.png){kind=logo}
The Solana Foundation has launched Solana Confidential Compute, an open-source toolkit that allows developers to build private AI applications and verify the hardware running them. Developed with AI platform Nito, the tools entered beta on Solana devnet on November 13.
The toolkit includes developer software and a program that records hardware checks on Solana. AI requests run off-chain, leaving private conversations and other sensitive information outside the blockchain.
In its November 13 announcement, the Solana Foundation said the toolkit uses trusted execution environments, or TEEs, to protect information while computers process it. These hardware environments isolate application code and data from the machine's operating system and operator.
The hardware produces a signed report showing which protected environment is running and what software it contains. A program on Solana checks that report and records a signing key, the hardware type and an expiration date.
The machine can then sign its AI responses using a key generated inside the protected environment. Applications can check the signature against the on-chain record before accepting an output. The hardware report also connects the signing key to a recent Solana slot.
The initial release supports Intel TDX and AMD SEV-SNP hardware attestations. However, support for NVIDIA GPU verification remains under development. The Foundation has not announced a mainnet launch date.
The Foundation also published a Confidential Chat developer template using Next.js and React. Under the documented setup, a user's browser checks the confidential machine's attestation before encrypting a message. The browser does not send the prompt if verification fails.
The encrypted request travels through an application server and Nito to a protected environment running the AI model. The environment reads the request and encrypts the response before returning it. The intermediaries cannot read either message under this design.
However, the template does not hide all information about a request. Its documentation says the application server can see the user's IP address, while Nito can see the account, model, request time and encrypted message size.
The Foundation's product documentation also states that its current confidential chat does not yet use the Solana on-chain verification program. Instead, the browser verifies the hardware report directly. The template requires a funded Nito account and an API key.
Solana and Nito have presented personal assistants, business-data tools and private trading agents as potential uses for the software. Developers could also build systems that accept trading instructions or automated decisions only when verified machines sign them.
The Foundation lists prediction-market rulings, insurance claims and governance decisions as other possible applications. A Solana program could check a machine's verification record before acting on its output.
The launch follows Solana's November 9 proposals for an AI inference market and the November 10 rollout of Solana Privacy Rings. Previous privacy work included confidential token vesting through Umbra and Streamflow.
The release remains limited to development and testing. The Foundation stated, “It hasn’t been audited yet, so use it for development for now.” Its product page says a security audit is underway and a mainnet version will follow that process. Developers must also implement authentication, rate limits and safeguards against automated misuse.
ALSO READ: Bitcoin Price Falls Toward USD 68,000 as Stablecoin Supply Shrinks USD 14.2B