planned, not live

TEE inference on CPUs.

TEE inference means your prompts run sealed inside a hardware enclave, a trusted execution environment. The host can't read them. You can verify that.

Pricing: TBD, set after we measure our own hardware.

TBDThroughput

TBDPricing

Not live yetAvailability

Create an account Read the docs

example request

curl https://api.cpuinference.ai/v1/chat/completions \
  -H "Authorization: Bearer $CPU_INFERENCE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "lfm3-1.7b",
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }'

Preview: this endpoint is not live yet.

How TEE inference works

TEE inference runs the model inside a trusted execution environment, so the host cannot see your data. This is the intended flow. It is not running on this preview.

Encrypt to an attested enclave

Your request is encrypted to a hardware enclave. The host that holds the machine should not be able to read the prompt.

Run inside a TEE

The model runs inside AMD SEV-SNP or Intel TDX, sealed from the host.

Verify before you send data

You can check the attestation report first, including the measurement and the vendor certificate chain.