Decisions

Decisions models evaluate a shared state against typed questions and return calibrated structured answers (choice | score | noul). They are not chat completions and not embeddings. Motivating catalog example: jev-1.13-ts (TypeSafe Jev / System One). After tag detection, models tagged Decision / decisions / typesafe / systemone / system-one classify as ModelType=DECISIONS (Decision-class wins over a coexisting LLM tag).

Headers

Same as other prompt modalities — see Chat, audio, and embeddings.

Header	Required	Purpose
`Authorization: Basic <base64(user:pass)>`	Yes	Local proxy-router auth. Permission method: `decisions`.
`session_id`	Yes for remote	Open Morpheus session. Omit the header for a local `models-config` Decisions row — do not send `0` / `0x0` (hash binding fails with 400).
`model_id`	Local only	Hex model id when calling a local Decisions provider row.
`Content-Type: application/json`	Yes

Request

POST /v1/decisions Normative body (TypeSafe System One / OpenRouter Decisions core):

{
  "state": "Ticket: invoice dispute, customer asks chargeback timeline.",
  "questions": {
    "dept": {
      "type": "choice",
      "instructions": "Route the ticket.",
      "criteria": {
        "billing": "Payment, invoice, refund",
        "technical": "Product bug or outage",
        "general": "Other"
      }
    },
    "urgent": {
      "type": "noul",
      "instructions": "Needs same-day human review?"
    },
    "severity": {
      "type": "score",
      "instructions": "Business severity",
      "criteria": ["low", "medium", "high"]
    }
  }
}

Required: state, questions.
Optional model in the body is overwritten by the provider models-config.modelName (embeddings parity).
Unknown request fields (e.g. OpenRouter provider, user) are forwarded to the upstream.
Non-streaming only in v0 — single JSON response.

curl -X POST 'http://localhost:8082/v1/decisions' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Basic YWRtaW46YWRtaW4=' \
  -H 'session_id: 0x...' \
  -d '{
    "state": "Ticket: invoice dispute, customer asks chargeback timeline.",
    "questions": {
      "dept": {
        "type": "choice",
        "instructions": "Route the ticket.",
        "criteria": {
          "billing": "Payment, invoice, refund",
          "technical": "Product bug or outage",
          "general": "Other"
        }
      },
      "urgent": { "type": "noul", "instructions": "Needs same-day human review?" }
    }
  }'

Response

{
  "model": "jev-1.13.0",
  "answers": {
    "dept": { "choice": "billing", "confidence": 0.91 },
    "urgent": { "noul": false, "confidence": 0.8 },
    "severity": { "score": 0.6, "confidence": 0.7 }
  },
  "usage": { "input_tokens": 42, "output_tokens": 12 }
}

Branch on answers.<id>.choice / .score / .noul — never on chat choices[].message.content.

Provider setup (summary)

Providers register a models-config.json row with:

Field	Example
`apiType`	`decisions`
`apiUrl`	Full URL: `https://api.typesafe.ai/v1/systemone` or `https://openrouter.ai/api/alpha/decisions`
`modelName`	Upstream id: `jev-1.13.0` or `typesafe/jev-1.13`

Details: models-config.json.

Auth

Scoped agent users need decisions in their rpcwhitelist (alongside or instead of chat / embeddings). See API authentication.

Get started

Inference API (hosted)

Concepts

Consumers

Prosumers

Providers - Resale

Providers - Full P-Node

AI knowledge

Ecosystem

Reference

Headers

Request

Response

Provider setup (summary)

Auth

​Headers

​Request

​Response

​Provider setup (summary)

​Auth

Headers

Request

Response

Provider setup (summary)

Auth