: ruff check src\nuv run ruff check src\n\n# Instead of: mypy src\nuv run mypy src\n

This ensures commands run in the correct virtual environment managed by uv.

"},{"location":"contributing/code-style/#type-safety","title":"Type Safety","text":""},{"location":"contributing/code-style/#pydantic-models","title":"Pydantic Models","text":""},{"location":"contributing/code-style/#async-patterns","title":"Async Patterns","text":" 
loop = asyncio.get_running_loop()\nresult = await loop.run_in_executor(None, cpu_bound_function, args)\n
"},{"location":"contributing/code-style/#common-pitfalls","title":"Common Pitfalls","text":"
  1. Blocking the event loop: Never use sync I/O in async functions
  2. Missing type hints: All functions must have complete type annotations
  3. Global mutable state: Use ContextVar or pass via parameters
  4. Import errors: Lazy-load optional dependencies (magentic, modal, embeddings)
"},{"location":"contributing/code-style/#see-also","title":"See Also","text":""},{"location":"contributing/error-handling/","title":"Error Handling & Logging","text":"

This document outlines error handling and logging conventions for The DETERMINATOR.

"},{"location":"contributing/error-handling/#exception-hierarchy","title":"Exception Hierarchy","text":"

Use custom exception hierarchy (src/utils/exceptions.py):

"},{"location":"contributing/error-handling/#error-handling-rules","title":"Error Handling Rules","text":" 
logger.error(\"Operation failed\", error=str(e), context=value)\n
"},{"location":"contributing/error-handling/#logging","title":"Logging","text":""},{"location":"contributing/error-handling/#logging-examples","title":"Logging Examples","text":"
logger.info(\"Starting search\", query=query, tools=[t.name for t in tools])\nlogger.warning(\"Search tool failed\", tool=tool.name, error=str(result))\nlogger.error(\"Assessment failed\", error=str(e))\n
"},{"location":"contributing/error-handling/#error-chaining","title":"Error Chaining","text":"

Always preserve exception context:

try:\n    result = await api_call()\nexcept httpx.HTTPError as e:\n    raise SearchError(f\"API call failed: {e}\") from e\n
"},{"location":"contributing/error-handling/#see-also","title":"See Also","text":""},{"location":"contributing/implementation-patterns/","title":"Implementation Patterns","text":"

This document outlines common implementation patterns used in The DETERMINATOR.

"},{"location":"contributing/implementation-patterns/#search-tools","title":"Search Tools","text":"

All tools implement SearchTool protocol (src/tools/base.py):

Example pattern:

class MySearchTool:\n    @property\n    def name(self) -> str:\n        return \"mytool\"\n    \n    @retry(stop=stop_after_attempt(3), wait=wait_exponential(...))\n    async def search(self, query: str, max_results: int = 10) -> list[Evidence]:\n        # Implementation\n        return evidence_list\n
"},{"location":"contributing/implementation-patterns/#judge-handlers","title":"Judge Handlers","text":""},{"location":"contributing/implementation-patterns/#agent-factory-pattern","title":"Agent Factory Pattern","text":""},{"location":"contributing/implementation-patterns/#state-management","title":"State Management","text":""},{"location":"contributing/implementation-patterns/#singleton-pattern","title":"Singleton Pattern","text":"

Use @lru_cache(maxsize=1) for singletons:

"},{"location":"contributing/implementation-patterns/#see-also","title":"See Also","text":""},{"location":"contributing/prompt-engineering/","title":"Prompt Engineering & Citation Validation","text":"

This document outlines prompt engineering guidelines and citation validation rules.

"},{"location":"contributing/prompt-engineering/#judge-prompts","title":"Judge Prompts","text":""},{"location":"contributing/prompt-engineering/#hypothesis-prompts","title":"Hypothesis Prompts","text":""},{"location":"contributing/prompt-engineering/#report-prompts","title":"Report Prompts","text":""},{"location":"contributing/prompt-engineering/#citation-validation","title":"Citation Validation","text":""},{"location":"contributing/prompt-engineering/#citation-validation-rules","title":"Citation Validation Rules","text":"
  1. Every reference URL must EXACTLY match a provided evidence URL
  2. Do NOT invent, fabricate, or hallucinate any references
  3. Do NOT modify paper titles, authors, dates, or URLs
  4. If unsure about a citation, OMIT it rather than guess
  5. Copy URLs exactly as provided - do not create similar-looking URLs
"},{"location":"contributing/prompt-engineering/#evidence-selection","title":"Evidence Selection","text":""},{"location":"contributing/prompt-engineering/#see-also","title":"See Also","text":""},{"location":"contributing/testing/","title":"Testing Requirements","text":"

This document outlines testing requirements and guidelines for The DETERMINATOR.

