YieldRails is an AI-driven treasury manager for DAOs, startups, and onchain organizations holding stablecoin liquidity on Arbitrum.
Many teams keep idle USDC in wallets or evaluate DeFi opportunities manually. This creates missed yield, operational friction, and unstructured risk. At the same time, giving an AI agent full control over treasury funds is unsafe. One bad recommendation, hallucinated action, or manipulated prompt should not be able to put organizational funds at risk.
YieldRails solves this through bounded agency: the AI can analyze, recommend, and act, but only within limits enforced directly onchain.
Organizations define treasury rules such as:
- Approved DeFi strategies
- Maximum exposure per protocol
- Minimum idle balance
- Per-action limits
- Risk tolerance
- Human approval thresholds
These rules are stored and enforced by vault and policy smart contracts. If an action breaks the policy, the contract rejects it automatically.
The AI agent can suggest treasury actions, explain the reasoning, estimate yield, show risk level, and prepare a transaction preview. But the final execution must always respect the organization’s onchain rules.
A DAO may allow the agent to move up to 2,000 USDC into a whitelisted lending strategy, but reject any action that exceeds exposure limits or reduces idle liquidity below the required threshold.
The agent can propose and act, but the contract sets the limits.
The buildathon MVP will include:
- Treasury dashboard
- Vault smart contract
- Policy enforcement contract
- AI recommendation layer
- Approval queue
- Tiered execution modes
- Transaction preview
- Onchain audit trail
The AI agent recommends actions, but every action requires human approval before execution.
Small actions can be executed automatically if they stay within predefined onchain limits.
A future mode where the agent can operate with broader autonomy under stricter aggregate caps.
Arbitrum provides a strong environment for DeFi, stablecoin liquidity, and agentic applications. YieldRails is designed to show how AI agents can safely interact with DeFi infrastructure while preserving transparency, user control, and smart-contract-enforced risk limits.
The key demo will show:
- A treasury with idle USDC.
- The AI agent analyzing available DeFi options.
- A valid recommendation passing the policy check.
- An out-of-bounds recommendation being rejected automatically by the smart contract.
This demonstrates the main principle of YieldRails:
AI can propose and act, but it cannot exceed the rails.
This project is currently in early buildathon development for the Arbitrum Open House London Buildathon.