Tria Agentic Trading Lets AI Agents Trade Without Losing Custody

Tria, a self-custodial neofinance platform serving more than 600,000 users, has launched Tria Agentic Trading, a feature that lets users connect AI agents such as Claude, ChatGPT, Codex and Cursor to execute cryptocurrency trading strategies automatically, according to a company announcement published on October 30, 2026.

Powered by the Tria Trade SDK, the feature enables perpetual futures trading on Hyperliquid and Decibel directly through users' existing Tria accounts, without transferring funds to a separate AI-controlled wallet. Users define the conditions for entering and exiting positions, and authorize an agent to execute supported trades when those conditions are met.

A trader could, for example, instruct an agent to monitor the price of Bitcoin, open a position at a specified entry point, and close it when a predetermined target is reached, removing the need to watch charts continuously and place each order by hand, according to the announcement.

The launch arrives as AI plays a growing role in how retail investors research markets and develop strategies. An April 2026 survey by Investing.com, cited in the announcement, found that 62 percent of United States retail investors surveyed had used AI to inform investment decisions, yet only 4.5 percent of those who had experimented with AI reported using automated trading algorithms. Tria positions Agentic Trading as a way to close that gap for everyday crypto traders.

Trade-only keys keep custody with the user

The key design choice in Agentic Trading is permission-based access. Users authorize agents through dedicated trading keys that allow them to open and close supported positions but cannot be used to withdraw funds, according to the announcement.

Traders can set expiration periods for these keys, monitor positions opened or closed by their agents within the Tria app, and revoke trading access whenever needed. Trades are executed through Hyperliquid and Decibel based on user-defined instructions, while custody and account permissions remain under the user's control.

That distinction matters because it separates execution from custody. An agent that can trade is not the same as an agent that can move money out. The trade-only key model is an attempt to give agents useful autonomy inside a bounded permission set, rather than handing them the keys to the account itself.

Tria chief executive Vijit Katta said traders in the company's private beta had been asking for the capability, and that the company is delivering it while continuing to build out the platform. Katta added that traders increasingly want to use AI to execute strategies as well as analyze markets, through tools they already know, while retaining control over funds and trading permissions.

Agents move from analysis to execution

Tria Agentic Trading was developed in response to requests from strategy-focused traders during Tria's private beta, according to the announcement. The pattern reflects a broader shift in how agents are being deployed: first as research assistants that summarize markets, then as executors that carry out multi-step instructions under human-set constraints.

That shift raises practical questions that go beyond any single platform. Who sets the strategy, who monitors the agent, and how quickly can a user pull the plug when conditions change? Tria's answer is built into the product mechanics: user-defined entry and exit conditions, expiring keys, in-app position monitoring, and revocation at any time.

The risks remain real, and the announcement states them plainly. AI-assisted trading does not guarantee profitable outcomes. Trade-only permissions prevent agents from withdrawing funds, but they do not prevent trading losses, and perpetual futures trading involves substantial financial risk, including the potential loss of trading collateral.

To get started, users select a trading venue, Hyperliquid or Decibel, within the Tria app, create a trading key for that venue, configure permissions, and follow the integration instructions for their preferred AI tool. Once connected, they can automate supported strategies and track trading activity directly through the app.

The broader agent economy is wrestling with the same permission problem in other domains. Recent coverage of agent identity and governance tools shows enterprises asking how to grant agents just enough access to do useful work without exposing the whole system. Tria's trade-only key is one consumer-facing answer to that question, and its reception among private-beta traders turned early adopters will be a useful signal for how far users are willing to let agents act on their behalf.

For now, Agentic Trading is a narrow but telling step: agents are no longer confined to suggesting trades. With bounded permissions and revocable keys, they can place them, while the user keeps custody of the funds.

Related coverage: Casap AI agents raise funds to fight fraud and Waters AI Link brings agents into the lab. Sources: Chainwire announcement via TradingView and SolarWinds on agentic enterprise systems.