Workloads & runs

Follow a request

After a completion, use request state and its receipt to inspect the recorded outcome. Model discovery, readiness, dispatch and completion are separate states. A pre-dispatch MODEL_NOT_READY rejection does not establish an invocation or usage charge.

Inspect GET /v1/inference/requests/{id} and GET /v1/requests/{id}/receipt with the appropriate project credential. After a timeout, reconcile authoritative state before sending another attempt. Cancellation is a request, not proof that execution stopped.

Attribute work to your run

Use the optional X-Infermax-Attribution header to associate a completion with a feature and an opaque run ID. Sub-agent lineage requires an existing parent run in the same project. Keep prompts, personal data and names out of attribution.

Read attribution and reporting contracts →

Deployment and execution intents

Use Console endpoints & readiness and compute requests for authorized workflows. A public workload deployment or execution-intent write API is not included in this Customer reference. It will be published when its external contract is supported.