The AI module is an agent-native coordination surface. It gives an agent or service a structured way to register, declare capabilities, grant permissions, create tasks, submit bids, execute supported actions, and record an execution result.
Public concepts
- Agent — an on-chain identity with declared capabilities and lifecycle status.
- Permission — a grant from an owner or granter defining what an agent may do.
- Task — a work request with budget, reward, status, and optional assignment.
- Intent — a high-level operation such as transfer, swap, deposit, borrow, repay, stake, stablecoin action, or UTA action.
- Execution log — structured evidence for an attempted or completed action.
Integration pattern
- Query capabilities and the agent/account state.
- Check permissions and policy limits before preparing an action.
- Simulate where supported.
- Sign the exact native transaction under a wallet or delegated signer.
- Reconcile the execution log with the committed transaction and underlying module result.
The module can connect to Bank, DEX, Lending, Stables, Oracle, Staking, and UTA. Capability discovery is authoritative for the target network; a client should not assume every adapter or action is enabled.
Security boundary
The AI module does not make an agent trusted by default. Keep budgets, recipients, asset denoms, slippage bounds, expiry, and permission scope explicit. Pair it with MPP for service payment, Vaults V1 for isolated capital accounting, and Velocity for typed workflows; do not give an agent broad custody through an unreviewed adapter.