Tria’s new agentic-trading feature makes a familiar security distinction more visible: software can be prevented from withdrawing money while retaining enough authority to lose it through trading. The company announced the feature on October 30, connecting compatible agents to perpetual-futures venues through a user’s existing account. The operational question is how narrowly that trading authority can be defined and supervised.
TL;DR
- Tria announced agentic perpetual-futures trading on October 30.
- Dedicated keys permit trading without withdrawal authority.
- Revoking access does not itself close positions or cancel every order.
A trading permission is still a financial permission
According to Tria’s announcement, dedicated keys can open and close supported positions on Hyperliquid or Decibel without granting withdrawal access. Users can set an access period and revoke the keys. The company also says the documented connection uses live funds, so this is an execution tool rather than a paper-trading demonstration.
The distinction matters because custody and economic exposure describe different things. Retaining the account’s master key can stop an agent from transferring assets to an outside wallet. It does not stop the agent from opening a position that moves against the account or repeatedly paying trading costs.
Imagine an agent given permission to trade a small market signal. If its interpretation changes, repeated orders could accumulate a much larger exposure than the owner intended. The account could remain self-custodial throughout. Preventing that outcome would require limits on position size or order behaviour, not merely a prohibition on withdrawals.
Credentials need a lifecycle
Tria’s credential documentation describes separate agent and delegate keys for the two venues, with revocable trading authority. This separation helps avoid handing an automation tool the master credentials that control the account itself. It also makes the access relationship something the owner can end independently of the strategy.
For any integration of this type, the lifecycle has several stages: granting authority, storing the credential, using it, rotating it and revoking it. A safe creation process is only one part of the problem. Logs, exported configuration and debugging tools can accidentally expose a key after it has been created.
The same reasoning applies to expiry. An expiration date places a boundary on future authorised activity. It does not explain what happens to orders submitted before that boundary. Account monitoring therefore needs to cover both the credential and the market state it has already created.
Stopping an agent and exiting a position differ
Tria specifically instructs users to inspect open orders and positions after revocation. That is a useful distinction for automated trading generally. Disabling the software’s ability to submit new instructions does not necessarily instruct the venue to unwind every existing commitment.
A hypothetical stop procedure might therefore have two independent confirmations. The first confirms that the key can no longer act. The second verifies the desired state of orders and positions. Combining both into an ambiguous green status indicator would hide information precisely when the operator most needs it.
Execution records matter for the same reason. An agent’s statement that it closed a position is not equivalent to a venue confirming that the order filled. A rejection, partial fill or disconnected service can leave exposure behind while the agent has moved on to the next task.
Permission design is only one part of the product
The feature makes it easier to connect software decisions to real markets, but it does not establish that the decisions are sound. A model can misunderstand a prompt, rely on stale data or respond incorrectly to an unexpected market condition. Restricted credentials reduce certain consequences without validating the strategy.
For developers, the useful measure is therefore the complete control loop: authorised intent, submitted action, confirmed outcome and remaining exposure. Each stage should be visible enough to reconstruct what happened. That is a stronger operating model than assuming the agent’s final message is the trading record.
Tria’s launch illustrates where agentic finance is becoming concrete. The new interface connects natural-language instructions with executable market permissions. The quality of the result will depend on how clearly that authority is bounded and how reliably the actual account state is checked afterward.
Written by the News Desk and edited by Samuel Rae.
Leave a Reply