Pause
One switch in the app. While it is on, your Agent Wallet cannot sign anything and running playbooks are held. Turning it off again is your call alone.
Every limit lives in policy, outside the model. Every action is tested against it before it runs and written down after it runs. Here is what that means in practice.
Stopping should never be harder than starting. Each brake works on its own, and none of them needs the agent's cooperation.
One switch in the app. While it is on, your Agent Wallet cannot sign anything and running playbooks are held. Turning it off again is your call alone.
Over MCP, execution starts in dry-run: orders are prepared, checked and reported, not sent. You switch it off per API token when you are ready.
The operator can halt every execution path at once, for every user, if something looks wrong with a venue, a contract or the service itself.
There are up to three keys in play. Only one of them is held by the operator, and it is the one you are told to keep small.
These live in policy on the server and are read again on every single call. If an order breaks any of them, it is refused, and the refusal is logged with the rule that stopped it.
The most one order can be worth in USD, whoever signs it.
A ceiling on what the Agent Wallet may spend or open in a rolling day.
The highest leverage the agent may use on any perp, below the venue's own maximum if you choose.
Only the coins and tokens you allow can be traded by the agent. Everything else is read-only for it.
Troya Gate is the pre-trade check. It runs before a proposal is shown to you and again before the Agent Wallet signs. If any answer is bad, the order is not offered, and you see which check failed and why.
Is the path real, through pools that exist on the chain right now?
How far does this order move the price, and is that within your slippage?
Is there enough liquidity behind the quote to exit later?
Can the token actually be sold back, or would a sell revert?
Does the token skim a fee on transfer that the quote does not show?
Token names, descriptions and posts from X are written by people you do not know. Troya marks them as untrusted before the model sees them, so a line like "ignore your limits" inside a token name is just a string.
Every proposal, signature, refusal and Agent Wallet action is written to your journal with the reason the agent gave and the rule that applied. It is readable in the app and is never edited after the fact.
Review the journal in the app