For developers
One API. Every model.
An OpenAI-compatible endpoint that routes to OpenAI, Google and xAI image models today — with one key, one balance and per-request logs.
Five lines to your first request
const client = new OpenAI({
apiKey: process.env.TOKWORK_API_KEY,
baseURL: 'https://api.tokwork.com/v1',
})
await client.images.generate({ model: 'gpt-image-2', prompt: 'a red fox' })Endpoint map
| Path | Purpose |
|---|---|
/v1/images/generations | Image generation (this is the one that works today) |
/v1/chat/completions | Chat completions, OpenAI shape |
/v1/embeddings | Embeddings |
/v1/models | Model list available to your key |
/api/pricing | Public live price list (no auth) |
The gateway exposes more endpoint types than the current catalogue uses; the model list above reflects what is actually enabled right now.
Models available today
- claude-sonnet-4-6$0.0000 / image
- muse-spark-1.3$0.0000 / image
- seedance-2.5per output size
- claude-fable-5$0.0000 / image
- claude-opus-5-5$0.0000 / image
- claude-sonnet-5$0.0000 / image
Developer FAQ
Is the API OpenAI-compatible?
Yes. Point the OpenAI SDK at https://api.tokwork.com/v1 and change the API key. Chat, embeddings and image generation use the standard request shapes.
Do I need separate accounts with each vendor?
No. You have a TokWork account and key; TokWork holds the upstream provider accounts and routes each request.
How is billing handled?
Prepaid balance, charged per request. Text models are metered by tokens; image models are charged per generated image, with some models priced by output resolution.
What happens when a provider is down?
Channels are health-checked and removed from the routing pool when they fail, so requests are served by a working channel where one exists.
Can I switch models without changing my code?
Yes. The model is a request parameter. Changing it is a string change, not a new integration.
Get a key
Create an account, top up, and copy a key from the console. No sales call required.