BM Librarian¶

BM Librarian is a sophisticated Python system for AI-powered biomedical literature research. It combines multiple specialized AI agents with robust database infrastructure to convert research questions into comprehensive, evidence-based medical reports with proper citations.

What is BM Librarian?¶

BM Librarian provides an intelligent platform for biomedical research that goes far beyond traditional literature search:

Multi-Agent AI Architecture: Specialized agents work together to query, score, cite, and synthesize biomedical literature
Comprehensive Database: Indexed PubMed abstracts, MedRxiv publications, and open access full texts with semantic search capabilities
Evidence-Based Reports: Generates publication-style reports with validated citations, preventing AI hallucination
Counterfactual Analysis: Identifies and integrates contradictory evidence for balanced conclusions

Key Features¶

Fact Checker System¶

Evaluates biomedical statements as yes/no/maybe based on literature evidence. Includes both CLI and desktop GUI with blind mode for unbiased human annotation.

PaperChecker System¶

Validates research abstracts through multi-strategy searches (semantic, HyDE, keyword) and generates counter-reports with evidence-based verdicts.

Multi-Agent Architecture¶

Agent	Role
QueryAgent	Converts natural language to database queries
DocumentScoringAgent	Rates document relevance (1-5 scale)
CitationFinderAgent	Extracts relevant passages from documents
ReportingAgent	Synthesizes citations into publication-style reports
CounterfactualAgent	Identifies contradictory evidence
EditorAgent	Creates balanced reports integrating all evidence
FactCheckerAgent	Evaluates biomedical statements with literature evidence
PaperCheckerAgent	Validates abstract claims against contradictory literature

Advanced Analytics¶

Multi-model query generation using up to 3 AI models simultaneously for 20-40% improved document retrieval
Query performance tracking showing which models find most relevant documents
Counterfactual analysis with progressive audit trails
Citation validation preventing AI fabrication

Infrastructure¶

PostgreSQL with pgvector for semantic search
SQLite-based task queue for memory-efficient processing
Automated database migrations at startup
PostgreSQL audit trail tracking complete research sessions
Local LLM integration via Ollama

Getting Started¶

Requirements¶

Python 3.12+
PostgreSQL 12+ with pgvector extension
Ollama service for local LLM

Installation¶

git clone https://github.com/hherb/bmlibrarian
cd bmlibrarian
uv sync

Usage¶

Research CLI:

uv run python bmlibrarian_cli.py

GUI Research Application:

uv run python bmlibrarian_research_gui.py

Fact-Checker CLI:

uv run python fact_checker_cli.py input.json -o results.json

Fact-Checker Review GUI:

uv run python fact_checker_review_gui.py --input-file results.db --blind

Configuration GUI:

uv run python bmlibrarian_qt.py

BM Librarian Lite & Medical Fact Checker¶

BM Librarian Lite is the lightweight companion to BM Librarian, designed for medical fact-checking without requiring local database installations or running local models.

The mobile and desktop apps are branded as Medical Fact Checker (MFC) to reflect their primary focus: answering medical questions and verifying health claims using the biomedical literature. Unlike the full BM Librarian system, MFC is not designed for comprehensive systematic literature reviews - mobile hardware constraints and the lack of large-volume local databases with powerful semantic search make this impractical on phones and tablets.

Key Differences from BM Librarian¶

Feature	BM Librarian	BM Librarian Lite / MFC
Database	PostgreSQL + pgvector	SQLite + sqlite-vec
LLM	Local via Ollama	Cloud APIs (Claude, OpenAI, etc.)
Search	Local indexed database	Direct PubMed/Europe PMC queries
Use Case	Systematic reviews, research	Quick fact-checking, Q&A
Platform	Desktop (requires setup)	Mobile apps + lightweight desktop

Cross-Platform Python App (BM Librarian Lite)¶

The desktop application runs on Windows, macOS, and Linux, built with PySide6 for a native look and feel.

Features:

Systematic literature review workflows with PubMed searching
Document scoring and citation extraction
Multi-model benchmarking to compare LLM performance
Quality assessment with evidence grading
Audit trail showing LLM reasoning
No PostgreSQL required - uses SQLite with sqlite-vec for vector search

Installation:

pip install bmlibrarian-lite

Medical Fact Checker for iOS¶

The iOS app brings medical fact-checking to iPhone and iPad with a native SwiftUI interface.

Features:

Search over 36 million PubMed articles
Dual scoring system combining LLM analysis with on-device Apple NLEmbedding
Hypothetical Document Embedding (HyDE) for improved search relevance
Multiple AI providers: Claude, OpenAI, DeepSeek, Groq, Mistral, or local Ollama
Budget controls with per-analysis and monthly spending limits
iCloud sync across devices
PDF export with full citations
Secure API key storage in Keychain

Availability: Available now on the App Store

Requirements: iOS 17.0+ or iPadOS 17.0+

Medical Fact Checker for macOS¶

The macOS version provides a native desktop experience optimized for keyboard-driven workflows.

Features:

Native macOS interface with keyboard navigation
Full-text article viewing with JATS XML rendering
Hybrid search across PubMed and Europe PMC
AppKit-based PDF generation
iCloud synchronization with iOS version
Optimized for Apple Silicon

Availability: Awaiting App Store approval

Requirements: macOS 14.0+ with Apple Silicon (M1 or later)

Medical Fact Checker for Android¶

The Android app brings the same fact-checking capabilities to Android devices with Material Design 3.

Features:

Material 3 design following Google guidelines
Medical fact-checking workflow
Room database for local persistence
Multiple search provider support (PubMed, Europe PMC)
Secure API key storage via EncryptedSharedPreferences
Support for Claude, OpenAI, DeepSeek, Groq, and Mistral

Availability: Submission to Google Play Store coming soon

Stay Updated¶

Check out our Blog for the latest development updates, release announcements, and tutorials.

Open Source¶

BM Librarian is free software released under the GNU General Public License v3.0. Visit our GitHub repository to contribute.

BM Librarian Lite is also open source - visit the bmlibrarian_lite repository for the Python desktop application and mobile app source code.