OpenType documentation

OpenType answers typed yes/no, choice and score questions about a JSON state in one model call per run, with a calibrated probability for every answer.

OpenType answers typed questions about a piece of JSON in one model call per run. You send a state (a support ticket, a contract clause, a database row) and a set of questions: yes/no (noul), one of N (choice), or an ordered level (score). You get back a probability for every answer you defined, so your code compares numbers to thresholds instead of parsing prose. This page is for developers who are evaluating or integrating the API: it shows one request, the conventions every request shares, and where to go next.

One request

A decision run is one POST /v1/runs. This one asks a single yes/no question about a ticket.

cURL

curl -sS https://api.opentype.dev/v1/runs \
  -H "Authorization: Bearer $OPENTYPE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ticket-4822-urgent" \
  -d '{
    "kind": "decision",
    "instructions": "You triage customer support tickets.",
    "state": {"ticket": "I was charged twice this month and nobody answers my emails.", "plan": "pro"},
    "questions": {
      "urgent": {"type": "noul", "instructions": "reply within the hour?"}
    },
    "max_output_tokens": 16
  }'

The answer to urgent comes back under decision.answers:

{"answered_within_labels": true, "label_mass": 0.991, "probability": 0.83, "type": "noul"}

probability is the probability of “yes”. label_mass and answered_within_labels tell you how much of the model’s mass landed on your labels, so you know when a number is trustworthy. Live responses report decision.model as "neon-1.1". The quickstart walks through a full response with all three question types.

What every request has in common

Base URL	https://api.opentype.dev. Every API route is under /v1/.
Auth	Authorization: Bearer otsk_..., an API key you create in the console. See Authentication.
Idempotency	POST /v1/runs requires an Idempotency-Key header of 1 to 255 bytes. See Idempotency.
Body	JSON with Content-Type: application/json, snake_case field names, at most 4 MiB for runs and router selections, 1 MiB elsewhere. Unknown fields are refused.
Model	Neon 1.1. The optional model field on decision runs takes neon-1.1 or neon-latest. See Models and pricing.
Money	Integers in micro-USD, in fields ending _micros. 1,000,000 micros is $1.
Timestamps	UTC strings in exactly YYYY-MM-DDTHH:MM:SSZ form, in fields ending _at.
IDs	Opaque prefixed strings: run_..., key_..., req_....
Request id	Every response carries an x-request-id header. Send your own value to have it echoed back.
Errors	{"error": {"code", "message", "request_id"}}. Branch on code. See Errors.

Start here

Quickstart

Create a key, send a decision run, and read the probabilities in about five minutes.

Create an account

Sign up, verify your email, and receive $5 of free credit.

Runs

What a run is, its states, what a response contains, and how replays work.

Decision questions

The noul, choice and score types, dependencies, bounds, and answer shapes.

Console

The playground, API keys, usage, and billing in the browser.

Guides

Triage, gating, retries, polling, spend control, and production hardening.

Errors

Every error code, what triggers it, and what to do next.

API reference

Every endpoint, generated from the OpenAPI document.

Related

• Quickstart - send your first decision run end to end.
• Decision runs - a full triage example with thresholds and gated questions.
• Models and pricing - what one run costs and how the arithmetic works.
• Errors - the error envelope and every code the API returns.