EnclaveDashboard
/PRIVATE INFERENCE

Private AI.
Verifiable execution.

The intended Enclave inference path verifies the environment before a secret is released, then returns a signed receipt alongside the answer.

Confidential computing infrastructure

Verify before release

A composite quote binds the confidential CPU environment to the GPU. Verification checks vendor roots, NVIDIA attestation material and the current measurement policy. An invalid quote prevents key release.

The deployment design uses a confidential virtual machine with Intel TDX or AMD SEV-SNP, plus NVIDIA confidential computing. GPU isolation alone is not the complete trust boundary.

Run signed, measured images

Only approved model and serving images run. Model weights decrypt after attestation. TLS terminates inside the confidential virtual machine so the host is outside the plaintext request path.

The gateway, key-release agent and receipt signer work together to bind each response to its measured execution environment.

Return the evidence

An InferenceReceipt contains modelHash, codeHash, inHash, outHash, attRef, ts and sig. The receipt anchoring path targets Arc AttestationVerifier and ModelRegistry contracts.

Operational stores retain hashes, receipt metadata and policies. Plaintext inputs and outputs are not intended to enter those stores.

Know what is proven

A receipt establishes provenance: which approved model and code executed in an attested environment. It does not establish that an answer is accurate, unbiased or fit for a specific decision.

The dashboard currently demonstrates this lifecycle with local, unsigned preview receipts. Production attestation and chain anchoring require live infrastructure.

Explore the dashboard
/ENCLAVE

Coming soon.

Enclave’s official channels are on the way.