Chess coach with AI

You lost that one for a reason. Here it is.

Play a game, then find out what actually went wrong. The engine in your pocket does the calculating. An AI coach turns it into a sentence you can use in the next game.

Free to start. iPhone, iPad and Android. No ads.

5…Nxd5?? +0.3 → +2.4
Blunder

Taking on d5 opens the door to 6.Nxf7. Your king gets dragged out with the white queen and bishop already pointing at it.

Play 5…Na5 instead. It costs a pawn and keeps the king at home.

The engine found the move. The coach wrote the reason.
Why another chess app

Engines are free. Understanding isn't.

Any phone can run an engine stronger than any human alive. It shows you the move and a number. It never tells you what you walked into, and a number is not a lesson. CheesCoach keeps the engine and adds the part that was missing: a plain sentence about the idea you missed, written for your rating.

“We're entering the endgame. King activity matters more than material now: walk him to the centre before you push anything.”

An actual comment, from an actual game review
After the game

The game ends. That is when it starts to teach.

Finish a game and CheesCoach runs it through the engine on your phone, marks the moves that cost you something, and comments them one by one as you step through. Bring games in from Lichess or paste any PGN.

Game review showing accuracy for both players, an evaluation graph split into opening, middlegame and endgame, and the engine's best move on the board

Accuracy, mistakes and where the game turned

The coach commenting a single move, with the evaluation beside it

A comment on every move you step through

Saved games list with filters for wins, losses, draws and Lichess imports

Your games, plus everything you import

Practice Lab

A game has three parts. So does the training.

Not a pile of exercises. Three labs, one for each phase of the game, each with the same shape: learn the idea first, then play it out against an opponent that will punish you for getting it wrong.

The three labs are Premium. The Italian Game in the opening trainer, and the first middlegame exercise, are free.

The Practice Lab screen with three cards: Opening Lab, Endgame Lab and Middlegame Lab

I — The opening

Your repertoire, until it sticks

Pick the openings you actually play. The trainer drills the moves with spaced repetition and tells you when a line is due for review. Then you play the same line out against Maia, a network trained on real club games, so it goes wrong the way humans go wrong.

Opening Trainer

The moves of your repertoire, drilled until you stop thinking about them.

Spaced repetition, three stars per line

Practise

Play the line for real against a rival that moves like a club player.

Maia networks by strength
Opening Trainer with a repertoire of openings for White and for Black, each with a star rating and a due-for-review marker

Your repertoire, by colour

A training position in the Giuoco Piano, with a green arrow showing the move to play and the candidate moves listed below

Play the move, not the theory

II — The middlegame

The plan when the book runs out Coming soon

Move fifteen, the theory is over and nobody taught you what to do next. The Middlegame Lab goes by pawn structure: the Carlsbad, the Benoni, the French chain and twenty-two more, each with its theory, the plan for both sides, and positions from real games where you name the structure and then carry the plan out.

Learn the structures

Each structure with its theory, the plans of both sides, and a question to check you read it right.

25 structures

Practise

Name the structure, say where the game is going, then play the plan out with a move budget.

1000 positions

Written and playable. The app still shows it as coming soon, and that is the truth: it opens once the first exercises have been through a round of testing.

Middlegame Lab with two cards: Learn the structures, twenty-five of them, and Practise with a thousand positions

Learn it, then play it

A grid of pawn structures: the Benko, the asymmetric Benoni, the Carlsbad, the Caro-Kann formation and the French chain

Benko, Benoni, Carlsbad, French chain

III — The endgame

The hundred endgames you have to know

Position by position: the goal, the technique in four lines, and then you play it against an engine at full strength that will not slip. Some you have to win, some you only have to hold. Guided mode warns you the moment a move throws the result away.

Learn the technique

King and pawns, rooks, queens, bishops and knights, each family in order of difficulty.

100 endgames

Practise

A fresh position every time, drawn from real games. Win some, hold others.

