Skip to main content

TypeSafe Jev (Three Calling Methods)

This document is based on real verified calls against this relay instance; all requests and responses are actual test results. Jev is TypeSafe's "System One" judgement model: it does not do free-text chat. Given a state and a set of questions, it returns structured judgements (choice / score / noul).

Basics​

ItemValue
Relay Base URLhttps://llmapi.cttai.art
AuthAuthorization: Bearer <API Key>
API KeyGenerate one on the API Console "Tokens" page (keep it secret, never share it)
Available modelsjev-1.13.0, jev-latest, jev-preview (currently all resolve to jev-1.13.0)
BillingBilled on input_tokens only ($0.042 / 1M input tokens); output is free

Note: jev-latest points to the latest stable version (use this by default); jev-preview points to the newest build and can be ahead of stable when a preview exists; jev-1.13.0 is a pinned version.

Method 1: List available models (connectivity check)​

GET https://llmapi.cttai.art/v1/models
Authorization: Bearer <API Key>

Verified response:

{
  "data": [
    { "id": "jev-1.13.0",  "object": "model", "owned_by": "task plugin", "supported_endpoint_types": ["openai"] },
    { "id": "jev-latest",  "object": "model", "owned_by": "task plugin", "supported_endpoint_types": ["openai"] },
    { "id": "jev-preview", "object": "model", "owned_by": "task plugin", "supported_endpoint_types": ["openai"] }
  ],
  "object": "list",
  "success": true
}

Note: supported_endpoint_types: ["openai"] is just a generic marker from new-api — it does not mean chat/completions works. See "Unsupported usage" at the end.

POST https://llmapi.cttai.art/typesafe/v1/systemone
Authorization: Bearer <API Key>
Content-Type: application/json

Request body​

{
  "model": "jev-latest",
  "state": "Customer: My payouts have been failing for 3 days, I am losing sales, please help ASAP",
  "questions": {
    "department": {
      "type": "choice",
      "instructions": "Which team should handle this?",
      "criteria": {
        "billing": "Payments, invoicing, refunds",
        "technical": "Bugs, outages, integrations",
        "sales": "Pricing, upgrades, new accounts"
      }
    },
    "urgent": {
      "type": "noul",
      "instructions": "Does this express urgency?"
    },
    "frustration": {
      "type": "score",
      "instructions": "How frustrated is the customer?",
      "criteria": ["Calm", "Frustrated", "Very angry"]
    }
  }
}

Field rules:

  • model: required, one of the three model names.
  • state: shared context, can be text / object / array / null (within the 64k context budget).
  • questions: required, non-empty object; each question type must be one of:
    • choice: criteria is an object, 1~255 options.
    • score: criteria is an array, 2~10 levels.
    • noul: criteria optional; if set, the keys can only be "true"/"false".
  • stream: not supported (must be false or omitted).

Verified response​

{
  "model": "jev-1.13.0",
  "answers": {
    "department": {
      "type": "choice",
      "choice": "billing",
      "confidence": 0.91,
      "probabilities": { "billing": 0.94, "sales": 0, "technical": 0.06 }
    },
    "urgent": {
      "type": "noul",
      "noul": 0.98
    },
    "frustration": {
      "type": "score",
      "score": 1.07,
      "confidence": 0.9,
      "legend": { "0": "Calm", "1": "Frustrated", "2": "Very angry" },
      "probabilities": { "0": 0, "1": 0.93, "2": 0.07 }
    }
  },
  "usage": { "input_tokens": 425, "output_tokens": 70 }
}

Answer field mapping:

Question typeReturned fields
choicechoice (matched option key) + confidence + probabilities (per-option probability)
scorescore (level number, 0-based) + confidence + legend (level names) + probabilities
noulnoul (0~1 confidence)
  • This endpoint is synchronous — request and response return immediately; usage.input_tokens is what is billed.
  • The result is not retained for the task query API; be sure to save this response.

curl example​

curl 'https://llmapi.cttai.art/typesafe/v1/systemone' \
  -H 'Authorization: Bearer <API Key>' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "jev-latest",
    "state": "Customer: My payouts have been failing for 3 days.",
    "questions": {
      "urgent": { "type": "noul", "instructions": "Does this express urgency?" }
    }
  }'

Method 3: Generic task API (submit + poll)​

POST https://llmapi.cttai.art/v1/tasks/typesafe   # submit a task
GET  https://llmapi.cttai.art/v1/tasks/{task_id}  # query task status

The request body is the same structure as Method 2 (model required).

Submit (verified)​

curl 'https://llmapi.cttai.art/v1/tasks/typesafe' \
  -H 'Authorization: Bearer <API Key>' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "jev-latest",
    "state": "Customer: My payouts have been failing for 3 days.",
    "questions": { "urgent": { "type": "noul", "instructions": "Does this express urgency?" } }
  }'

Submit response (returns the public gateway task ID; task_id is the same as id):

{
  "created_at": 1789967908,
  "id": "task_HA2KPl4UIKSwr7ZOCcGQStelcz357SpI",
  "model": "jev-latest",
  "status": "queued",
  "task_id": "task_HA2KPl4UIKSwr7ZOCcGQStelcz357SpI"
}

Query status (verified)​

GET https://llmapi.cttai.art/v1/tasks/task_HA2KPl4UIKSwr7ZOCcGQStelcz357SpI
{
  "created_at": 1789967908,
  "fail_reason": "",
  "finished_at": 1789967908,
  "platform": "typesafe",
  "progress": "100%",
  "status": "SUCCESS",
  "task_id": "task_HA2KPl4UIKSwr7ZOCcGQStelcz357SpI"
}

Note: the generic task query only returns standard status fields — it does not return the answer body (this plugin uses a synchronous route). To get judgements, use Method 2 and save the synchronous response; task records are only for status and billing verification.

Unsupported usage​

The following usages were verified to be unavailable — do not use them:

UsageResult
POST /v1/chat/completions503 model_not_found (Jev does not chat)
stream: trueError (TypeSafe System One does not support streaming)
Polling tasks for answersThe query only returns status, not answers

Appendix: Python example (verified runnable)​

import json, urllib.request, ssl

BASE = "https://llmapi.cttai.art"
KEY = "<API Key>"
CTX = ssl.create_default_context()

def call(path, method="GET", body=None):
    data = json.dumps(body, ensure_ascii=False).encode() if body is not None else None
    req = urllib.request.Request(BASE + path, data=data, method=method,
                                 headers={"Authorization": "Bearer " + KEY,
                                          "Content-Type": "application/json"})
    with urllib.request.urlopen(req, context=CTX, timeout=30) as resp:
        return resp.status, json.loads(resp.read().decode("utf-8"))

# Method 2: native route
status, resp = call("/typesafe/v1/systemone", "POST", {
    "model": "jev-latest",
    "state": "Customer: My payouts have been failing for 3 days.",
    "questions": {"urgent": {"type": "noul", "instructions": "Does this express urgency?"}},
})
print(status, json.dumps(resp, ensure_ascii=False, indent=2))

Tip: all content in this document comes from real verified calls against the relay instance; you can verify directly with curl or any OpenAI-compatible client.