Private AI Gateway
An OpenAI-compatible gateway for attested confidential inference that publishes hardware attestation for itself, verifies its upstream model providers before forwarding a prompt, and signs a receipt for every response.
Private AI Gateway lets a client verify, before trusting a response, that the gateway is a specific, unmodified workload running inside a genuine trusted execution environment, that the upstream provider it forwarded a prompt to was itself verified, and that the response is cryptographically bound to that verified identity through a signed per-request receipt. If verification is required and cannot be completed, the gateway does not send the prompt.
It is the reference implementation of the Attested Confidential Inference specification, and is also the workload that a companion launcher tool can fetch, build and run inside a dstack v2 application VM. Security auditors are directed to the project's security claim, request-flow and auditor checklist documentation, while new users and agent developers are pointed to a shorter mental model and a local smoke-test path before reading provider-specific integration details.