"},{"location":"contributing/testing/#test-structure","title":"Test Structure","text":""},{"location":"contributing/testing/#test-markers","title":"Test Markers","text":"

The project uses pytest markers to categorize tests. These markers are defined in pyproject.toml:

"},{"location":"contributing/testing/#running-tests-by-marker","title":"Running Tests by Marker","text":"
# Run only unit tests (excludes OpenAI tests by default)\nuv run pytest tests/unit/ -v -m \"not openai\" -p no:logfire\n\n# Run HuggingFace tests\nuv run pytest tests/ -v -m \"huggingface\" -p no:logfire\n\n# Run all tests\nuv run pytest tests/ -v -p no:logfire\n\n# Run only local embedding tests\nuv run pytest tests/ -v -m \"local_embeddings\" -p no:logfire\n\n# Exclude slow tests\nuv run pytest tests/ -v -m \"not slow\" -p no:logfire\n

Note: The -p no:logfire flag disables the logfire plugin to avoid conflicts during testing.

"},{"location":"contributing/testing/#mocking","title":"Mocking","text":""},{"location":"contributing/testing/#tdd-workflow","title":"TDD Workflow","text":"
  1. Write failing test in tests/unit/
  2. Implement in src/
  3. Ensure test passes
  4. Run checks: uv run ruff check src tests && uv run mypy src && uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire
"},{"location":"contributing/testing/#test-command-examples","title":"Test Command Examples","text":"
# Run unit tests (default, excludes OpenAI tests)\nuv run pytest tests/unit/ -v -m \"not openai\" -p no:logfire\n\n# Run HuggingFace tests\nuv run pytest tests/ -v -m \"huggingface\" -p no:logfire\n\n# Run all tests\nuv run pytest tests/ -v -p no:logfire\n
"},{"location":"contributing/testing/#test-examples","title":"Test Examples","text":"
@pytest.mark.unit\nasync def test_pubmed_search(mock_httpx_client):\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=5)\n    assert len(results) > 0\n    assert all(isinstance(r, Evidence) for r in results)\n\n@pytest.mark.integration\nasync def test_real_pubmed_search():\n    tool = PubMedTool()\n    results = await tool.search(\"metformin\", max_results=3)\n    assert len(results) <= 3\n
"},{"location":"contributing/testing/#test-coverage","title":"Test Coverage","text":""},{"location":"contributing/testing/#terminal-coverage-report","title":"Terminal Coverage Report","text":"
uv run pytest --cov=src --cov-report=term-missing tests/unit/ -v -m \"not openai\" -p no:logfire\n

This shows coverage with missing lines highlighted in the terminal output.

"},{"location":"contributing/testing/#html-coverage-report","title":"HTML Coverage Report","text":"
uv run pytest --cov=src --cov-report=html -p no:logfire\n

This generates an HTML coverage report in htmlcov/index.html. Open this file in your browser to see detailed coverage information.

"},{"location":"contributing/testing/#coverage-goals","title":"Coverage Goals","text":""},{"location":"contributing/testing/#see-also","title":"See Also","text":""},{"location":"getting-started/examples/","title":"Examples","text":"

This page provides examples of using The DETERMINATOR for various research tasks.

"},{"location":"getting-started/examples/#basic-research-query","title":"Basic Research Query","text":""},{"location":"getting-started/examples/#example-1-drug-information","title":"Example 1: Drug Information","text":"

Query: 

What are the latest treatments for Alzheimer's disease?\n

What The DETERMINATOR Does: 1. Searches PubMed for recent papers 2. Searches ClinicalTrials.gov for active trials 3. Evaluates evidence quality 4. Synthesizes findings into a comprehensive report

"},{"location":"getting-started/examples/#example-2-clinical-trial-search","title":"Example 2: Clinical Trial Search","text":"

Query: 

What clinical trials are investigating metformin for cancer prevention?\n

What The DETERMINATOR Does:

  1. Searches ClinicalTrials.gov for relevant trials
  2. Searches PubMed for supporting literature
  3. Provides trial details and status
  4. Summarizes findings
"},{"location":"getting-started/examples/#advanced-research-queries","title":"Advanced Research Queries","text":""},{"location":"getting-started/examples/#example-3-comprehensive-review","title":"Example 3: Comprehensive Review","text":"

Query:

Review the evidence for using metformin as an anti-aging intervention, \nincluding clinical trials, mechanisms of action, and safety profile.\n

What The DETERMINATOR Does: 1. Uses deep research mode (multi-section) 2. Searches multiple sources in parallel 3. Generates sections on: - Clinical trials - Mechanisms of action - Safety profile 4. Synthesizes comprehensive report

"},{"location":"getting-started/examples/#example-4-hypothesis-testing","title":"Example 4: Hypothesis Testing","text":"

Query: 

Test the hypothesis that regular exercise reduces Alzheimer's disease risk.\n

What The DETERMINATOR Does: 1. Generates testable hypotheses 2. Searches for supporting/contradicting evidence 3. Performs statistical analysis (if Modal configured) 4. Provides verdict: SUPPORTED, REFUTED, or INCONCLUSIVE

