Docs

1 · Get an API key

Sign in to the console and open Tokens. Create a key, copy it and store it somewhere safe — the console only shows it in full once. Keys are bearer credentials: anyone who has one can spend your balance, so treat it like a password and give a separate key to each application.

You also need a positive balance before the first call: top up in Wallet in the console. Requests stop when the balance reaches zero instead of running up a bill.

2 · Point your client here

Everything lives behind one base URL and one header:

Base URL   https://api.tokwork.com/v1
Header     Authorization: Bearer <your-api-key>

Because the surface is OpenAI-compatible, most SDKs only need the base URL and the key changed. Requests in the Anthropic and Gemini formats are accepted as well.

3 · Make the first call

Generate an image from the command line:

curl https://api.tokwork.com/v1/images/generations \
  -H "Authorization: Bearer $TOKWORK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "An orange cat at sunset, watercolour",
    "size": "1024x1024"
  }'

The response carries the image twice — once inline, once as a hosted URL:

{
  "created": 1791610498,
  "data": [
    {
      "b64_json": "iVBORw0KGgoAAAANSUhEUg…",   // base64 PNG
      "url": "https://…/1791610498419702800.png",      // hosted copy
      "width": 1024,
      "height": 1024,
      "revised_prompt": "An orange cat at sunset, watercolour"
    }
  ],
  "usage": { … }                                     // per-request usage detail
}

With the official OpenAI SDKs, keep your code and change two lines: base_url to https://api.tokwork.com/v1 and api_key to your TokWork key.

Image sizes and prices

Image models are billed per image, and the price depends on the output size — the request’s size is matched to a tier by its long edge:

Leave size empty or set it to auto and the cheapest tier is used. Per-model prices, including the size breakdown, are on the pricing page and in the console — they are generated from the gateway, so they are current.

Endpoint reference

GET /v1/models is the source of truth for what your key can call right now — it is the same list that powers the console and the pricing page. Today the gateway serves image generation models; text and chat models are enabled model by model, and a model that is not enabled yet answers with 503 model_not_found rather than a silent fallback.

Errors

Errors use the OpenAI error envelope, and every response carries a request id you can quote to support:

{
  "error": {
    "message": "No available channel for model … (request id: …)",
    "type": "new_api_error",
    "code": "model_not_found"
  }
}

A failed request is not charged. Every successful one is logged with the model, the size and the amount deducted, so you can reconcile spend in Logs.

Need help?

Mail support@tokwork.com with the request id and the model name, or use the contact form. We answer within one business day.