About Stylish

Stylish goes beyond "is this a good move?" and tells you "what kind of move is this?"

What it does

• Style of every move: 19 style scores such as aggressive, tactical, positional, prophylactic and defensive
• Risky choices: traps, gambles, sound sacrifices and speculative sacrifices
• Human-based judgment: the human-move model Maia3 estimates how likely a human at your rating is to play a move, so traps and gambles are judged by real human standards
• Player profile: across several games you get your 4-letter chess type, similar famous players and style by opening
• Play the coach: play a bot and get feedback on every move

How it works

Stylish analyzes games you load from Chess.com or Lichess (or paste as PGN) right inside your browser. The chess engine Stockfish 19 scores every move on 19 styles, and Maia3, a neural network trained on real human moves, estimates how likely a human at the same rating is to play each move. Your 4-letter type comes from four axes built from the 8 main styles plus Maia3 originality.

The cutoffs weren't guessed. We analyzed real games from 479 Lichess players chosen across rating bands, from beginners to masters, the same way, and used the middle of that distribution. Because style changes with strength (lower-rated players play more checks and captures, for example), you are compared with players at your own rating. So each letter means "which side you lean toward among players of similar strength", and beginners and masters alike get all 16 types. For an accurate result we recommend 5 or more games, ideally 10.

Everything runs inside your browser. Your games are never sent to a Stylish server. See the privacy policy for details.

Limitations

• Trap and gamble labels rely on the human-move model (Maia3). If its predictions differ from real people, the labels can be off.
• A lightweight browser engine and limited depth are used, so moves that need very deep lines may be evaluated differently.
• Style weights were tuned on representative positions and classic games, and are still being refined.
• "Similar players" compares you with averages from 24 games each of 10 famous players, analyzed the same way (depth 10). The sample is small, so treat it as a fun reference. (Game sources: antlr/grammars-v4 examples on GitHub and CRAN bigchess examples. Only aggregate numbers are included in the site.)

Open source

Stylish is GPLv3 open source. The source code is on GitHub. Open source used: Stockfish 19 (GPLv3), stockfish.js (GPLv3), Maia3 (CSSLab, GPLv3), chessground (GPLv3), chess.js (BSD), onnxruntime-web (MIT), Lichess chess-openings (CC0).

Contact

For bug reports, suggestions or anything else, use the contact page.