"},{"location":"getting-started/examples/#mcp-tool-examples","title":"MCP Tool Examples","text":""},{"location":"getting-started/examples/#using-search_pubmed","title":"Using search_pubmed","text":"
Search PubMed for \"CRISPR gene editing cancer therapy\"\n
"},{"location":"getting-started/examples/#using-search_clinical_trials","title":"Using search_clinical_trials","text":"
Find active clinical trials for \"diabetes type 2 treatment\"\n
"},{"location":"getting-started/examples/#using-search_all","title":"Using search_all","text":"
Search all sources for \"COVID-19 vaccine side effects\"\n
"},{"location":"getting-started/examples/#using-analyze_hypothesis","title":"Using analyze_hypothesis","text":"
Analyze whether vitamin D supplementation reduces COVID-19 severity\n
"},{"location":"getting-started/examples/#code-examples","title":"Code Examples","text":""},{"location":"getting-started/examples/#python-api-usage","title":"Python API Usage","text":"
from src.orchestrator_factory import create_orchestrator\nfrom src.tools.search_handler import SearchHandler\nfrom src.agent_factory.judges import create_judge_handler\n\n# Create orchestrator\nsearch_handler = SearchHandler()\njudge_handler = create_judge_handler()\n
# Run research query\nquery = \"What are the latest treatments for Alzheimer's disease?\"\nasync for event in orchestrator.run(query):\n    print(f\"Event: {event.type} - {event.data}\")\n
"},{"location":"getting-started/examples/#gradio-ui-integration","title":"Gradio UI Integration","text":"
import gradio as gr\nfrom src.app import create_research_interface\n\n# Create interface\ninterface = create_research_interface()\n\n# Launch\ninterface.launch(server_name=\"0.0.0.0\", server_port=7860)\n
"},{"location":"getting-started/examples/#research-patterns","title":"Research Patterns","text":""},{"location":"getting-started/examples/#iterative-research","title":"Iterative Research","text":"

Single-loop research with search-judge-synthesize cycles:

from src.orchestrator.research_flow import IterativeResearchFlow\n
async for event in flow.run(query):\n    # Handle events\n    pass\n
"},{"location":"getting-started/examples/#deep-research","title":"Deep Research","text":"

Multi-section parallel research:

from src.orchestrator.research_flow import DeepResearchFlow\n
async for event in flow.run(query):\n    # Handle events\n    pass\n
"},{"location":"getting-started/examples/#configuration-examples","title":"Configuration Examples","text":""},{"location":"getting-started/examples/#basic-configuration","title":"Basic Configuration","text":"
# .env file\nLLM_PROVIDER=openai\nOPENAI_API_KEY=your_key_here\nMAX_ITERATIONS=10\n
"},{"location":"getting-started/examples/#advanced-configuration","title":"Advanced Configuration","text":"
# .env file\nLLM_PROVIDER=anthropic\nANTHROPIC_API_KEY=your_key_here\nEMBEDDING_PROVIDER=local\nWEB_SEARCH_PROVIDER=duckduckgo\nMAX_ITERATIONS=20\nDEFAULT_TOKEN_LIMIT=200000\nUSE_GRAPH_EXECUTION=true\n
"},{"location":"getting-started/examples/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/installation/","title":"Installation","text":"

This guide will help you install and set up DeepCritical on your system.

"},{"location":"getting-started/installation/#prerequisites","title":"Prerequisites","text":""},{"location":"getting-started/installation/#installation-steps","title":"Installation Steps","text":""},{"location":"getting-started/installation/#1-install-uv-recommended","title":"1. Install uv (Recommended)","text":"

uv is a fast Python package installer and resolver. Install it using the standalone installer (recommended):

Unix/macOS/Linux: 

curl -LsSf https://astral.sh/uv/install.sh | sh\n

Windows (PowerShell): 

powershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n

Alternative methods: 

# Using pipx (recommended if you have pipx installed)\npipx install uv\n\n# Or using pip\npip install uv\n

After installation, restart your terminal or add ~/.cargo/bin to your PATH.

"},{"location":"getting-started/installation/#2-clone-the-repository","title":"2. Clone the Repository","text":"
git clone https://github.com/DeepCritical/GradioDemo.git\ncd GradioDemo\n
"},{"location":"getting-started/installation/#3-install-dependencies","title":"3. Install Dependencies","text":"

Using uv (recommended):

uv sync\n

Using pip:

pip install -e .\n
"},{"location":"getting-started/installation/#4-install-optional-dependencies","title":"4. Install Optional Dependencies","text":"

For embeddings support (local sentence-transformers):

uv sync --extra embeddings\n

For Modal sandbox execution:

uv sync --extra modal\n

For Magentic orchestration:

uv sync --extra magentic\n

Install all extras:

uv sync --all-extras\n
"},{"location":"getting-started/installation/#5-configure-environment-variables","title":"5. Configure Environment Variables","text":"

Create a .env file in the project root:

# Required: At least one LLM provider\nLLM_PROVIDER=openai  # or \"anthropic\" or \"huggingface\"\nOPENAI_API_KEY=your_openai_api_key_here\n\n# Optional: Other services\nNCBI_API_KEY=your_ncbi_api_key_here  # For higher PubMed rate limits\nMODAL_TOKEN_ID=your_modal_token_id\nMODAL_TOKEN_SECRET=your_modal_token_secret\n

See the Configuration Guide for all available options.

