Skip to main content
Confidential compute means your data stays private even from the machine running it. The Singularity Cloud Network achieves this with hardware Trusted Execution Environments (TEEs), remote attestation, and end-to-end encryption for AI inference.

​
Trusted Execution Environments (TEEs)

A TEE is a hardware-isolated enclave (e.g. Apple Secure Enclave, Intel TDX/SGX, AMD SEV-SNP, AWS Nitro) where code and data are protected from the host OS, the operator, and other processes. Memory is encrypted and access-controlled by the CPU itself. On the network, every compute node runs inside a TEE — so the operator hosting your workload cannot inspect it.

​
Remote attestation

Before a node can serve, it produces a hardware attestation — a signed proof, rooted in the CPU vendor, that it is a genuine enclave running the expected software. The network verifies this; only attestation-verified nodes receive work. When you run inference, the response is tagged with the serving node’s attestation so you can confirm it ran in a real enclave (see Confidential inference).

​
End-to-end encryption (AI inference)

For Grid inference, privacy goes one step further than “TEE on the server”:
  • Your client fetches the serving node’s public key and encrypts the prompt in your browser/app, sealed to that specific enclave.
  • The network relays only ciphertext — the orchestrator never sees your prompt or the response in the clear.
  • The node decrypts inside the enclave, runs the model, and seals the reply back to your key.
So the plaintext exists only in two places: your device, and the hardware enclave. Not in transit, not in any database, not with the platform.
This is why Grid chat history is never stored server-side — the platform cannot read it by design. Any history is kept client-side.

​
What this protects against

  • A malicious or compromised operator reading your prompts.
  • The platform logging or training on your data.
  • In-transit interception (the relay only ever holds ciphertext).

​
What to still consider

  • The model output is only as private as where you send it next.
  • Attestation proves the enclave is genuine and running approved software; always verify the attestation badge for sensitive workloads.