Skip to main content

​
Base URL

The API follows the OpenAI chat completions request and response shapes, so OpenAI-compatible SDKs work by changing the base URL. Protégé-specific fields are additive: task on the request, and a protege object on the response.

​
Authentication

Every request carries a bearer token.
x-api-key: $PROTEGE_API_KEY is accepted as an alternative. Keys are scoped to an environment. Use separate keys for production and staging, so staging traffic does not enter a task’s production history. This is the only credential involved. Protégé is managed, so there is no bring-your-own-key header and no provider account to configure.
The key authorises real inference spend. Never expose it in a browser, a mobile binary, or anything else a user can read. Call Protégé from your server.

​
Content type

Requests are JSON. Streaming responses are server-sent events.

​
Versioning

The version is in the path. /v1 will not change in a way that breaks an existing request body. Additive changes ship without a version bump, so parse responses leniently. Unknown response fields should be ignored rather than treated as errors, since the protege object gains fields over time.

​
Rate limits

600 requests per key per minute, returned on every response. A looser per-IP ceiling applies before a key is verified. A 429 means the window is exhausted. Back off using x-ratelimit-reset-requests rather than retrying immediately.

​
Request IDs

Every response carries x-request-id. Include it when reporting a problem; it is what lets us find the exact call, its route and its trace.

Chat completions

The endpoint, and the task parameter in full.