On-chain Verification
Every agent decision produces a Groth16 proof verified by our vrFy… Anchor program. Light Protocol state compression lets us commit millions of execution traces per day at lamport-level cost.
Deploy autonomous agents as Solana programs. Every decision lands in a 400ms slot, every action costs a fraction of a cent, every state transition is proven by an on-chain SVM verifier. Built for the only L1 where agent economies actually clear.
Three primitives that turn Solana's speed and density into an agent-native execution layer.
Every agent decision produces a Groth16 proof verified by our vrFy… Anchor program. Light Protocol state compression lets us commit millions of execution traces per day at lamport-level cost.
Agents speak the language of Solana: CPI into Jupiter for best-route swaps, Pyth for sub-second price feeds, Kamino and MarginFi for leverage, Drift for perps. One transaction, one slot, atomic.
Model inference runs inside AMD SEV-SNP enclaves, identical hardware Jito uses for its block engine. Signed transactions ship as Jito bundles — atomic, front-run resistant, landed with tipped priority.
A transparent pipeline, deterministic by design, finalized in a single Solana slot.
Declare the agent's model, strategy, CPI targets, and compute-unit budget in a agent.nexus.yaml manifest. Anchor IDL is generated automatically.
The registry program mints a PDA for your agent, stakes its $FORGE bond, and routes compute to the operator closest to your target leader schedule.
Inference runs in an SEV-SNP enclave reading Pyth and Switchboard. The signed instruction set ships as a Jito bundle with a tipped priority fee.
A Groth16 proof of the inference trace is submitted to the NexusForge verifier program and committed to a Light Protocol state tree — auditable by any RPC client.
Autonomous agents don't work on chains with 12-second blocks and $5 gas. They need sub-second finality to react to market state, fees denominated in lamports so strategy margins survive, and parallel execution so a thousand agents don't serialize into a single mempool. Solana is the only L1 that clears all three. Firedancer pushes the ceiling higher. We built on the floor that exists today.
Everything institutional teams need, out of the box.