Observer
Helius LaserStream consumer. No signing key. No policy access.
Audit · Autonomous · Private
RISKCLAW
Onchain Policy. Sealed Enforcement.
Treasury operators encrypt a risk threshold once. Three sealed agent zones read, compute, and execute — without any of them ever seeing the plaintext. Squads-signed, Swig-delegated, Arcium-sealed.
100%
On-chain enforcement
0
Plaintext leaked
3
Sealed agent zones
Architecture
Three zones. Three keypairs. One signer.
Privacy in RiskClaw isn't a UX layer — it's a key-separation guarantee enforced both onchain and offchain. The Observer is allowed to read everything but holds no key. The Analyst can ask Arcium "is the threshold breached?" but never sees the answer in plaintext form. The Guardian is the only address that can sign — and only inside the limits that Swig delegation permits.
Analyst
Calls Arcium MPC. Receives only {breached, score}. Never decrypts.
Guardian
Holds the only signing key. Bounded by Swig delegation onchain.
Sealed at rest.
Sealed in compute.
Sealed at execution.
Agents enforce what they cannot see.
How it works
Encrypted policy → MPC check → bounded rebalance
On submit, your drawdown threshold is encrypted via Arcium and stored as a 64-byte ciphertext on the risk_policyaccount. The Analyst queues a threshold check inside Arcium's MPC; only the public answer leaves the circuit. If breached, the Guardian fires a single bounded execute_rebalance via Swig — slippage and notional caps enforced by the program, not by trust.
Colosseum Frontier · Devnet only