You hold the keys
You connect a wallet you control. There is no mode of this product in which we take custody of your assets, and no upgrade path that introduces one.
Most of what goes wrong in this category goes wrong because somebody else was holding the assets. So we do not.
You connect a wallet you control. There is no mode of this product in which we take custody of your assets, and no upgrade path that introduces one.
Removing a wallet's authority is one action in the console and takes effect immediately. You do not have to ask us, and there is no notice period.
The agent's authority is to trade inside your limits. It has no route to send your assets anywhere, including to us.
An autonomous trader that respects its limits voluntarily is one bug away from not respecting them. Every boundary here is checked before capital moves, outside the code that wants to trade.
Checked before each commitment, across the whole fleet.
Reaching it ends the day. The agent cannot vote to continue.
A fill outside your tolerance is refused, not accepted and explained.
How much may be in flight at once, however many agents are live.
There is no code path by which the agent raises its own ceiling.
The stop is the control you will care about most on the worst day, so it is the one with the fewest moving parts.
Not queued, not scheduled, not subject to a confirmation flow.
Every agent and every swarm, not the one you were looking at.
Phone, laptop, anywhere you can open the console.
Who stopped it and when, written to your record.
Custody is the big risk. These are the ordinary ones, handled the ordinary way.
Signing material is held in environment configuration, never in source. There are no default passwords and no fallback keys — a missing secret stops the service rather than quietly using a published one.
Account data is scoped per account at the database level, so a query that forgot its filter returns nothing rather than somebody else's activity.
Internal instructions between our own services are signed over their exact contents, so a message that was altered in flight is refused rather than acted on.
Identity can be federated through your organisation's provider, so joiners and leavers are handled where that already happens.
Viewing, operating and administering are separate rights. The administration console is not reachable from the member app at all.
Machine credentials are named, scoped and individually revocable, and every use is logged against the key that made it.
The dangerous failure for an automated trader is not an outage. It is continuing to act on information that has gone stale — which looks exactly like working normally until the losses arrive.
If the agent cannot confirm the state of the market or of your limits, it does nothing. Doing nothing is always available and never wrong by more than an opportunity.
An agent that has not reported in is stood down rather than assumed healthy. We would rather tell you it stopped than tell you it is fine without knowing.
Outages and degradations land on your record like everything else, so what happened during one is a fact rather than a reconstruction.
We do not sell your activity, we do not trade against it, and we do not aggregate it into a signal for anyone else. Your record exists so that you can read it and take it with you.
Security pages in this category tend to imply more than they say. These are the things we are not telling you.
This is trading. It loses sometimes. The loss floor bounds how much in a day; it does not make losses impossible.
Software has bugs and markets have conditions nobody modelled. What we can tell you is how the system behaves when it is uncertain: it stops.
Where we hold a certification we will name it and show you the report. Until then we would rather say nothing than imply one.
Connect a wallet you control and set limits the agent cannot widen. Everything else is its job.
Non-custodial. Stoppable at any moment. Trading involves risk of loss.