A gateway proxy that keeps the key

cloud.ai

Requests are forwarded to the Vercel AI Gateway from your server. The key never enters the bundle, and every call is rate limited per session before anything is forwarded — which is what makes it safe to leave on a public URL.

Model identifiers are the gateway’s own provider/model strings, passed through untouched. There is no abstraction over model names, so a model released today works today.

Editor

Output

Press Run.

Network — real requests to this site’s own backend

No requests yet.

chat(prompt | messages, options?)

Returns a value that is both awaitable and async-iterable: await it for the whole reply, or iterate it for tokens as they arrive. Options take model, system, temperature and maxOutputTokens.

generateImage(prompt, options?)

Returns base64 images with their MIME types. Takes model, size, n.

transcribe(audio, options?)

Accepts a Blob, ArrayBuffer, typed array or base64 string. Returns text and duration.