Agent Brain RAG Server

Agent Brain (formerly doc-serve) is an intelligent document indexing and semantic search system designed to give AI agents long-term memory.

AI agents need persistent memory to be truly useful. Agent Brain provides the retrieval infrastructure that enables context-aware, knowledge-grounded AI interactions.

Installation

pip install agent-brain-rag

Quick Start

1. Set environment variables:

   export OPENAI_API_KEY=your-key
   export ANTHROPIC_API_KEY=your-key
   

2. Start the server:

   agent-brain-serve
   

The server will start at http://127.0.0.1:8000.

Note: The legacy command doc-serve is still available but deprecated. Please use agent-brain-serve for new installations.

Search Capabilities

Agent Brain provides multiple search strategies to match your retrieval needs:

Features

• Document Indexing: Load and index documents from folders (PDF, Markdown, TXT, DOCX, HTML)
• AST-Aware Code Ingestion: Smart parsing for Python, TypeScript, JavaScript, Java, Go, Rust, C, C++
• Multi-Strategy Retrieval: Semantic, keyword, hybrid, and graph-based search
• OpenAI Embeddings: Uses text-embedding-3-large for high-quality embeddings
• Claude Summarization: AI-powered code summaries for better context
• Chroma Vector Store: Persistent, thread-safe vector database
• FastAPI: Modern, high-performance REST API with OpenAPI documentation

Prerequisites

• Python 3.10+
• OpenAI API key (for embeddings)
• Anthropic API key (for summarization)

GraphRAG Configuration (Feature 113)

Agent Brain supports optional GraphRAG (Graph-based Retrieval-Augmented Generation) for enhanced relationship-aware queries.

Enabling GraphRAG

Set the environment variable to enable graph indexing:

export ENABLE_GRAPH_INDEX=true

Configuration Options

Query Modes

With GraphRAG enabled, you have access to additional query modes:

• graph: Query using only the knowledge graph (entity relationships)
• multi: Combines vector search, BM25, and graph results using RRF fusion

Example: Graph Query

# CLI
agent-brain query "authentication service" --mode graph

# API
curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"query": "authentication service", "mode": "graph", "top_k": 10}'

Example: Multi-Mode Query

# CLI
agent-brain query "user login flow" --mode multi

# API
curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"query": "user login flow", "mode": "multi", "top_k": 5}'

Rebuilding the Graph Index

To rebuild only the graph index without re-indexing documents:

curl -X POST "http://localhost:8000/index?rebuild_graph=true" \
  -H "Content-Type: application/json" \
  -d '{"folder_path": "."}'

Optional Dependencies

For enhanced GraphRAG features, install optional dependency groups:

# For Kuzu graph store (production workloads)
poetry install --with graphrag-kuzu

# For enhanced entity extraction
poetry install --with graphrag

Two-Stage Reranking (Feature 123)

Agent Brain supports optional two-stage retrieval with reranking for improved search precision. When enabled, the system:

1. Stage 1: Retrieves more candidates than requested (e.g., 50 candidates for top_k=5)
2. Stage 2: Reranks candidates using a cross-encoder model for more accurate relevance scoring

Enabling Reranking

Set the following environment variables:

# Enable two-stage reranking (default: false)
ENABLE_RERANKING=true

# Choose provider (default: sentence-transformers)
RERANKER_PROVIDER=sentence-transformers  # or "ollama"

# Choose model (default: cross-encoder/ms-marco-MiniLM-L-6-v2)
RERANKER_MODEL=cross-encoder/ms-marco-MiniLM-L-6-v2

# Stage 1 retrieval multiplier (default: 10)
RERANKER_TOP_K_MULTIPLIER=10

# Maximum candidates for Stage 1 (default: 100)
RERANKER_MAX_CANDIDATES=100

# Batch size for reranking inference (default: 32)
RERANKER_BATCH_SIZE=32

Provider Options

YAML Configuration

You can also configure reranking in config.yaml:

reranker:
  provider: sentence-transformers
  model: cross-encoder/ms-marco-MiniLM-L-6-v2
  params:
    batch_size: 32

Graceful Degradation

If the reranker fails (model unavailable, timeout, etc.), the system automatically falls back to Stage 1 results. This ensures queries never fail due to reranking issues.

Response Fields

When reranking is enabled, query results include additional fields:

• rerank_score: The cross-encoder relevance score
• original_rank: The position before reranking (1-indexed)

Example response:

{
  "results": [
    {
      "text": "Document content...",
      "source": "docs/guide.md",
      "score": 0.95,
      "rerank_score": 0.95,
      "original_rank": 5,
      "chunk_id": "chunk_abc123"
    }
  ]
}

Development Installation

cd agent-brain-server
poetry install

Configuration

Copy the environment template and configure:

cp ../.env.example .env
# Edit .env with your API keys

Required environment variables:

• OPENAI_API_KEY: Your OpenAI API key for embeddings
• ANTHROPIC_API_KEY: Your Anthropic API key for summarization

Running the Server

# Development mode
poetry run uvicorn agent_brain_server.api.main:app --reload

# Or use the entry point
poetry run agent-brain-serve

API Documentation

Once running, visit:

• Swagger UI: http://127.0.0.1:8000/docs
• ReDoc: http://127.0.0.1:8000/redoc
• OpenAPI JSON: http://127.0.0.1:8000/openapi.json

API Endpoints

Health

• GET /health - Server health status
• GET /health/status - Detailed indexing status

Indexing

• POST /index - Start indexing documents from a folder
• POST /index/add - Add documents to existing index
• DELETE /index - Reset the index

Querying

• POST /query - Semantic search query
• GET /query/count - Get indexed document count

Example Usage

Index Documents

curl -X POST http://localhost:8000/index \
  -H "Content-Type: application/json" \
  -d '{"folder_path": "/path/to/docs"}'

Query Documents

curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"query": "How do I configure authentication?", "top_k": 5}'

Architecture

agent_brain_server/
├── api/
│   ├── main.py           # FastAPI application
│   └── routers/          # Endpoint handlers
├── config/
│   └── settings.py       # Configuration management
├── models/               # Pydantic request/response models
├── indexing/
│   ├── document_loader.py  # Document loading
│   ├── chunking.py         # Text chunking
│   └── embedding.py        # Embedding generation
├── services/
│   ├── indexing_service.py # Indexing orchestration
│   └── query_service.py    # Query execution
└── storage/
    └── vector_store.py     # Chroma vector store

Development

Running Tests

poetry run pytest

Code Formatting

poetry run black agent_brain_server/
poetry run ruff check agent_brain_server/

Type Checking

poetry run mypy agent_brain_server/

Documentation

• User Guide - Getting started and usage
• Developer Guide - Contributing and development
• API Reference - Full API documentation

Release Information

• Current Version: See pyproject.toml
• Release Notes: GitHub Releases
• Changelog: Latest Release

Related Packages

• agent-brain-cli - Command-line interface for Agent Brain

License

MIT