The Solana Foundation has launched Solana Confidential Compute, an open-source toolkit developed in collaboration with Nito that enables developers to create private AI applications and verify their computation hardware on Solana.
This toolkit is currently in the Solana devnet testing phase, with a mainnet version planned for release.
- The toolkit is currently in beta on the Solana devnet, with plans to launch on mainnet.
- Supports Intel TDX and AMD SEV-SNP attestation; NVIDIA GPU verification is still in development.
- AI processing occurs off-chain, while applications can view hardware verification records on Solana.
- These tools have not yet been audited and are currently primarily intended for developers building and testing applications.
On November 13, the Solana Foundation announced that this is a software development kit and an on-chain verifier developed by Nito in coordination with the Foundation. Nito is integrating this infrastructure into its private AI platform, and other teams can also use the same open-source components.
The foundation wrote on X: “Private AI, verified onchain,” and introduced these tools as suitable for applications involving personal conversations, sensitive business information, and private trading strategies.
This beta version supports hardware attestation from Intel TDX and AMD SEV-SNP; NVIDIA GPU attestation is still under development. The foundation also stated on the same thread that this version has not been audited and is intended for development use only. The announcement noted that the mainnet version is coming soon, but no release date was provided.
How does Solana Confidential Compute verify the hardware behind AI?
According to the foundation, confidential computing protects information while it is being processed by a machine. The system uses a Trusted Execution Environment, or TEE, to isolate code and data from the surrounding infrastructure, including the host operating system and its operators.
Inside the confidential virtual machine, the software generates a signing key and stores it within protected hardware. The hardware-signed report associates this key with the most recent Solana slot, and the verification program checks this report before recording the key, hardware type, and expiration date.
When a machine-generated answer is produced, it is signed using this key. The application receiving the answer can first verify its signature, then query the corresponding record on Solana, and only then accept the result.
The foundation stated that AI computation occurs off-chain. Publishing verification records does not require disclosing users' conversations or private inputs.
Private chats are encrypted before the prompt leaves the browser.
The Solana chat template provides a development example using Next.js and React. Its documentation states that the browser first verifies the secret environment before encrypting the message and sending it to its key; if verification fails, the prompt is not sent.
Encrypted messages are transmitted through the application server and Nito to a secure environment, where the model reads the prompts and generates responses. The responses are re-encrypted before being returned to the browser. As designed in the documentation, no intermediate components can access the message content.
However, privacy is subject to clear limitations. The template document states that the application server can see visitors' IP addresses, and Nito can access account information, models, request times, and the size of encrypted messages.
Developers still rely on the web code provided, hardware manufacturer attestations, and published software running in that environment. Verification can only identify a specific published software version, but cannot prove that the version is free of vulnerabilities.
This example requires a funded Nito account and an API key. Authentication, rate limiting, and bot protection are left to the developer to implement.
Developers can build agents around verified outputs.
The foundation lists personal assistants, agents handling confidential business data, and private trading strategies as potential applications on the X chain. The foundation states that developers can check the proof status of the signing key before accepting the signature result.
The announcement also outlined additional possibilities, including AI adjudication for market prediction or insurance claims, accepting transaction instructions only from verified machines, and AI-driven governance decisions.
On November 9, Solana outlined key developer priorities. As crypto.news previously reported, its 11 startup ideas include an open AI inference market where users can verify which model processed a request and track provider performance.
Another proposal calls for the introduction of a wallet-based financial assistant capable of understanding holdings and past decisions, and obtaining explicit user approval before executing any actions.
Solana's privacy solutions cover payments and token distribution.
Helius launched Solana Privacy Rings on November 10, enabling encrypted balances and confidential transfers on mainnet. The initial permissionless rings will conceal assets and amounts while keeping sender and receiver addresses visible.
Previously, Umbra and Streamflow launched confidential token vesting on May 26. This integration uses Arcium’s encrypted computation infrastructure to encrypt the allocation amount, recipient address, and unlock schedule.
In the United States, SEC Chairman Paul Atkins discussed financial privacy in a speech in December 2025, stating in his personal capacity that zero-knowledge proofs and selective disclosure can help users demonstrate compliance without revealing their full financial history.


