By Hamza Faisal
The Solana Foundation has launched Solana Confidential Compute, an open-source toolkit that lets developers build private AI applications and prove — on the blockchain — that the hardware running them is what it claims to be. The toolkit, built in collaboration with the Nito team, was announced on November 13.
The idea tackles a quiet problem in AI: when you send a sensitive prompt to an app, you have to trust that the server on the other end is running the software it says it is, on hardware nobody has tampered with. Solana's answer is to put the proof on-chain, while keeping the actual computation — and your data — off it.
How the verification works
The system runs AI workloads inside trusted execution environments, or TEEs — sealed-off areas of a processor that isolate code and data from the host operating system. A machine inside a confidential virtual machine generates a signing key that exists only inside that protected hardware. The hardware then signs a report tying the key to a recent Solana slot, and an on-chain verification program checks the report and records the key, its hardware type and an expiry date.
When the machine produces an answer, it signs the output with that key. Anyone receiving the answer can verify the signature and check the key's record on Solana before trusting the result — without ever seeing the prompt or the private inputs behind it. The AI computation itself happens off-chain; only the verification record lives on the blockchain.
What developers get today
The kit ships with a verifier program live on Solana's devnet, Rust and TypeScript software development kits, a daemon that runs inside the enclave, a command-line tool and a Docker image. It currently supports hardware attestations from Intel TDX and AMD SEV-SNP, along with AWS Nitro Enclaves and Google Confidential Space; verification for Nvidia GPUs is still in development.
The Foundation pitched the toolkit for applications handling personal conversations, sensitive business information and private trading strategies, and showed off a demo chat app built on it. Nito, the team that built the toolkit with the Foundation, will fold the infrastructure into its own private-AI platform, and other teams can reuse the same components.
Beta only — for now
There are limits worth noting. The toolkit is in beta on Solana's devnet, and the Foundation stresses that it has not yet been audited and is intended for development use only. A mainnet launch is planned once the security audit is complete, but no date has been given — and verification proves only that a published software release ran in a given environment, not that the code itself is bug-free.
Still, the launch puts Solana in the middle of a growing race to make AI trustworthy by construction rather than by promise — and, if the audit holds up, gives developers a way to let users check the machine behind the answer for themselves.
By Hamza Faisal
