Endpoints

Every route this API mounts, what it needs, and the ones people expect that are deliberately absent.

Base URL in this deployment: https://api.llmeu.com

Method Path Credential Notes
GET /healthz none Liveness. Reads nothing.
GET /readyz none Readiness: database, first-party endpoints, managed runtimes. Answers 503 while a GPU is still loading weights.
GET /v1/models key · inference Only the models this key may actually call under its organization's default policy.
GET /v1/models/{id} key · inference One model, with its licence and whether it is hosted here.
POST /v1/chat/completions key · inference The OpenAI-compatible chat endpoint, streaming or not.
POST /v1/completions none Not offered. Answers 501 not_implemented and names the endpoint to use instead.
POST /v1/embeddings key · inference Embeddings from an embedding model. This deployment serves them from its deterministic local backend, and the response says so: llmeu.mock is true in the body.
POST /v1/playground none Anonymous, mock backend only, and limited per IP address.
GET /platform/v1/me key · platform_read The organization this credential belongs to.
GET /platform/v1/api-keys key · platform_read The organization's keys, without secrets.
POST /platform/v1/api-keys key · platform_write Create a key. The secret appears in this response and never again.
DELETE /platform/v1/api-keys/{id} key · platform_write Revoke a key. Effective on the next request.
GET /platform/v1/traces key · platform_read Recorded metadata for past calls. No prompt or completion bodies: retention is zero.
GET /platform/v1/traces/{id} key · platform_read One trace, including the routing decision.
GET /platform/v1/usage key · platform_read Usage aggregated by day and by model.
GET /platform/v1/policies key · platform_read The organization's policies.
GET /platform/v1/policies/{id} key · platform_read One policy.
POST /platform/v1/policies key · platform_write Create a policy.
PUT /platform/v1/policies/{id} key · platform_write Edit a policy. Applies to the next request, not to one already in flight.
GET /platform/v1/status none Deployment status. It takes no credential, so treat it as non-confidential.

What is not here

There is no /v1/files, /v1/batches, /v1/assistants, /v1/fine_tuning, /v1/images, /v1/audio and no /v1/moderations. They are not mounted, so they answer 404 unknown_route rather than a 501 that would imply they are coming. There are no webhooks and no SDK: any OpenAI-compatible client is the SDK.

Status without a key

This endpoint takes no credential, which makes the deployment's own status public. It is written here rather than discovered later, because it is a product decision: if your policy needs an authenticated status probe, change it deliberately. The response never contains organization data.