"},{"location":"getting-started/installation/#6-verify-installation","title":"6. Verify Installation","text":"

Run the application:

uv run gradio run src/app.py\n

Open your browser to http://localhost:7860 to verify the installation.

"},{"location":"getting-started/installation/#development-setup","title":"Development Setup","text":"

For development, install dev dependencies:

uv sync --all-extras --dev\n

Install pre-commit hooks:

uv run pre-commit install\n
"},{"location":"getting-started/installation/#troubleshooting","title":"Troubleshooting","text":""},{"location":"getting-started/installation/#common-issues","title":"Common Issues","text":"

Import Errors: - Ensure you've installed all required dependencies - Check that Python 3.11+ is being used

API Key Errors: - Verify your .env file is in the project root - Check that API keys are correctly formatted - Ensure at least one LLM provider is configured

Module Not Found: - Run uv sync or pip install -e . again - Check that you're in the correct virtual environment

Port Already in Use: - Change the port in src/app.py or use environment variable - Kill the process using port 7860

"},{"location":"getting-started/installation/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/mcp-integration/","title":"MCP Integration","text":"

The DETERMINATOR exposes a Model Context Protocol (MCP) server, allowing you to use its search tools directly from Claude Desktop or other MCP clients.

"},{"location":"getting-started/mcp-integration/#what-is-mcp","title":"What is MCP?","text":"

The Model Context Protocol (MCP) is a standard for connecting AI assistants to external tools and data sources. The DETERMINATOR implements an MCP server that exposes its search capabilities as MCP tools.

"},{"location":"getting-started/mcp-integration/#mcp-server-url","title":"MCP Server URL","text":"

When running locally:

http://localhost:7860/gradio_api/mcp/\n
"},{"location":"getting-started/mcp-integration/#claude-desktop-configuration","title":"Claude Desktop Configuration","text":""},{"location":"getting-started/mcp-integration/#1-locate-configuration-file","title":"1. Locate Configuration File","text":"

macOS: 

~/Library/Application Support/Claude/claude_desktop_config.json\n

Windows: 

%APPDATA%\\Claude\\claude_desktop_config.json\n

Linux: 

~/.config/Claude/claude_desktop_config.json\n

"},{"location":"getting-started/mcp-integration/#2-add-the-determinator-server","title":"2. Add The DETERMINATOR Server","text":"

Edit claude_desktop_config.json and add:

{\n  \"mcpServers\": {\n    \"determinator\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#3-restart-claude-desktop","title":"3. Restart Claude Desktop","text":"

Close and restart Claude Desktop for changes to take effect.

"},{"location":"getting-started/mcp-integration/#4-verify-connection","title":"4. Verify Connection","text":"

In Claude Desktop, you should see The DETERMINATOR tools available: - search_pubmed - search_clinical_trials - search_biorxiv - search_all - analyze_hypothesis

"},{"location":"getting-started/mcp-integration/#available-tools","title":"Available Tools","text":""},{"location":"getting-started/mcp-integration/#search_pubmed","title":"search_pubmed","text":"

Search peer-reviewed biomedical literature from PubMed.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search PubMed for \"metformin diabetes\"\n

"},{"location":"getting-started/mcp-integration/#search_clinical_trials","title":"search_clinical_trials","text":"

Search ClinicalTrials.gov for interventional studies.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search clinical trials for \"Alzheimer's disease treatment\"\n

"},{"location":"getting-started/mcp-integration/#search_biorxiv","title":"search_biorxiv","text":"

Search bioRxiv/medRxiv preprints via Europe PMC.

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results (default: 10)

Example: 

Search bioRxiv for \"CRISPR gene editing\"\n

"},{"location":"getting-started/mcp-integration/#search_all","title":"search_all","text":"

Search all sources simultaneously (PubMed, ClinicalTrials.gov, Europe PMC).

Parameters: - query (string): Search query - max_results (integer, optional): Maximum number of results per source (default: 10)

Example: 

Search all sources for \"COVID-19 vaccine efficacy\"\n

"},{"location":"getting-started/mcp-integration/#analyze_hypothesis","title":"analyze_hypothesis","text":"

Perform secure statistical analysis using Modal sandboxes.

