The live agent is Claude
The five strategies here are deterministic templates you can read line by line. When you go live, the rebalance itself is drafted by Claude — Anthropic's AI model — running as Claude Code in your own terminal.
Claude proposes an order, a deterministic guard re-checks it against your risk policy, and nothing is sent until you sign. The agent never holds your key.
How the guardrails workRules you can read, not a black box
Each template is a short, fixed rule set with bounded settings. Agents only buy stock tokens with USDG and sell back to USDG. Never leverage, never shorting.
Real Chainlink SPY prices on Robinhood Chain. Fills at the next update's price, with pool fee and a price-impact allowance. Simulated results, not a forecast. Open in the studio →
Your vault, your limits
Each live agent trades from its own vault. The vault may only swap inside the limits written into the contract. It can never withdraw. Only your wallet can, at any time.
The example on the left uses the real limits of the live mainnet vault: trades that break a limit are refused on-chain, whoever sends them.
See every limitFour steps. Every one is checked.
The agent decides on a new oracle price. The vault re-checks everything on-chain before any swap, and refuses if anything is off.
On each new Chainlink price it applies its rules and records BUY, SELL or HOLD with a reason.
Off-chain, a deterministic guard re-checks the order against your risk policy. You approve it.
On-chain: oracle price, 30-minute TWAP, weight cap, daily turnover and the drawdown breaker. Then the pool swaps.
Every decision and every swap is written down, so you can see exactly why the agent acted.
Loading the latest paper decision…
Is this real money?
On Mainnet, yes: real USDG and stock tokens on Robinhood Chain. The Testnet tab uses valueless test tokens.
Who holds the private key?
You do. This site is static and every transaction is signed by your wallet. The AI agent never holds your key, and this site never asks for one.
Has it been audited?
By AI only (three rounds, every finding reproduced and fixed). No human audit firm has reviewed it, so use small amounts.
Is ANT AGENTS made by or affiliated with Anthropic?
No. The live trading agent runs on Claude, Anthropic's AI model, through Claude Code — but ANT AGENTS is an independent project, not affiliated with, endorsed by, or audited by Anthropic.
Is this an official Robinhood product?
No. It is an independent project using public contracts on Robinhood Chain, Chainlink and Uniswap, with no affiliation to any of them.
Ready to open your vault?
Read-only until you connect. Start small on mainnet.