Sudoku infrastructure

Sudoku puzzles, engineered for production.

A pre-generated pool of solvable, uniquely-difficult Sudoku puzzles — served over a fast edge API, exported in bulk for print, or wired straight into an AI pipeline with MCP. Built to remove one problem from your stack, not add one.

  • Guaranteed unique solution
  • Near-duplicate filtering
  • Calibrated difficulty, Easy → Expert
5difficulty tiers, calibrated
3ways in — API, bulk export, MCP
0near-duplicates in the pool
24/7served from the edge

Built for

Puzzle infrastructure for teams shipping content, not players looking to compete.

Three ways to pull from the same quality-gated puzzle pool, depending on where in your stack you need it.

Websites & apps

Pull ready-to-play puzzles on demand over a REST API — a guaranteed unique solution, calibrated difficulty, and image export, without running a generator of your own.

Read the API docs

Publishers & print

Generate thousands of unique, quality-gated puzzles offline and export straight into book-ready PDFs — no near-duplicates slipping into the same volume.

Download the generator

AI & content pipelines

An MCP server exposes generate, solve, rate and explain as tools an agent can call directly — no bespoke API wrapper to write and maintain.

See the MCP docs

How it works

Puzzles are generated once, quality-gated hard, and served many times.

Generation is CPU-heavy, so it happens in batches ahead of time — not on every request.

1

Generate

The Python engine produces puzzles in bulk, then runs a dedup + near-duplicate + QA gate before anything is kept — the problem that started this whole project.

2

Pool

Passing batches sync to Cloudflare D1, R2 and KV — metadata, files and a fast cache, ready to serve globally from the edge.

3

Serve

Pulled over the API, exported in bulk for a print run, or called through MCP by an agent — whichever fits your stack.

Why this exists

“GridHearth publishes free Sudoku online and paperback puzzle books on Amazon. We needed a way to produce genuinely unique puzzles at volume — not just re-shuffled duplicates — for our own AI-managed web content and book pipeline. This is the infrastructure we built to solve that. It runs our own books and web content today; we’re opening it up for other teams solving the same problem.”
— Built and operated by GridHearth

Ready to look at the docs?

Authentication, rate limits, every endpoint, and copy-pasteable examples in curl, Python and TypeScript.