Heavy computation executed off-chain, folded into a single proof, and certified on Solana in one transaction. No trusted operator. No blind trust.
AI inference, analytics, aggregation and ranking systems blow past what a Solana program can run inside a transaction. So today they run on a server you're asked to trust.
SP1 emits a succinct proof for every sub-computation. Nova folds them — recursively, instance by instance — into a single SNARK. The verifier certifies that one folded proof in constant time, however much work it represents.
Submit a circuit id, an input blob and a callback. The job queues to an SP1 host pool.
The SP1 v3 zkVM runs the guest program and emits a succinct proof per sub-computation.
Nova / SuperNova folds the sub-proofs into a single Groth16 SNARK of constant size.
The Anchor program runs verify_proof on mainnet and writes a ProofRecord PDA.
Certified facts publish as feeds. dApps read them via CPI, API or SDK.
Many provers do the expensive work in parallel. Solana only ever sees the folded result — a few kilobytes, a single transaction, permanent.
Protocol architecture, the proof engine and verifier integration on Solana mainnet.
The developer CLI, SDK and Compute API for submitting jobs and reading certified results.
Nova folding at scale — constant on-chain cost across unbounded work.
The proof marketplace, AI verification modules and ecosystem composition via CPI.
A permissionless proving network, cross-chain certification and ecosystem grants.