planned, not live
Attestation
CPU Inference is meant to run TEE inference on CPUs, inside a hardware enclave. The planned platforms are AMD SEV-SNP and Intel TDX. You should be able to verify a remote attestation report before you send data. This page describes that planned flow. It is not live, and nothing here is an audit or a certification.
The planned flow
Fetch the attestation report
Call a placeholder endpoint, GET /v1/attestation. The host below is also a placeholder and will not return a real report on this preview.
curl https://api.cpuinference.ai/v1/attestation
Check the measurement
The report should include a measurement of the enclave. You compare that with the build you expect. If it does not match, do not send a prompt.
Check the vendor certificate chain
The report should come with a vendor certificate chain, from AMD or Intel, depending on the platform. You verify that chain before you trust the measurement.
Then send the request
Only after those checks should a client encrypt the request to the attested enclave and call the inference API. The host should not be able to read the prompt. That path is planned, and it is not live here.
What this preview does not do
This preview does not issue attestation reports, does not run a TEE, and does not serve models. The endpoint above is a placeholder. Do not treat any example document in the console as a real attestation.