Reactor Review: Should You Let an AI Agent Trade?

Reactor is an AI-assisted interface for building trading agents, monitoring them in a simulated environment and sending live orders through connected venues. You can describe a strategy in a sentence or assemble indicators on a canvas, joining conditions with AND or OR. The site lists 18 indicators and shows examples using RSI, EMA and MACD signals. Agents make decisions on live prices while using virtual money in the sandbox, and Reactor says each decision and fill is logged.

The sandbox starts with a $10,000 virtual balance. After 14 days, an agent can graduate to a wallet of its own; you can also publish a performance page with metrics such as profit and loss and drawdown. Reactor names Hyperliquid, Aster and Lighter for trading. The site also presents swapping, staking and other interface areas, but does not explain their full scope. Treat the displayed agent rankings and sample returns as sandbox records, not evidence that an agent will make money live.

Reactor does not publish a platform fee schedule. Its swap interface shows an example Ethereum-to-Arbitrum transfer with a quoted network fee of $0.42 and an estimated time of about two minutes, but that is a sample quote, not a fixed charge. The interface says it will show the rate, time and network fee before a swap is confirmed.

Live trading can carry costs beyond Reactor's own charges: venue trading fees, funding payments on perpetuals, price slippage and network fees may apply. Reactor does not provide a consolidated breakdown of those costs or say whether it adds a spread or execution charge. The interface includes sample leverage figures, but those are not a promise of what every market or account can use. Check the order preview and the relevant venue's terms before confirming a live trade. If you cannot see the total cost and how an order will be routed, keep the agent in the sandbox.

Reactor says live orders leave from a connected wallet, which points to wallet-based trading rather than simply depositing funds into an account. That alone does not establish the full custody model: Reactor does not explain wallet permissions, key handling or what control it has over an agent's trading wallet. Read the connection and signing prompts carefully, and do not approve permissions you do not understand.

Reactor does not publish an audit, a licence, named team details or a track record of security incidents in the product information. The displayed REACT token sale also lacks published sale terms and a clear explanation of token utility. These omissions are meaningful risks, not proof that the product is unsafe. Before using real funds, confirm how you can revoke access, what an agent can do with its wallet and how you can stop it. Start with the sandbox; do not treat a visible profit record as a substitute for independent security assurance.

You can build and inspect agents without a wallet, which makes it possible to explore the strategy workflow before signing in. Reactor describes a plain-language prompt that becomes editable signal nodes, plus a canvas for combining indicators. The sandbox logs decisions and uses live prices with virtual funds. A connected wallet is needed to sign in for live trading, and the product names Hyperliquid, Aster and Lighter as venues.

This setup may be useful if you want to watch an agent's reasoning trail instead of handing it a strategy and seeing only an end result. The site also lists several AI model providers, including Anthropic, OpenAI, Grok, Qwen and others; it does not explain which model handles a particular task or whether model use adds a cost. The interface includes a swap flow and a trading terminal, but Reactor does not publish details on mobile support, service speed or full chain coverage. Test the controls and review logs before considering a live agent.

The main limitation is that Reactor's most appealing evidence is simulated. Sandbox prices may be live, but virtual fills do not establish that a strategy will get the same execution, costs or results with real money. Markets move quickly, automated rules can behave unexpectedly, and leveraged perpetual positions can be liquidated. A 14-day test is a useful checkpoint, not a guarantee of safety or profitability.

The product also leaves important questions open: platform charges, the full custody and permission model, audit status, licence, team, token-sale terms and complete chain support. Its example swap quote and agent performance figures should not be mistaken for standard terms or expected returns. If you proceed, keep the sandbox separate from live trading until you understand the wallet setup and venue costs. Use only money you can afford to lose, and make sure you can pause or revoke an agent before funding it.