Parameters: - hypothesis (string): Hypothesis to analyze - data (string, optional): Data description or code

Example: 

Analyze the hypothesis that metformin reduces cancer risk\n

"},{"location":"getting-started/mcp-integration/#using-tools-in-claude-desktop","title":"Using Tools in Claude Desktop","text":"

Once configured, you can ask Claude to use DeepCritical tools:

Use DeepCritical to search PubMed for recent papers on Alzheimer's disease treatments.\n

Claude will automatically: 1. Call the appropriate DeepCritical tool 2. Retrieve results 3. Use the results in its response

"},{"location":"getting-started/mcp-integration/#troubleshooting","title":"Troubleshooting","text":""},{"location":"getting-started/mcp-integration/#connection-issues","title":"Connection Issues","text":"

Server Not Found: - Ensure DeepCritical is running (uv run gradio run src/app.py) - Verify the URL in claude_desktop_config.json is correct - Check that port 7860 is not blocked by firewall

Tools Not Appearing: - Restart Claude Desktop after configuration changes - Check Claude Desktop logs for errors - Verify MCP server is accessible at the configured URL

"},{"location":"getting-started/mcp-integration/#authentication","title":"Authentication","text":"

If DeepCritical requires authentication: - Configure API keys in DeepCritical settings - Use HuggingFace OAuth login - Ensure API keys are valid

"},{"location":"getting-started/mcp-integration/#advanced-configuration","title":"Advanced Configuration","text":""},{"location":"getting-started/mcp-integration/#custom-port","title":"Custom Port","text":"

If running on a different port, update the URL:

{\n  \"mcpServers\": {\n    \"deepcritical\": {\n      \"url\": \"http://localhost:8080/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#multiple-instances","title":"Multiple Instances","text":"

You can configure multiple DeepCritical instances:

{\n  \"mcpServers\": {\n    \"deepcritical-local\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    },\n    \"deepcritical-remote\": {\n      \"url\": \"https://your-server.com/gradio_api/mcp/\"\n    }\n  }\n}\n
"},{"location":"getting-started/mcp-integration/#next-steps","title":"Next Steps","text":""},{"location":"getting-started/quick-start/","title":"Single Command Deploy","text":"

Deploy with docker instandly with a single command :

docker run -it -p 7860:7860 --platform=linux/amd64 \\\n    -e DB_KEY=\"YOUR_VALUE_HERE\" \\\n    -e SERP_API=\"YOUR_VALUE_HERE\" \\\n    -e INFERENCE_API=\"YOUR_VALUE_HERE\" \\\n    -e MODAL_TOKEN_ID=\"YOUR_VALUE_HERE\" \\\n    -e MODAL_TOKEN_SECRET=\"YOUR_VALUE_HERE\" \\\n    -e NCBI_API_KEY=\"YOUR_VALUE_HERE\" \\\n    -e SERPER_API_KEY=\"YOUR_VALUE_HERE\" \\\n    -e CHROMA_DB_PATH=\"./chroma_db\" \\\n    -e CHROMA_DB_HOST=\"localhost\" \\\n    -e CHROMA_DB_PORT=\"8000\" \\\n    -e RAG_COLLECTION_NAME=\"deepcritical_evidence\" \\\n    -e RAG_SIMILARITY_TOP_K=\"5\" \\\n    -e RAG_AUTO_INGEST=\"true\" \\\n    -e USE_GRAPH_EXECUTION=\"false\" \\\n    -e DEFAULT_TOKEN_LIMIT=\"100000\" \\\n    -e DEFAULT_TIME_LIMIT_MINUTES=\"10\" \\\n    -e DEFAULT_ITERATIONS_LIMIT=\"10\" \\\n    -e WEB_SEARCH_PROVIDER=\"duckduckgo\" \\\n    -e MAX_ITERATIONS=\"10\" \\\n    -e SEARCH_TIMEOUT=\"30\" \\\n    -e LOG_LEVEL=\"DEBUG\" \\\n    -e EMBEDDING_PROVIDER=\"local\" \\\n    -e OPENAI_EMBEDDING_MODEL=\"text-embedding-3-small\" \\\n    -e LOCAL_EMBEDDING_MODEL=\"BAAI/bge-small-en-v1.5\" \\\n    -e HUGGINGFACE_EMBEDDING_MODEL=\"sentence-transformers/all-MiniLM-L6-v2\" \\\n    -e HF_FALLBACK_MODELS=\"Qwen/Qwen3-Next-80B-A3B-Thinking,Qwen/Qwen3-Next-80B-A3B-Instruct,meta-llama/Llama-3.3-70B-Instruct,meta-llama/Llama-3.1-8B-Instruct,HuggingFaceH4/zephyr-7b-beta,Qwen/Qwen2-7B-Instruct\" \\\n    -e HUGGINGFACE_MODEL=\"Qwen/Qwen3-Next-80B-A3B-Thinking\" \\\n    registry.hf.space/dataquests-deepcritical:latest python src/app.py\n   ```\n\n## Quick start guide\n\nGet up and running with The DETERMINATOR in minutes.\n\n## Start the Application\n\n```bash\ngradio src/app.py\n

Open your browser to http://localhost:7860.

