The workflow editor

The editor is a canvas. You build a workflow by adding task nodes and connecting them; each node is one action against one service.

Task nodes

Open the task library and choose a task to add it to the canvas. A node’s form holds its inputs. Its output port hands the result to whatever it connects to.

A node’s form is generated from the task’s own contract, so the fields you see are exactly what that task accepts — and a field whose value should come from an earlier step takes a reference instead of a value.

Connections

Drag from one node’s output port to another node’s input port to connect them. The connection is data flow: the upstream task’s result is offered to the downstream task as its input.

Execution follows the connections, not the arrangement on the canvas. A task runs once everything it depends on has finished.

Groups and loops

A group holds related tasks and runs them as one step. Its result becomes available to later tasks when all of its tasks have finished, so a task inside the group cannot refer to the group’s own output.

A loop runs its body once per item in a list. Each pass is its own execution of the body, with that item as its input.

Branches

A switch chooses one path based on a value. A guard stops the run when its condition is not met, before the tasks after it do any work. Both are tasks you add from the library like any other.

Building with AI

Build with AI composes a workflow from a plain-English description and places it on the canvas, where you can edit it like one you built yourself. A conversation is charged once, not per message. Check the result before running it: it is a draft, not a guarantee.