Overview

Verify Phala gateway attestation, pinned TLS, signed receipts, and upstream evidence.

Private AI verification connects the gateway’s hardware evidence to the exact request and response bytes. The attestation report identifies the Phala gateway. The response receipt cites a verified upstream session, whose evidence coverage varies by provider.

Every model on https://inference.phala.com/v1 uses the TEE-only serving policy: upstream verification and channel binding are required before forwarding. A failed check refuses the request. Ordinary clients still need a verification client to check this from their own machine.

​

Start with a verification client

• For Node.js or Bun, use the TypeScript ACI verifier: connectAci(), the scoped aci.fetch, and verifyReceipt() after each response.
• For coding agents, use Private AI Proxy: attestation and pinned TLS before traffic, receipt audits after delivery.
• Use Verify a Response to understand the checks and inspect existing artifacts. cURL and jq inspection alone does not verify them.

​

Verification workflow

1. Verify the gateway before traffic

   Generate a fresh nonce, verify the TDX quote and JCS keyset digest, check not_after, replay RTMR3 and the measured compose hash, and apply your release policy. Source labels alone are not a trust root.

2. Pin the request channel

   Use hostname-validated TLS pinned to the SPKI listed for inference.phala.com in the verified keyset.

3. Verify the receipt

   Match workload_keyset_digest, verify the receipt signature under its key_id, and compare request.received.body_hash and response.returned.body_hash with the exact bytes exchanged.

4. Audit the cited upstream session

   Require upstream.verified.result: verified and required: true. Recompute the full session’s content address and verify its evidence, channel binding, and claims under your policy.

A passing audit proves the exchange used the attested gateway and cited upstream channel. It proves a reviewed release only with an independently obtained compose allowlist. It does not imply that every provider proves GPU binding, model-weight provenance, or the same software measurements.

​

Canonical and legacy endpoints

Purpose	Canonical endpoint	Legacy alias
Gateway attestation	GET /v1/aci/attestation?nonce=	GET /v1/attestation/report
Per-response proof	GET /v1/aci/receipts/{id}	GET /v1/signature/{id}
Upstream security context	GET /v1/aci/sessions/{session_id}	None

Use the canonical /v1/aci/* endpoints for new integrations.

Attestation

Verify a Response

TypeScript ACI Verifier

Private AI Proxy