Play chess against any model
Pick a model and play. The board sits beside the conversation, every move is validated before it lands, and you can ask why in the middle of the game.
How it works
Three steps, and none of them is “configure”.
- 01
Pick a side
Open yourself, or ask it to move first. It sets the game up and records which side is yours.
- 02
Play on the board
Click a piece and click a square, or drag it. Legal moves are shown as you go, and an illegal one is refused before it ever reaches the model.
- 03
Ask in the middle of the game
“Why that move?” — it answers, then plays on. Asking a question does not restart anything.
What it can do
Every move is checked
Moves are validated against the position by a real chess engine rather than trusted because a model said them. An illegal one is refused, and the model is told why.
Swap opponents mid-game
Change the model between moves and watch the play change. The position is a file, so nothing is lost when the opponent is.
The game is a file you keep
Standard PGN in your workspace. Step back through the moves on the board, or take the file somewhere else.
How it is set up
The mechanics, so you know what you are getting before you sign in.
- Environment
- A validated chess engine on the server, and an interactive board: click to move, legal-move indicators, a promotion picker, captured material and a material score.
- Tools
- A move tool that validates and applies, plus a shell and file reads for the PGN itself.
- Model
- Any chat model. This is the cheapest way to feel how differently they think — the same position, a different opponent.
- Starting files
- game.pgn, created when the session starts, with headers naming which side is yours.
- What persists
- The PGN is saved after every move. Close the tab in the middle of a game and come back to the position.
Things people ask it
- “Let’s play. You are white — make your first move.”
- “I will play white: 1. e4. Your move, and tell me what you are planning.”
- “Why did you play that instead of taking the knight?”
- “Switch to a different model and carry on from here.”
What it will not do
Every one of these is a real constraint we have hit, not a roadmap item.
- Models are not engines. Expect a strong, human-feeling opponent that blunders — not Stockfish.
- One game per session. The PGN is what makes a session a game, so another game means another session.
- No clock, no rating and no opening book.
Which model does this best
Measured over real runs of this agent, per model. A run counts as finished when it produced the answer on its own — nothing failed, nobody was asked to approve anything, and it did not run out of steps. Read all three columns together: a model that finishes fast by giving up scores badly on the first, and one that completes everything by grinding scores badly on the second.
| Model | Completion rate | Median cost | Median time | Runs |
|---|---|---|---|---|
| claude-sonnet-5anthropic | 100% | $0.010 | 13s | 9 |
- Completion rate
- Runs that finished the job without you having to step in.
- Median cost
- What a typical run costs, in credits.
- Median time
- Wall clock, from the first message to the answer.
Medians over the runs behind each row. A model appears once it has 3 runs on this agent, and the run count is shown so you can judge how much a figure rests on.