Chess Coach

A self-hosted, single-user web app that analyzes your Chess.com games with a local Stockfish chess engine and an OpenAI (GPT-4.1) coach. Engine analysis runs locally and free; the coach chat uses the OpenAI API.

Features

• Sign-in with just a Chess.com username — no password or sign-up step.
• Game sync: pulls games directly from Chess.com and runs full Stockfish engine analysis on every move.
• Game dashboard: all synced games with opponent, result, color, date, and per-game blunder/mistake/inaccuracy counts.
• Game detail view: move-by-move breakdown with eval before/after, centipawn loss, move classification, and the engine's suggested best move, alongside a replayable board.
• AI coach analysis: written coaching commentary per game.
• Blunder puzzles: a cross-game queue of your own flagged blunders/mistakes — the position is shown right before the mistake and you have to find the fix yourself before the answer is revealed, prioritized toward rapid/daily games over bullet/blitz time-pressure scrambles.
• Coach chat: conversational AI coach scoped to a single game or a global chat across your whole history, with saved conversations, streamed responses, and a "deep analysis" mode that aggregates real per-move stats across recent games.
• Insights/stats: accuracy trend, win rate by color, win rate by opening, and blunder rate by game phase (opening/middlegame/ endgame), filterable by time range.
• Repertoire diagnostic: win rate and opening-phase accuracy broken down per opening AND per color you played it as (not just opening name alone), sorted worst-performing first, so you know exactly which specific branch of your repertoire to review.
• Learned lessons: after each lost game's analysis, the coach distills one reusable rule from it and remembers it — every future coach chat (per-game or global) gets your accumulated lessons injected as context, so a repeated mistake actually gets flagged as familiar instead of treated as new each time.
• Sparring: live practice games against Maia (a neural-net engine trained on real human games at each rating level, so it plays and blunders like an actual human opponent, not a dumbed-down-but-still-perfect Stockfish). Sessions are steered into your worst-performing opening/color combo from the repertoire diagnostic, and each of your moves is checked locally in real time — a flagged blunder can be taken back and retried. Finished games are analyzed and saved exactly like a synced game.