My kids play chess on Chess.com. It is a great product, but the account, the cloud and the ads are not something I want to hand to a nine-year-old. So I built Patzer: a self-hosted, open-source alternative that runs on a Pi, a NAS or an old laptop.
What it does:
- Game Review — pull your public Chess.com or Lichess games, or paste a PGN. Bundled Stockfish classifies every move (Brilliant / Great / Inaccuracy / Mistake / Blunder), gives accuracy, estimated Elo and an eval graph, plus a "what's the threat?" probe that answers in plain words.
- Play vs Bot — Stockfish at seven named tiers from Kid to Stockfish max, all standard time controls, with kid-mode blunder warnings.
- Play vs Friend — real-time PvP between profiles on the same server over WebSocket: draw offers, takebacks, one-click rematch.
- AI Coach — point it at your own Ollama or vLLM host. The coach explains why a move works or fails, shows the better move, and connects the mistake to your own recent games ("4 of your last 10 games: leaving a piece unprotected"). It never contradicts the engine: an answer that praises a blunder is caught before you see it.
- Family-ready — multi-user with an admin console, per-profile language, kid mode, "horsey" piece names for the youngest profiles.
The design choice I keep coming back to: the LLM never analyses. The server works out the facts — what a move allows, hangs or wins, the opponent's best answer, your recurring mistakes — and the model only turns that into coaching. So a small local model is enough, and a hallucinated piece is impossible.
One container:
docker run -d -p 8800:8800 -v patzer-data:/app/data \
--name patzer ghcr.io/SikamikanikoBG/patzer:latest
Everything runs on your LAN. Games, profiles and settings never leave home. MIT licensed, English-first with Bulgarian, Spanish, German and Russian out of the box.
Top comments (0)