Schemami v1.0.0 · stable release

Recipes, structured to taste.

Turn recipes into portable data without flattening what makes them recipes: exact quantities, authored methods, choices, dependencies, evidence, and explicit refusal when facts are missing.

Use with AI Integrate an SDK

No runtime AI requiredNo global ingredient registryOffline deterministic core

Choose your path

I have a recipe

Convert it with an AI you choose.

Give a capable chat a photo, PDF, page, transcription, or text. It asks only necessary questions and returns an untrusted candidate for deterministic validation.

I build software

Parse, admit, calculate, and compare.

Use the TypeScript, Go, or Swift SDK. Exact source bytes enter; typed results or stable pointer-addressed problems come out.

6deterministic Calculus operations

3SDK implementations

1shared conformance authority

0network calls in the core

A recipe stays a recipe

Structure where structure earns its keep.

Schemami preserves source-language prose while giving quantities, formulas, method order, conditions, alternatives, components, evidence, and lineage machine-readable shapes.

Portable recipe

{
  "schemami": "1",
  "collection": "personal",
  "id": "pao-de-centeio",
  "revision": 1,
  "content_language": "pt-PT",
  "title": "Pão de centeio",
  "ingredients": [
    {
      "id": "rye-flour",
      "name": "Farinha de centeio",
      "quantity": { "kind": "measured", "value": "300", "unit": "g" }
    }
  ],
  "method": {
    "sequence": [
      {
        "kind": "step",
        "id": "mix",
        "instruction": "Misturar a farinha com os restantes ingredientes."
      }
    ]
  }
}

Deterministic boundary

source or application data
  → untrusted Schemami candidate
  → strict parse + admission
  → Calculus / Diff
  → application presentation