p playfulbit
Public API · no API key required

Sudoku solving for AI agents.

Give an AI agent a Sudoku board and get the next verified human-style move: a digit placement, a candidate elimination, or a bounded solving trace.

Try a verified hint View the request format

Built for ChatGPT, AI assistants, puzzle apps, and image-to-Sudoku workflows.

next human-style move● verified
placement related cells

One API. Two kinds of help.

Sudoku solving is more useful when an agent can explain the move instead of only returning the final answer.

placement

Fill the next cell

Receive a verified digit placement such as R5C5 = 5, together with the technique name and a human-readable explanation.

elimination

Remove a candidate

Receive a safe candidate deduction such as removing 8 from R7C5, plus highlights and links for a visual explanation.

Simple request format

Send a grid.
Get a reason.

The API accepts an 81-character grid. Use 0 for blank cells. No account or API key is required.

Check API health
POST /v1/solve/next
Content-Type: application/json

{
  "grid": "530070000600195000098000060800060003400803001700020006060000280000419005000080079"
}

{
  "status": "progress",
  "action": {
    "techniqueName": "Naked Single",
    "placements": [{ "cell": "R5C5", "value": 5 }]
  }
}
Visual explanation

Give the user a board, not just text.

Create a temporary HTML preview that preserves the exact verified action. It shows the full candidate context, technique highlights, placements, eliminations, and links without bundling the Sudoku solver.

Open a visual preview
POST /v1/previews
Content-Type: application/json

{
  "grid": "530070000600195000098000060800060003400803001700020006060000280000419005000080079"
}

{
  "previewUrl": "https://sudoku-agent-api.playfulbit.com/v1/previews/<id>",
  "expiresAt": "2026-09-22T00:00:00.000Z",
  "rendererVersion": "1.1.0"
}

Designed for the agent workflow

Let the AI handle the conversation while Sudoverse supplies deterministic, solver-backed puzzle logic.

Read the boardAccept a grid directly, or use vision to transcribe a Sudoku image and confirm ambiguous cells.
Ask for one verified moveCall /v1/solve/next to keep the user in control and make the explanation easy to follow.
Show the techniqueUse the returned action, target cells, highlights, and links to render a lightweight board overlay, or create a temporary preview with /v1/previews.
Continue when askedCall /v1/solve/trace for a bounded lesson sequence or guided solution.

Try a verified hint

This small live example calls the public Sudoverse Agent API from your browser.

Open visual example
The result will appear here. Try the example grid, or paste your own 81-character board.

Questions agents ask

Does the API only return deductions?

No. An action can contain verified placements (fills), candidate eliminations, or both. The response also includes the technique, explanation, candidate state, and visual coordinates.

Can I send a Sudoku image?

The API accepts a normalized 81-character grid. An AI agent with vision can read the image, confirm uncertain cells with the user, then call the API.

What does the visual preview show?

The preview is a small browser page with the current board, all legal candidates in empty cells, and the exact technique context returned by the solver. Placements, candidate removals, highlights, and links use separate visual treatments.

How long does a generated preview last?

Previews created through /v1/previews are retained for 24 hours. The response includes expiresAt; agents should present the link as temporary.

Can it solve the whole puzzle?

Use the trace endpoint for a bounded sequence of human-style steps. It returns each before-and-after grid and the action taken.

Who built Sudoverse?

Sudoverse is made by playfulbit, the same author behind sudoku.playfulbit.com and the Sudoku StepPilot app.