Every vault decision,
cryptographically attested.
When our AI agents commit to a hedge, allocation, or rebalance, the decision is proven correct with a zero-knowledge STARK — no trusted setup, post-quantum secure by construction, verifiable by anyone.
How it works
Trace the computation
Each agent decision (allocation percentages, hedge sizing, risk score) is compiled into an execution trace — a matrix of intermediate states the prover walks through.
Commit + prove
AIR constraints on the trace are extended and folded through FRI (Fast Reed-Solomon IOP). Merkle-committed with SHA-256. Fiat-Shamir non-interactivity.
Verify anywhere
Proofs are 10–50 KB. Verification is 50–200 ms off-chain and constant-time — no interaction with the prover, no trust in the AI, no revealing the underlying data.
Security parameters
- Field
- Goldilocks-64 (default) · NIST P-521 prime available
- Commitments
- SHA-256 Merkle trees · Fiat-Shamir non-interactive
- Soundness
- ~180 bits effective (FRI queries + grinding)
- Trusted setup
- None — hash-based commitments only
- Post-quantum
- Yes — no discrete-log or factoring assumption
- Acceleration
- CUDA (CuPy / Numba) with CPU fallback
Post-quantum security comes from the hash-based commitment structure, not field size. Effective soundness of ~180 bits from FRI queries + grinding is well above post-quantum requirements.
Verify a proof
Paste a proof hash (0x…) or transaction digest from the vault's activity log. We'll check whether it's recorded on-chain and return its details.