Describe the workflow. Get a graph you can edit. The AI Composer turns a plain-language description into a workflow on the canvas — a draft you own and change, not a black box you have to trust.
SlackOAuth or Bearer tokenstored once
StripeAPI keystored once
HubSpotOAuth or Bearer tokenstored once
NotionOAuth or Bearer tokenstored once
A sentence becomes a graph. On the canvas, where you can see every step.
Describe what you want to happen — which tool it starts from, what should be read, what should be written — and the Composer assembles it as a workflow on the visual canvas. You get nodes and the edges between them, laid out as a graph rather than as a wall of configuration.
The point is not that a description is faster to type than a form. It is that the result is inspectable. A graph shows you what will run and in what order before anything runs at all, which is a different kind of confidence from watching output appear and hoping it was right.
- Plain language in, an editable workflow graph out.
- Every step visible on the canvas before it runs.
It is a draft, and we mean draft. Review is part of the design, not an apology for it.
What the Composer produces is a starting graph. It is meant to be read and changed: rename a node, drop a step you did not want, point one at a different tool, add the branch the description left implicit. Nothing about it is locked.
That framing is deliberate rather than defensive. A generated workflow that cannot be edited would be a demo; one that opens as a draft you finish is a tool. Correctness still depends on your review, and the canvas exists so that review is a matter of looking rather than of testing.
- Every node and edge is editable after generation.
- Correctness depends on review — the graph is the thing you review.
Ask once, or run it forever. The same description covers both.
Most work starts as a one-off. You ask the assistant for something, it runs the actions across your connected tools, and you get the result. Nothing needs to be built for that.
The Composer is what happens when the one-off turns out to be recurring. The same description that got you the result once becomes a workflow you can keep, schedule, and hand to a teammate — without you having to rebuild it as a diagram from memory.
- No workflow needed to get the first result.
- The description you already used becomes the workflow you keep.
Every kind of key, counted from the registry. The vault holds the credentials behind all 179 integrations.
Built for the teams that run on it. See the vault at work behind a real workflow.
AI Composer FAQ
An editable workflow on the visual canvas, generated from your plain-language description. It is a draft graph: you can rename, remove, re-point, and extend any node before or after you run it.