Overview

Base URL

https://reapi.ai/api/v1

All endpoints accept and return JSON. The API mirrors OpenAI's conventions where it makes sense — many existing OpenAI clients work by changing only base_url.

Authentication

Every request must carry an API key as a Bearer token:

Authorization: Bearer rk_live_xxxxxxxxxxxx

Create keys in your dashboard. See Authentication for full details.

Asynchronous tasks (image, video)

Generation calls are asynchronous. The pattern is:

1. POST the model endpoint (e.g. /api/v1/images/generations) → returns a task_id immediately.
2. GET /api/v1/tasks/{task_id} periodically until status is completed or failed.

┌──────────┐        ┌──────────┐        ┌──────────┐
│  POST    │  →     │processing│  →     │ completed│
│ /images/ │        └──────────┘        │  failed  │
│generations│                           └──────────┘
└──────────┘                                  │
                                              ▼
                                       output / error

Recommended polling cadence: 1 request every 1–2 seconds.

For details see Tasks. Per-model request schemas are on each model's page in the Models sidebar.

Credits

Each call deducts credits from your balance. Failed tasks are automatically refunded by the worker the moment the workflow ends in failure — atomically, before any poll observes the failed status. Refund is one-shot; re-polling never refunds twice.

Pricing per model is configured in your account dashboard.

Idempotency

reAPI does not deduplicate by Idempotency-Key. Every successful HTTP POST to a generation endpoint creates a new task and charges credits — by design, so no upstream provider call is ever silently skipped.

If you need retry safety, generate the same payload only once on your side or rely on the natural recovery path: when a task fails, credits are refunded automatically and you can resubmit.

The Idempotency-Key header is currently accepted and recorded for telemetry but does not alter request behavior. Don't rely on it to prevent duplicate charges.

Rate limits

Requests are limited per user, in 1-second windows. The default is 10 requests / second / user. When exceeded, you get HTTP 429 with code 50001 and a Retry-After header.

Polling GET /api/v1/tasks/{id} shares the same limit, so don't poll faster than once every 1–2 seconds per task.

Response format

Successful responses are JSON of the resource type. Errors follow a standard shape:

{
  "error": {
    "code": 20002,
    "message": "Both `model` and `prompt` are required",
    "request_id": "req_8a4f0d2e-1c8b-4f1a-9e2d-3b7c5a6f0a1b"
  }
}

code is a 5-digit numeric identifier (see Errors for the full catalog). request_id is unique per call — include it in support requests.