Modular RAG System

A modular, pluggable RAG (Retrieval-Augmented Generation) system with MCP (Model Context Protocol) server support.

Features

• Modular Architecture: Pluggable components for LLM, Embedding, Vector Store, Reranker, and Evaluator
• Multi-provider Support: OpenAI, Azure OpenAI, Ollama, DeepSeek, and more
• MCP Server Integration: Seamless integration with Copilot and Claude Desktop
• Hybrid Search: Combined dense and sparse retrieval with RRF fusion
• Multi-modal Support: Image understanding and caption generation
• Full Observability: Trace collection and web-based dashboard
• Zero External Dependencies: Local-first design, runs offline

Quick Start

Installation

# Create virtual environment
python -m venv .venv

# Activate virtual environment
# On Windows:
.\.venv\Scripts\activate
# On macOS/Linux:
source .venv/bin/activate

# Install dependencies
pip install -e .

# Install dev dependencies (optional)
pip install -e ".[dev]"

Configuration

1. Create local config:

cp config/settings.yaml config/settings.local.yaml

2. Set environment variables:

export AZURE_API_ENDPOINT="https://..."
export AZURE_API_KEY="..."
export OPENAI_API_KEY="..."

Running Tests

# Run smoke tests
python -m compileall src
pytest tests/unit/test_smoke_imports.py

# Run all tests
pytest

# Run with coverage
pytest --cov=src --cov-report=html

Architecture

See DEV_SPEC.md for detailed architecture documentation.

Development Status

Currently implementing Phase A: Engineering skeleton and test base

• ✅ Directory structure initialized
• ⏳ Pytest setup and smoke tests
• ⏳ Configuration loading and validation

License

MIT