ragwatcher

Local RAG over a directory. Incremental. Served as MCP tool + CLI.

• Zero-config default (drops into ~/notes, works).

• Offline, CPU-only baseline (FastEmbed ONNX). GPU / API models opt-in.

• CLI is a peer to the MCP server, not a wrapper.

Install

uv tool install ragwatcher
# or
pipx install ragwatcher

Optional extras: [lance] (LanceDB store), [pdf-pro] (pymupdf), [remote] (qdrant/openai/cohere).

Related MCP server: Corpus-KB

Usage

ragwatcher serve ~/notes                  # MCP server (stdio)
ragwatcher index ~/notes                  # one-shot sync
ragwatcher query ~/notes "what is X"      # one-shot query
ragwatcher stats ~/notes                  # index summary
ragwatcher doctor ~/notes                 # health check
ragwatcher purge ~/notes --yes            # wipe .rag_index/

Every data-producing command supports --json for scripting.

Configuration

Precedence (low → high): defaults → $XDG_CONFIG_HOME/ragwatcher/config.toml → <DIR>/.ragwatcher.toml → RAGWATCHER_* env vars → CLI flags.

See SPEC.md § 5 for the full schema.

Development

uv sync --extra dev
uv run pytest -q --ignore=tests/integration
uv run ruff check src tests
uv run mypy src

Integration tests (real embed model, ~3s):

uv run pytest tests/integration -m slow

License

MIT