Patzer
An open-source, self-hosted chess platform deployable on a Raspberry Pi, NAS, or older laptop. It integrates the Stockfish engine and a local LLM-based coach, and supports game review, human-versus-computer play, and multi-user management for households.
Timeline (last 90 days)
Patzer is a self-hosted, open-source alternative to Chess.com that can run on a Raspberry Pi, NAS, or old laptop.
Patzer is released under the MIT license, defaults primarily to English, and has built-in support for Bulgarian, Spanish, German, and Russian.
Patzer's game records, user profiles, and settings never leave the home LAN.
Patzer's game review feature can fetch public games from Chess.com or Lichess, or accept a PGN notation pasted directly by the user.
Patzer's built-in Stockfish engine classifies each move as brilliant, good, inaccuracy, mistake, or blunder, and provides accuracy, estimated Elo, and a position evaluation curve.
Patzer's human-vs-computer mode offers seven named difficulty levels ranging from Kid to Stockfish max, with blunder warnings added for children's mode.
Patzer's friend battle mode implements real-time same-server PvP via WebSocket, supporting draw offers, take-backs, and one-click rematches.
Patzer's AI coach can be pointed at the user's own Ollama or vLLM host to explain moves and connect them to error patterns from recent games.
In Patzer's architectural design, the LLM is never responsible for analysis; Stockfish computes the facts, and the LLM is solely responsible for translating those facts into instructional language.
By dividing responsibilities so that the engine handles truth and the model handles expression, a small local model is sufficient for the task, with no need for expensive GPUs or cloud APIs.
3 more timeline events
All Facts (13)
Patzer is a self-hosted, open-source alternative to Chess.com that can run on a Raspberry Pi, NAS, or old laptop.
50%UnverifiedPatzer is released under the MIT license, defaults primarily to English, and has built-in support for Bulgarian, Spanish, German, and Russian.
50%UnverifiedPatzer's game records, user profiles, and settings never leave the home LAN.
50%UnverifiedPatzer's game review feature can fetch public games from Chess.com or Lichess, or accept a PGN notation pasted directly by the user.
50%UnverifiedPatzer's built-in Stockfish engine classifies each move as brilliant, good, inaccuracy, mistake, or blunder, and provides accuracy, estimated Elo, and a position evaluation curve.
50%UnverifiedPatzer's human-vs-computer mode offers seven named difficulty levels ranging from Kid to Stockfish max, with blunder warnings added for children's mode.
50%UnverifiedPatzer's friend battle mode implements real-time same-server PvP via WebSocket, supporting draw offers, take-backs, and one-click rematches.
50%UnverifiedPatzer's AI coach can be pointed at the user's own Ollama or vLLM host to explain moves and connect them to error patterns from recent games.
50%UnverifiedIn Patzer's architectural design, the LLM is never responsible for analysis; Stockfish computes the facts, and the LLM is solely responsible for translating those facts into instructional language.
50%UnverifiedBy dividing responsibilities so that the engine handles truth and the model handles expression, a small local model is sufficient for the task, with no need for expensive GPUs or cloud APIs.
50%UnverifiedPatzer has a safeguard in place: if the AI's response praises a move that is clearly a blunder, that response is intercepted before the user sees it.
50%UnverifiedThe entire Patzer system is packaged in a single Docker container and can be deployed with one docker run command, with the image at ghcr.io/SikamikanikoBG/patzer:latest.
50%UnverifiedThe Patzer project repository is located at github.com/SikamikanikoBG/patzer.
50%