Every major model, one endpoint, one bill.
An OpenAI-compatible API in front of every provider worth calling. Swap models by changing a string, not a client library — and see what each call cost before the invoice.
What you get
OpenAI-compatible
Point your existing client at a different base URL. Chat, embeddings, images, speech and transcription.
Bring your own key
Use your provider account and pay a small platform rate, or use ours and keep one bill. Your keys are stored sealed.
Priced per call, visibly
Every request is itemised by model in your usage, not folded into a single number.
Reserved before it runs
Credits are held before the provider is called, so an empty balance is a clean refusal rather than a bill you did not expect.
Refunded when it fails
A provider timeout costs nothing. The reservation goes back.
Limits you control
Output tokens, input truncation and per-call ceilings, set per workspace.
Try it without writing anything
The playground runs a real call in the browser against whichever model you pick.
Open the playground