Conduit

Self-hosted RAG system that exposes an MCP interface for CLI agents (Claude Code, Codex, Gemini). Ingests code files, documentation, and notes into a vector database, then serves relevant context on demand.

Features

• Document ingestion — upload files via API or admin UI, automatic chunking and embedding
• Semantic search — query your knowledge base with natural language
• MCP interface — plug into any MCP-compatible CLI agent
• Admin UI — Streamlit panel for uploading documents and browsing the collection

Tech Stack

• Python 3.12 / uv
• FastAPI — REST API for ingestion and search
• Streamlit — admin panel
• Qdrant — vector database
• BAAI/bge-m3 — embedding model (self-hosted)

Quick Start

# install dependencies
uv sync

# configure environment
cp .env.example .env
# edit .env with your Qdrant and embedding service addresses

# run the API server
uv run uvicorn conduit.api.app:app --reload

# run the admin UI
uv run streamlit run ui/app.py

Development

uv run pytest                         # run tests
uv run ruff check src/ tests/ ui/     # lint
uv run ruff format src/ tests/ ui/    # format
uv run mypy src/                      # type check

Project Structure

src/conduit/
├── api/          # FastAPI app, routes, dependencies
├── core/         # Domain logic (chunking, embeddings, retrieval)
├── db/           # Qdrant client wrapper
├── mcp/          # MCP interface (TBD)
├── models/       # Pydantic schemas
└── config.py     # Configuration (pydantic-settings)
ui/               # Streamlit admin panel
tests/            # pytest tests

License