"},{"location":"getting-started/quick-start/#first-research-query","title":"First Research Query","text":"
  1. Enter a Research Question

Type your research question in the chat interface, for example: - \"What are the latest treatments for Alzheimer's disease?\" - \"Review the evidence for metformin in cancer prevention\" - \"What clinical trials are investigating COVID-19 vaccines?\"

  1. Submit the Query

Click \"Submit\" or press Enter. The system will: - Generate observations about your query - Identify knowledge gaps - Search multiple sources (PubMed, ClinicalTrials.gov, Europe PMC) - Evaluate evidence quality - Synthesize findings into a report

  1. Review Results

Watch the real-time progress in the chat interface: - Search operations and results - Evidence evaluation - Report generation - Final research report with citations

"},{"location":"getting-started/quick-start/#authentication","title":"Authentication","text":""},{"location":"getting-started/quick-start/#huggingface-oauth-recommended","title":"HuggingFace OAuth (Recommended)","text":"
  1. Click \"Sign in with HuggingFace\" at the top of the app
  2. Authorize the application
  3. Your HuggingFace API token will be automatically used
  4. No need to manually enter API keys
"},{"location":"getting-started/quick-start/#manual-api-key","title":"Manual API Key","text":"
  1. Open the Settings accordion
  2. Enter your API key:
  3. OpenAI API key
  4. Anthropic API key
  5. HuggingFace API key
  6. Click \"Save Settings\"
  7. Manual keys take priority over OAuth tokens
"},{"location":"getting-started/quick-start/#understanding-the-interface","title":"Understanding the Interface","text":""},{"location":"getting-started/quick-start/#chat-interface","title":"Chat Interface","text":""},{"location":"getting-started/quick-start/#status-indicators","title":"Status Indicators","text":""},{"location":"getting-started/quick-start/#settings","title":"Settings","text":""},{"location":"getting-started/quick-start/#example-queries","title":"Example Queries","text":""},{"location":"getting-started/quick-start/#simple-query","title":"Simple Query","text":"
What are the side effects of metformin?\n
"},{"location":"getting-started/quick-start/#complex-query","title":"Complex Query","text":"
Review the evidence for using metformin as an anti-aging intervention, \nincluding clinical trials, mechanisms of action, and safety profile.\n
"},{"location":"getting-started/quick-start/#clinical-trial-query","title":"Clinical Trial Query","text":"
What are the active clinical trials investigating Alzheimer's disease treatments?\n
"},{"location":"getting-started/quick-start/#next-steps","title":"Next Steps","text":""},{"location":"overview/architecture/","title":"Architecture Overview","text":"

The DETERMINATOR is a powerful generalist deep research agent system that uses iterative search-and-judge loops to comprehensively investigate any research question. It stops at nothing until finding precise answers, only stopping at configured limits (budget, time, iterations). The system automatically determines if medical knowledge sources are needed and adapts its search strategy accordingly. It supports multiple orchestration patterns, graph-based execution, parallel research workflows, and long-running task management with real-time streaming.

"},{"location":"overview/architecture/#core-architecture","title":"Core Architecture","text":""},{"location":"overview/architecture/#orchestration-patterns","title":"Orchestration Patterns","text":"
  1. Graph Orchestrator (src/orchestrator/graph_orchestrator.py):
  2. Graph-based execution using Pydantic AI agents as nodes
  3. Supports both iterative and deep research patterns
  4. Node types: Agent, State, Decision, Parallel
  5. Edge types: Sequential, Conditional, Parallel
  6. Conditional routing based on knowledge gaps, budget, and iterations
  7. Parallel execution for concurrent research loops
  8. Event streaming via AsyncGenerator[AgentEvent] for real-time UI updates

  9. Fallback to agent chains when graph execution is disabled

  10. Deep Research Flow (src/orchestrator/research_flow.py):

  11. Pattern: Planner \u2192 Parallel Iterative Loops (one per section) \u2192 Synthesis
  12. Uses PlannerAgent to break query into report sections
  13. Runs IterativeResearchFlow instances in parallel per section via WorkflowManager
  14. Synthesizes results using LongWriterAgent or ProofreaderAgent
  15. Supports both graph execution (use_graph=True) and agent chains (use_graph=False)
  16. Budget tracking per section and globally

  17. State synchronization across parallel loops

  18. Iterative Research Flow (src/orchestrator/research_flow.py):

  19. Pattern: Generate observations \u2192 Evaluate gaps \u2192 Select tools \u2192 Execute \u2192 Judge \u2192 Continue/Complete
  20. Uses KnowledgeGapAgent, ToolSelectorAgent, ThinkingAgent, WriterAgent
  21. JudgeHandler assesses evidence sufficiency
  22. Iterates until research complete or constraints met (iterations, time, tokens)

  23. Supports graph execution and agent chains

  24. Magentic Orchestrator (src/orchestrator_magentic.py):

  25. Multi-agent coordination using agent-framework-core
  26. ChatAgent pattern with internal LLMs per agent
  27. Uses MagenticBuilder with participants: searcher, hypothesizer, judge, reporter
  28. Manager orchestrates agents via OpenAIChatClient
  29. Requires OpenAI API key (function calling support)
  30. Event-driven: converts Magentic events to AgentEvent for UI streaming

  31. Supports long-running workflows with max rounds and stall/reset handling

  32. Hierarchical Orchestrator (src/orchestrator_hierarchical.py):

  33. Uses SubIterationMiddleware with ResearchTeam and LLMSubIterationJudge
  34. Adapts Magentic ChatAgent to SubIterationTeam protocol
  35. Event-driven via asyncio.Queue for coordination

