A sentence in, transactions out.
You describe an outcome. What comes back is a numbered plan you read before anything spends. This page is what happens in between, and why each part is shaped the way it is.
You describe the outcome, not the steps
Type what you want in ordinary words. The agent works out which network, which market, which pool, what to approve, and the order everything has to happen in so money is not stranded halfway.
In most real requests the compiled plan contains something the person never mentioned: a chain crossing, a parallelisation, or a pull from their own wallet instead of a deposit. That is the part actually doing work.
It compiles to a plan, and saving one spends nothing
You get a numbered list of steps. Each names the operation it uses, the network it lands on, the amount, and where the value goes next. Every amount appears twice — as the raw integer the machine uses, and as “0.5 ETH” in plain units. Read the plain-units column: a plan where every amount is a thousand times too small is consistent with itself and passes every other check.
Saving a workflow runs nothing. Running it is a separate, deliberate press.
The list of things it can do is closed
The agent selects from a catalogue of operations across a handful of protocols and three networks. It cannot construct a call to an address that is not in it — not because it is told not to, but because there is no code path that would let it.
A plan naming anything outside the list is refused when it is written, rather than attempted and failed halfway. That is the safety property, and it is also why a mistake costs thirty seconds instead of gas and a half-finished position.
And it keeps going without you
A workflow can start on a schedule, at a price, or when another workflow finishes. Nothing is agreed in advance: each firing rebuilds and re-prices the plan against the market at that moment, so a schedule set up in March does not carry March’s prices into June.
A run survives you closing the tab, and survives us shipping an update mid-flight, because progress lives in a database rather than in the memory of a running process.
Bring one strategy you are tired of running by hand.
We onboard a small number of funds and operators each month, and we sit with each one to build their first loop.