Give AI agents
rules, not trust.
The on-chain control layer for AI-managed tokenized securities. Every action is checked against your covenant before funds can move.
AI agents are getting wallets.
They should not get unlimited authority.
One key. Unlimited risk.
Wallet permissions are binary. Once an agent can sign, users have little control over what it buys, where it sends funds, or what it discloses.
Bounded authority. Verifiable safety.
Users write an on-chain mandate once. Agents can act quickly inside it and physically cannot cross its boundaries.
Attack the agent.
The covenant holds.
Run real demo scenarios against Covenant #0001. Safe actions pass. Unsafe actions are refused before execution.
Execution receipts prove what happened.
Refusal proofs prove what didn’t.
Covenant Prime does not merely stop unsafe transactions. It creates a portable on-chain record of the attempted action, violated rule, agent, amount, and timestamp.
Agent attempted to transfer 450 USDC to a recipient outside the permitted perimeter.
- Agent
- 0x71C2...9F4A
- Covenant
- #0001
- Asset
- USDC
- Amount
- 450.00
0x2cc2806cd1ccccccccccccccccccccccccccccccccccccccccccccccccccccccOne covenant.
The full asset lifecycle.
Trade & rebalance
Bounded execution across tokenized stocks and RWAs.
Vote & claim
Governance and corporate actions with explicit permission.
Repay & distribute
Manage credit repayments and tokenized cash flows.
Disclose & audit
Permissioned data access with a complete proof trail.
Plug enforceable authority into any agent.
Standard Solidity contracts. Simple action requests. Stable reason codes. Deploy on Arbitrum or any EVM-compatible chain.
Read the architecture// Agent proposes an action
router.proposeAction({
covenantId: 1,
actionType: BUY,
asset: mNVDA,
amount: 200e6
});
// Allowed → execute + receipt
// Refused → immutable proof