  36. Supports sub-iteration patterns for complex research tasks

  37. Legacy Simple Mode (src/legacy_orchestrator.py):

  38. Linear search-judge-synthesize loop
  39. Uses SearchHandlerProtocol and JudgeHandlerProtocol
  40. Generator-based design yielding AgentEvent objects
  41. Backward compatibility for simple use cases
"},{"location":"overview/architecture/#long-running-task-support","title":"Long-Running Task Support","text":"

The system is designed for long-running research tasks with comprehensive state management and streaming:

  1. Event Streaming:
  2. All orchestrators yield AgentEvent objects via AsyncGenerator
  3. Real-time UI updates through Gradio chat interface
  4. Event types: started, searching, search_complete, judging, judge_complete, looping, synthesizing, hypothesizing, complete, error

  5. Metadata includes iteration numbers, tool names, result counts, durations

  6. Budget Tracking (src/middleware/budget_tracker.py):

  7. Per-loop and global budget management
  8. Tracks: tokens, time (seconds), iterations
  9. Budget enforcement at decision nodes
  10. Token estimation (~4 chars per token)
  11. Early termination when budgets exceeded

  12. Budget summaries for monitoring

  13. Workflow Manager (src/middleware/workflow_manager.py):

  14. Coordinates parallel research loops
  15. Tracks loop status: pending, running, completed, failed, cancelled
  16. Synchronizes evidence between loops and global state
  17. Handles errors per loop (doesn't fail all if one fails)
  18. Supports loop cancellation and timeout handling

  19. Evidence deduplication across parallel loops

  20. State Management (src/middleware/state_machine.py):

  21. Thread-safe isolation using ContextVar for concurrent requests
  22. WorkflowState tracks: evidence, conversation history, embedding service
  23. Evidence deduplication by URL
  24. Semantic search via embedding service
  25. State persistence across long-running workflows

  26. Supports both iterative and deep research patterns

  27. Gradio UI (src/app.py):

  28. Real-time streaming of research progress
  29. Accordion-based UI for pending/done operations
  30. OAuth integration (HuggingFace)
  31. Multiple backend support (API keys, free tier)
  32. Handles long-running tasks with progress indicators
  33. Event accumulation for pending operations
"},{"location":"overview/architecture/#graph-architecture","title":"Graph Architecture","text":"

The graph orchestrator (src/orchestrator/graph_orchestrator.py) implements a flexible graph-based execution model:

Node Types:

Edge Types:

Graph Patterns:

Execution Flow:

  1. Graph construction from nodes and edges
  2. Graph validation (no cycles, all nodes reachable)
  3. Graph execution from entry node
  4. Node execution based on type
  5. Edge evaluation for next node(s)
  6. Parallel execution via asyncio.gather()
  7. State updates at state nodes
  8. Event streaming for UI
"},{"location":"overview/architecture/#key-components","title":"Key Components","text":""},{"location":"overview/architecture/#research-team--parallel-execution","title":"Research Team & Parallel Execution","text":"

The system supports complex research workflows through:

  1. WorkflowManager: Coordinates multiple parallel research loops
  2. Creates and tracks ResearchLoop instances
  3. Runs loops in parallel via asyncio.gather()
  4. Synchronizes evidence to global state

  5. Handles loop failures gracefully

  6. Deep Research Pattern: Breaks complex queries into sections

  7. Planner creates report outline with sections
  8. Each section runs as independent iterative research loop
  9. Loops execute in parallel
  10. Evidence shared across loops via global state

  11. Final synthesis combines all section results

  12. State Synchronization: Thread-safe evidence sharing

  13. Evidence deduplication by URL
  14. Global state accessible to all loops
  15. Semantic search across all collected evidence
  16. Conversation history tracking per iteration
"},{"location":"overview/architecture/#configuration--modes","title":"Configuration & Modes","text":"

Note: The UI provides separate controls for orchestrator mode and graph research mode. When using graph-based orchestrators (iterative/deep/auto), the graph research mode determines the specific pattern used within the graph execution.

"},{"location":"overview/features/","title":"Features","text":"

The DETERMINATOR provides a comprehensive set of features for AI-assisted research:

"},{"location":"overview/features/#core-features","title":"Core Features","text":""},{"location":"overview/features/#multi-source-search","title":"Multi-Source Search","text":""},{"location":"overview/features/#mcp-integration","title":"MCP Integration","text":""},{"location":"overview/features/#authentication","title":"Authentication","text":""},{"location":"overview/features/#secure-code-execution","title":"Secure Code Execution","text":""},{"location":"overview/features/#semantic-search--rag","title":"Semantic Search & RAG","text":""},{"location":"overview/features/#orchestration-patterns","title":"Orchestration Patterns","text":"