45 000+ drills
The hundred endgames grouped into families: king and pawns, rook, queen, bishop and knight endgames, each with a progress bar

Five families, twenty-five solved

The Inside the square endgame: the goal, a rigour selector between guided and free, and the technique explained in four numbered steps

The goal, then the technique in four steps

Courses

If you'd rather be told where to start

A ladder of guided courses, 164 lessons so far, each one a mix of theory, puzzles, an opening to learn and a game against a bot you have to win to move on. The first level teaches the rules from scratch and is free.

Level 01

First Steps

How the pieces move, checkmate, what a plan is.

29 lessons, free
Level 02

Beginner

Tactics and the first openings worth knowing.

44 lessons, 0–1000
Level 03

Intermediate

Strategy, structures and the endgames that decide games.

91 lessons, 1000–1500
Level 04

Advanced

Calculation and long plans.

1500–2000, coming soon
The courses screen: a winding path with First Steps, Beginner and Intermediate completed, and Advanced still locked

One path, one lesson at a time

Play

Somebody to play, whatever mood you're in

Twenty rivals with faces and ratings, from a beginner who hangs pieces to a grandmaster who doesn't. Real people online. Puzzles when you have three minutes. A board to analyse anything you like.

Twenty rivals

Five levels, four opponents each. They blunder like the rating on their card.

HU Chess

Chess and Go online, with a clock, a rating and a monthly ranking.

Puzzles

A daily one, by theme, a streak to keep alive and a three-minute storm.

Analysis board

Set up any position, get the engine lines, ask the coach what it means.

New game setup with the rival, the colour to play and the coach mode switched on
Turn the coach on before you start
Choose your opponent: beginner and intermediate rivals with portraits and win-loss records
Pick a rival, keep the score
The exercises screen with the daily puzzle, puzzles by theme, streak and storm modes
Daily puzzle, themes, streak, storm
Puzzle streak in progress: find the best move without missing one
Streak: one miss and it's over
RoyalChess daily challenge: rebuild the target position in as few moves as possible
RoyalChess: the daily challenge
Under the hood

One engine cannot do all three jobs

Judging a position, playing like a person and fitting inside an iPhone are different problems. They run on your device, not on a server, so analysis works on the underground with no signal.

Stockfish 18

The one that judges

On Android it runs on the phone as its own process, with its neural evaluation. It is what marks your blunders, draws the evaluation graph and finds the move you missed, and it is what the strongest rivals play with.

Caissa

The one on iPhone and iPad

Not a whim. Stockfish is GPLv3 and iOS does not let an app start a separate process, so linking it inside the app would put the whole app under that licence. Caissa does the same job under a licence that lets it live in there, and it analyses and plays on the device just the same.

Maia

The one that plays like a person

Not a strong engine turned down. Each level is a different network trained on real games by players of that strength, so it goes wrong where people go wrong and plays the move a club player of that rating would actually pick. It is the rival in the Opening Lab.

One more, for honesty: below about 1300 the official Stockfish cannot be made weaker — its weakest setting still plays at roughly 1320. The beginner rivals run on Fairy-Stockfish instead, which can, so a bot labelled 800 loses the way an 800 loses rather than by throwing a piece at random.

How the coach is powered

Bring your own key, or don't think about it

The engine runs on your phone. The explanations come from a language model, and you choose how you pay for that: credits inside the app, or your own API key with one of five providers, billed by them at cost.

Credits

Buy a pack and forget about it. Free calls to start with, so you can see what the coach sounds like before paying anything.

Your key

Paste an API key and the app talks to that provider directly. Typical cost of a full game review is a fraction of a cent.

Privacy

Keys are encrypted on the device and never leave it except to reach the provider you chose. No ads, no trackers, no selling anybody's games.

How to set up a key

The engine is yours, and it's open

Stockfish, Fairy-Stockfish and the Maia networks are GPL and run on the device. The sources and the exact builds are published.

Read the open source notice

Your next game

Take the coach with you

Start free on iPhone, iPad or Android. No ads, and no subscription needed to play, analyse or take the first course.

Sign in and your games, ratings and progress follow you to the other device.

The CheesCoach home screen on an iPhone: puzzle, streak and storm ratings, and the cards for playing, learning and the Practice Lab

The home screen, on an iPhone. Same app on Android and iPad.