Overview

How API keys, inference, and c8s attestation work in the Confidential Inference API.

The Confidential Inference API uses one public gateway in a confidential c8s cluster.

Client → gateway → inference router → inference workers

The gateway authenticates the API key. The inference router and workers do not receive that key.

Use the service

  1. Obtain an API key from Confidential AI.
  2. Call /v1/models to get the available model ID.
  3. Call /v1/chat/completions or /v1/completions with the key.

Verify the deployment

Call /attestation with a fresh 32-byte nonce before you send sensitive input. This endpoint does not require an API key.

The response contains c8s evidence for the gateway, router, and inference workers. An independent verifier checks the confidential VM base image, Intel TDX evidence, nonce, c8s operator key, allowlist, workload identity, OCI image digest, and effective command.

Attestation proves what c8s approved and launched. It does not prove that the service is healthy or that model output is correct.