typesafe.pro API

api.typesafe.pro is a simple HTTP endpoint for evaluating text and structured application state with TypeSafe System One models. Send a state and typed questions; receive structured answers with confidence and token usage.

The API is compatible with the TypeSafe evaluation endpoint. See the TypeSafe API reference for the full request and response schema.

The website also provides verified, interactive use-case guides with sandboxed visual demos. Their storage and public API, the local LLM7/Jev review session (npm run content:review), deployment, and admin publication statistics are documented in Generated use-case pages.

Call POST https://api.typesafe.pro/v1/systemone with JSON. Authentication is optional: requests without a token use anonymous access. If you have an API token, send it as a Bearer token to receive the limits for your account.

curl https://api.typesafe.pro/v1/systemone \
  --request POST \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "jev-latest",
    "state": "The light is on.",
    "questions": {
      "is_on": {
        "type": "noul",
        "instructions": "Is the light on?"
      }
    }
  }'

Example response:

{
  "model": "jev-1.13.0",
  "answers": {
    "is_on": {
      "type": "noul",
      "noul": 0.98
    }
  },
  "usage": {
    "input_tokens": 25,
    "output_tokens": 8
  }
}

The value returned for a noul answer is a probability from 0 to 1.

Use a token in the Authorization header when one has been issued to you:

curl https://api.typesafe.pro/v1/systemone \
  --request POST \
  --header 'Authorization: Bearer YOUR_API_TOKEN' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "jev-latest",
    "state": {"ticket": "My payment has failed for three days."},
    "questions": {
      "is_urgent": {
        "type": "noul",
        "instructions": "Is this support request urgent?"
      }
    }
  }'

Missing, malformed, duplicate, or unknown tokens receive anonymous access. Do not put a token in a URL, browser bundle, public repository, or client-side app.

The questions object can combine several question types in one request.

{
  "model": "jev-latest",
  "state": {
    "message": "I was charged twice and need help today.",
    "plan": "pro"
  },
  "questions": {
    "department": {
      "type": "choice",
      "instructions": "Which team should handle this request?",
      "criteria": {
        "billing": "Charges, invoices, refunds, or duplicate payments.",
        "technical": "Product bugs, outages, or integrations.",
        "sales": "Pricing, upgrades, or product information."
      }
    },
    "urgency": {
      "type": "score",
      "instructions": "How urgent is this request?",
      "criteria": ["Routine", "Soon", "Urgent"]
    },
    "needs_response_today": {
      "type": "noul",
      "instructions": "Does this request need a response today?"
    }
  }
}

state may be a string, JSON object, or JSON array. The response uses the same keys you chose in questions.

Each request can contain up to 64,000 context tokens and a body up to 10 MiB. Requests are queued when shared upstream capacity is busy. A higher account tier is served ahead of a lower tier; requests within the same tier stay FIFO.

Limits may change as the service evolves. 429 means the request rate limit or queue deadline was reached. 503 means the queue or the shared limiter is full. Both responses include Retry-After; wait at least that long and retry with exponential backoff. Do not retry a non-idempotent request automatically unless your application can safely handle duplicate work.

The API forwards successful and upstream validation responses as received. Its own errors are JSON objects such as:

{"error":"rate_limit_exceeded"}

Common statuses:

Every response includes X-Request-ID. You may provide your own value using 1-128 letters, digits, _, or -; otherwise the API generates one. Include this ID when reporting a problem.

curl --include https://api.typesafe.pro/health

The endpoint returns a small JSON status document and does not call the model.

Keep API tokens server-side. The service retains operational activity and error events for seven days, including real client IP addresses for abuse investigation and usage statistics. Client token identifiers are hashed. Error records omit authorization values and request bodies and keep only a bounded, sanitized portion of an upstream error response. These records are available only through the authenticated admin API.

The read-only dashboard at /admin uses the protected /admin/api routes on the API host. See admin operations for configuration, secrets, retention, and local verification.

For full model behavior, question types, and response shapes, use the TypeSafe API reference. When contacting support, include the X-Request-ID, approximate time of the request, and the HTTP status.

The gateway forwards only POST /v1/systemone. It checks the decoded path and method before reading a body or performing admission. /health and /admin remain local routes; unknown paths are never forwarded. Browser CORS preflight is supported for POST /v1/systemone.

Send Content-Type: application/json (parameters such as charset=utf-8 are accepted). Request compression is unsupported: omit Content-Encoding or use identity. The existing 10 MiB body limit and client rate limits apply.

Before reserving an upstream key, the gateway validates JSON and the documented model, state, and questions structure, including noul, choice (1–255 options), and score (2–10 levels). Duplicate JSON keys and non-finite numbers are rejected. Additional fields and query parameters are preserved, and valid bodies are forwarded byte for byte. Model availability and semantic constraints are checked by upstream. Anonymous access remains supported.

Local errors use the existing error field:

Schema errors include the first failing field as a details.path array and a details.reason string, without its input value. These permanent errors have no Retry-After header. They retain CORS, X-Request-ID, and Cache-Control: no-store headers and appear under their own error codes in the admin dashboard. Early route rejections have client tier unknown because authentication was skipped.