Decide API

The split

Engine

A System One model first, an LLM fallback second, schema validation always. Every answer says which provider answered, the model version, whether it was calibrated, and its confidence.

Rulebook

What correct means here, and the mistakes we've already caught. It grows one job at a time.

1. R01Answers never leave the schema you asked for.
2. R02Every response says whether the provider's probabilities were calibrated.
3. R03State is treated as data, never as instructions.
4. R04Flat distributions come back flagged: your options weren't distinguishable.
5. R05Launches only after TypeSafe confirms in writing that its terms allow it. Until then: the recipe APIs, or TypeSafe directly.

Endpoint

POST /api/v1/decide  # $0.01 per call, USDC via x402

{
  "state": {
    "message": "I was charged twice for order A-104. Please refund the duplicate."
  },
  "questions": {
    "team": {
      "type": "choice",
      "instructions": "Which team should handle this",
      "criteria": {
        "billing": "Payments, refunds, duplicate charges",
        "technical": "Bugs",
        "other": "Anything else"
      }
    },
    "refund": {
      "type": "noul",
      "instructions": "The customer is explicitly asking for a refund"
    }
  }
}

Intake

• An x402-capable client paying USDC, or an API key on an invoiced plan

UNIT

One decision call: up to 16 questions over up to 8k tokens of state

REVIEW LAYER

Automated: schema errors and flat distributions are flagged in the response.

DELIVERY

JSON over HTTPS.

TURNAROUND

~100–800 ms

How we find buyers

• x402 Bazaar listing
• llms.txt and agent card on our domain
• MCP server