Security

The vault contracts hold the funds; the agent holds narrow permissions. This page lists exactly what is enforced on-chain and where to verify it.

What can the agent do?

• The agent can move liquidity to a new range within contract-set bounds.
• The agent can harvest and compound fees.
• The agent cannot withdraw funds to any address.
• The agent cannot rebalance more often than the vault's on-chain cooldown.
• Replacing a vault's strategy waits behind a 48-hour on-chain timelock.

This list is the same source of truth shown in the safety card on every vault page.

Audits and reviews

Before deployment, every contract went through multiple independent AI-agent security audits — each a fresh, adversarial review of the full source, with findings fixed and re-reviewed before broadcast. The contracts have not yet been audited by an external security firm. Changes to reviewed code re-enter review before deployment.

Timelock and ownership

Replacing a vault's strategy — the only way to change what code manages the funds — must be proposed on-chain and waits 48 hours before it can execute, publicly visible the whole time. The performance fee is hard-capped at 15% by an immutable constant; below that cap, fee and rail adjustments by the owner take effect immediately.

During beta the owner is a single team-controlled account. Pausing deposits and triggering an emergency exit are deliberately wider: the owner, the guardian, and the agent itself may all pull those brakes, because they only ever reduce risk. None of them can move funds to an external address.

Emergency behaviour

If a safety bound is breached, the vault evacuates all liquidity from the pool and holds both tokens. Deposits stop; withdrawals — always available in-kind — continue working the whole time.