Guide
How the MAZChess AI Coach Works
Post-game pipeline: server Stockfish analysis, move tags, LLM summary in plain English, and personalized puzzles from your blunders.
Updated 2026-06-15 · Part of improvement series
After your game ends
MAZChess queues server-side Stockfish analysis on finished games. Each move gets an evaluation, centipawn loss, and quality tag (blunder, mistake, inaccuracy, best). Results appear on your game review page with an eval graph and mistake navigation.
The coach layer
Premium users get an AI-generated summary stored in your game analysis: a short overview plus per-player strengths, key moments with lessons, and practical tips. The model reads structured eval data — not raw PGN hallucination — so feedback ties to real mistakes.
From review to training
Insights roll up themes across games. Premium puzzles can be built from positions you actually blundered — closing the loop between review and daily training.
Frequently asked questions
- Which LLM does MAZChess use?
- The platform supports OpenRouter and OpenAI backends configured server-side. Summaries are cached per game so repeat views are instant.
- Can I regenerate the coach summary?
- Premium users can refresh the coach summary from the review page to get a new generation.