API reference

Base URL, authentication, the headers on every response, and what is on this tab.

Everything on this tab is generated from the API’s own specification, so a parameter listed here is a parameter the service accepts today.

Base URL

https://compute.prentis.ai/v1

Anthropic SDKs append /v1/messages themselves, so give them the origin (https://compute.prentis.ai) rather than the base above.

Authentication

Authorization: Bearer $PRENTIS_API_KEY

/v1/messages also accepts x-api-key: $PRENTIS_API_KEY. See Authentication.

Endpoints

POST /chat/completions	The one you want. OpenAI-compatible, streaming or not.
POST /completions	Legacy text completion, for prompts that are not a conversation.
POST /messages	Anthropic-compatible. Same models, Anthropic’s wire format.
GET /models	What this key can call.
GET /models/{model}	One model by id.

On every response

Header	Always?	What it is
x-request-id	yes	The handle that reaches the whole path your request took, on successes and errors alike. Quote it when you ask us about a call.
x-provider-class	on success	primary or fallback — whether you got the first choice or a standby. It never names who served it.
x-config-version	on success	Which routing configuration was in effect. Useful only when comparing two calls that behaved differently.
Retry-After	on 429 / 503	Seconds to wait. Honour it before falling back to your own backoff.

Two conventions worth knowing before you read on

Unknown parameters are ignored, not rejected. Sending an OpenAI parameter this API does not implement will not fail your call. The exceptions are marked not supported on each endpoint — those return INVALID_REQUEST, because silently ignoring them would change the answer in a way you could not detect.

Errors are OpenAI-shaped, and code is the stable part. Messages get reworded; codes are append-only and a retired one is never reused. Branch on error.code. The full list is on Errors.