Request best-effort cancellation of a running request by its id. The gateway aborts the request if this instance is serving it for the caller’s org, and fans the id out over Redis so whichever instance holds it aborts too. There is no request body.

​
Authentication

Authenticated with a secret key (rd_sk_…) or a client token. A client token must carry an inference scope — ai.chat or ai.embed — or the request fails 403 permission_error (insufficient_scope); a feedback-only token makes no requests and so may cancel none. A publishable key is rejected 403 (publishable_key_mint_only). A caller may only cancel requests within its own org, and a client token is additionally scoped to its own end user — it cannot cancel another end user’s request in the same tenant.

​
Path parameters

​
Response

202 Accepted:
The response is a uniform 202 whether the id was actively held, already finished, unknown, owned by another tenant, or owned by another end user — so it is never an existence oracle. 202 means the cancel signal was accepted, not that a request was necessarily aborted.
An in-flight request that is successfully aborted finalizes with outcome canceled: a non-streaming call returns 499 invalid_request_error (request_canceled); a streaming call receives a terminal error frame with code request_canceled.

​
Errors

Failures use the error envelope: 401 authentication_error (bad or missing key), 403 permission_error (publishable_key_mint_only, insufficient_scope).