sensei-ai 0.1.2

Intelligent documentation agent for AI coding assistants

Sensei

Intelligent documentation agent for AI coding assistants

Sensei solves the context pollution problem by being a specialized agent that orchestrates multiple knowledge sources and returns curated, accurate documentation with code examples.

Features

• Multi-Source Documentation: Searches Context7, nia.ai, deepcon.ai, Sourcegraph, and Tavily
• Intelligent Synthesis: Uses Claude Sonnet via DSPy to synthesize information from multiple sources
• FastAPI REST API: Query Sensei via HTTP endpoints
• MCP Server: Expose Sensei as an MCP tool for AI coding agents
• Feedback Loop: Rate responses to improve future results
• Self-Hostable: Run locally with SQLite or deploy with PostgreSQL

Quick Start

Installation

1. Clone the repository

git clone https://github.com/yourusername/sensei.git
cd sensei

2. Install dependencies

# Make sure you have Python 3.13 installed
uv sync

3. Configure environment

cp .env.example .env
# Edit .env and add your API keys

4. Run Sensei

# Run both API and MCP servers (default)
uv run sensei

# Or run just the API server
uv run sensei api

# Or run just the MCP server
uv run sensei mcp

Configuration

Sensei requires several API keys to function. Copy .env.example to .env and configure:

# Required: Claude API for the reasoning engine
ANTHROPIC_API_KEY=sk-ant-...

# Documentation sources (at least one recommended)
CONTEXT7_API_KEY=...
NIA_API_KEY=...
DEEPCON_API_KEY=...
TAVILY_API_KEY=...

# Database (default: SQLite)
DATABASE_URL=sqlite+aiosqlite:///./sensei.db

# API Server
API_HOST=0.0.0.0
API_PORT=8000

Usage

REST API

Once the API server is running, you can query Sensei via HTTP:

# Query for documentation
curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"query": "How do I authenticate with OAuth in FastAPI?"}'

# Health check
curl http://localhost:8000/health

# Rate a response
curl -X POST http://localhost:8000/rate \
  -H "Content-Type: application/json" \
  -d '{"query_id": "abc-123", "rating": 5, "feedback": "Very helpful!"}'

MCP Server

When running in MCP mode, Sensei exposes two tools:

• sensei_query: Query for documentation and code examples
• sensei_rate: Rate a previous response

AI coding agents can use these tools via the Model Context Protocol.

Docker Deployment

Using Docker

# Build the image
docker build -t sensei .

# Run API server only
docker run -p 8000:8000 --env-file .env sensei api

# Run with custom environment variables
docker run -p 8000:8000 \
  -e ANTHROPIC_API_KEY=sk-ant-... \
  -e DATABASE_URL=sqlite+aiosqlite:///./sensei.db \
  sensei api

Using Docker Compose

# Start both API and MCP servers
docker-compose up -d

# View logs
docker-compose logs -f

# Stop services
docker-compose down

Development

Running Tests

# Run all tests
uv run pytest

# Run with coverage
uv run pytest --cov=sensei --cov-report=html

# Run specific test file
uv run pytest tests/test_api.py -v

Project Structure

sensei/
\x00\x00 sensei/              # Main package
   \x00\x00 agent/          # DSPy agent and tools
   \x00\x00 api/            # FastAPI server
   \x00\x00 database/       # SQLAlchemy models and storage
   \x00\x00 mcp/            # MCP server implementation
   \x00\x00 config.py       # Configuration management
   \x00\x00 main.py         # Entry point
\x00\x00 tests/              # Test suite
\x00\x00 docs/               # Documentation
\x00\x00 Dockerfile          # Docker image definition
\x00\x00 docker-compose.yml  # Docker Compose configuration
\x00\x00 pyproject.toml      # Project dependencies

Architecture

Sensei uses a multi-layer architecture:

1. Agent Layer (DSPy): Orchestrates tool selection and response synthesis
2. Tool Layer: Wraps external APIs (Context7, Sourcegraph, etc.)
3. API Layer (FastAPI): Exposes HTTP endpoints
4. MCP Layer: Provides MCP protocol interface
5. Storage Layer (SQLAlchemy): Stores queries and ratings

Roadmap

• Implement actual Context7 and Tavily MCP integrations
• Add nia.ai and deepcon.ai API endpoints
• Implement streaming responses
• Add query caching
• Support additional LLM models (OpenAI, Gemini)
• Build feedback-based prompt optimization
• Define open .codemap format

License

MIT

Contributing

Contributions welcome! Please read our contributing guidelines and submit pull requests to our repository.