Docs

1. Step 1

   Create an API key

   Sign up, open your organization settings, and create an API key. The secret is shown once — store it somewhere safe.

2. Step 2

   Send the request

   POST to the classify endpoint with a managed model ID from the model catalog. The example below classifies one string.

   Browse the model catalog

3. Step 3

   See it in the dashboard

   The request appears in your organization’s logs and usage with its outcome, latency, and exact cost. Logs arrive with a short delay.

curl -X POST https://api.truepositive.ai/v1/classify \
  -H "Authorization: Bearer $TRUEPOSITIVE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "truepositive/moderation", "input": "hello" }'

Replace $TRUEPOSITIVE_API_KEY with your organization API key. This page never embeds a live secret.

Every request carries an organization API key as a Bearer token in the Authorization header. Keys start with tp_live_, are created and revoked in organization settings, and one key authenticates every managed model. Usage is attributed to the organization that owns the key — never to an individual project.

Batches return one result per input string, in order, identified by its index. A batch counts each string as one billed item.

• Keys are up to 128 characters: letters, digits, and ., _, :, -.
• Repeating a key with the same request body replays the recorded response — the retry is not charged again.
• Reusing a key with a different body is rejected with idempotency_conflict.
• If the original request is still processing, the retry is rejected with idempotency_in_progress; try again shortly.

curl -X POST https://api.truepositive.ai/v1/classify \
  -H "Authorization: Bearer $TRUEPOSITIVE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: order-4187-first-check" \
  -d '{ "model": "truepositive/moderation", "input": "hello" }'

{
  "id": "req_9f4c2ab1",
  "model": "truepositive/moderation",
  "results": [
    { "index": 0, "scores": { "Safe": 0.98, "Harassment": 0.01 } }
  ],
  "latency_ms": 141,
  "usage": {
    "input_items": 1,
    "cost": { "amount": "0.001000", "currency": "USD" }
  }
}

• results[].scores maps each label to a score between 0 and 1. An empty map means no label applied.
• usage.cost.amount is the exact USD debit for the request, to six decimal places; usage.input_items counts the billed strings.
• id identifies the request across logs, usage, and support.

The example above is illustrative; scores, latency, and cost come from your actual requests.

{
  "id": "req_9f4c28d2",
  "error": { "code": "insufficient_credits" }
}

• When payload storage is on, request and response payloads are kept for the configured period — 1 to 365 days — then deleted.
• When payload storage is off, payloads are never stored. Request metadata and usage are always logged either way.
• Changing the setting affects future requests; earlier requests keep the setting that was active when they were received.
• Owners and admins manage retention in organization settings.