Skip to main content
The pages in this section are generated from the platform’s own OpenAPI specification, so they describe the API as it is actually served rather than as it was last written down.

Base URL

Authentication

Every request carries an API key as a bearer token:
Create and scope a key under API keys. Keys belong to an organization, so a request is always attributed to one tenant.

Request headers

Each endpoint page below lists the headers that endpoint reads, and the SDKs take them as arguments. This is the overview. Idempotency-Key is read by the three endpoints that meter, because it is what billing dedupes on. All but Authorization are optional, none of them changes how a request is served, and none of them is charged. Successful chat completions answer with an X-Request-Id header naming the platform’s request id, which is the id you attach a score to afterward. It is the Idempotency-Key you sent, when you sent one.

Compatibility

The chat and embedding endpoints are OpenAI-compatible. Any OpenAI client works by changing one base_url, which is the fastest way to try the platform. The native endpoints — Memory, scores, threads and the audit log — have no OpenAI equivalent and are reached through the same key. For typed access in PHP, TypeScript or Python, see the SDKs. They are generated from this same specification.