Orchestrator Modes: - simple: Legacy linear search-judge loop - advanced (or magentic): Multi-agent coordination (requires OpenAI API key) - iterative: Knowledge-gap-driven research with single loop - deep: Parallel section-based research with planning - auto: Intelligent mode detection based on query complexity

Graph Research Modes (used within graph orchestrator): - iterative: Single research loop pattern - deep: Multi-section parallel research pattern - auto: Auto-detect pattern based on query complexity

Execution Modes: - use_graph=True: Graph-based execution with parallel and conditional routing - use_graph=False: Agent chains with sequential execution (backward compatible)

"},{"location":"overview/features/#real-time-streaming","title":"Real-Time Streaming","text":""},{"location":"overview/features/#budget-management","title":"Budget Management","text":""},{"location":"overview/features/#state-management","title":"State Management","text":""},{"location":"overview/features/#multimodal-input--output","title":"Multimodal Input & Output","text":""},{"location":"overview/features/#advanced-features","title":"Advanced Features","text":""},{"location":"overview/features/#agent-system","title":"Agent System","text":""},{"location":"overview/features/#search-tools","title":"Search Tools","text":""},{"location":"overview/features/#error-handling","title":"Error Handling","text":""},{"location":"overview/features/#configuration","title":"Configuration","text":""},{"location":"overview/features/#testing","title":"Testing","text":""},{"location":"overview/features/#ui-features","title":"UI Features","text":""},{"location":"overview/features/#gradio-interface","title":"Gradio Interface","text":""},{"location":"overview/features/#mcp-server","title":"MCP Server","text":""},{"location":"overview/features/#development-features","title":"Development Features","text":""},{"location":"overview/features/#code-quality","title":"Code Quality","text":""},{"location":"overview/features/#documentation","title":"Documentation","text":""},{"location":"overview/quick-start/","title":"Quick Start","text":"

Get started with DeepCritical in minutes.

"},{"location":"overview/quick-start/#installation","title":"Installation","text":"
# Install uv if you haven't already (recommended: standalone installer)\n# Unix/macOS/Linux:\ncurl -LsSf https://astral.sh/uv/install.sh | sh\n\n# Windows (PowerShell):\npowershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"\n\n# Alternative: pipx install uv\n# Or: pip install uv\n\n# Sync dependencies\nuv sync\n
"},{"location":"overview/quick-start/#run-the-ui","title":"Run the UI","text":"
# Start the Gradio app\nuv run gradio run src/app.py\n

Open your browser to http://localhost:7860.

"},{"location":"overview/quick-start/#basic-usage","title":"Basic Usage","text":""},{"location":"overview/quick-start/#1-authentication-required","title":"1. Authentication (REQUIRED)","text":"

Authentication is mandatory - you must authenticate before using the application. The app will display an error message if you try to use it without authentication.

HuggingFace OAuth Login (Recommended): - Click the \"Sign in with HuggingFace\" button at the top of the app - Your HuggingFace API token will be automatically used for AI inference - No need to manually enter API keys when logged in

Manual API Key (Alternative): - Set environment variable HF_TOKEN or HUGGINGFACE_API_KEY before starting the app - The app will automatically use these tokens if OAuth login is not available - Supports HuggingFace API keys only (OpenAI/Anthropic keys are not used in the current implementation)

"},{"location":"overview/quick-start/#2-start-a-research-query","title":"2. Start a Research Query","text":"
  1. Enter your research question in the chat interface
  2. Text Input: Type your question directly
  3. Image Input: Click the \ud83d\udcf7 icon to upload images (OCR will extract text)
  4. Audio Input: Click the \ud83c\udfa4 icon to record or upload audio (STT will transcribe to text)
  5. Click \"Submit\" or press Enter
  6. Watch the real-time progress as the system:
  7. Generates observations
  8. Identifies knowledge gaps
  9. Searches multiple sources
  10. Evaluates evidence
  11. Synthesizes findings
  12. Review the final research report
  13. Audio Output: If enabled, the final response will include audio synthesis (TTS)

Multimodal Features: - Configure image/audio input and output in the sidebar settings - Image OCR and audio STT/TTS can be enabled/disabled independently - TTS voice and speed can be customized in the Audio Output settings

"},{"location":"overview/quick-start/#3-mcp-integration-optional","title":"3. MCP Integration (Optional)","text":"

Connect DeepCritical to Claude Desktop:

  1. Add to your claude_desktop_config.json: 

    {\n  \"mcpServers\": {\n    \"deepcritical\": {\n      \"url\": \"http://localhost:7860/gradio_api/mcp/\"\n    }\n  }\n}\n

  2. Restart Claude Desktop

  3. Use DeepCritical tools directly from Claude Desktop
"},{"location":"overview/quick-start/#available-tools","title":"Available Tools","text":"

Note: The application automatically uses all available search tools (Neo4j, PubMed, ClinicalTrials.gov, Europe PMC, Web search, RAG) based on query analysis. Neo4j knowledge graph search is included by default for biomedical queries.

"},{"location":"overview/quick-start/#next-steps","title":"Next Steps","text":""}]} \ No newline at end of file