The building blocks, and an agent that wields them.
A workflow is a graph of steps with a trigger. Four step families cover the work; the AI steps are where Actuant departs from a conventional flow engine.
Triggers
Start a run on a webhook, a schedule, polling, or manually. Every run carries its trigger context through to the last step.
Actions
Do work in the world: typed HTTP requests and a growing library of app connectors, with credentials kept in an encrypted vault.
Logic
Control flow without code: branch on conditions, loop over collections, filter, and merge results back together.
AI agents
A first-class Agent node that plans, selects tools, loops until the goal is met, and reports its reasoning — not a prompt bolted to a box.
Observability
Every run records step inputs, outputs, and tool calls. Debugging an agent is as clear as reading a stack trace.
Any model, your key
Run each node on the model you choose and bring your own key. A provided model is available instantly for anyone starting out.
The Agent node makes the calls a rule can’t.
Give it a goal and the tools it’s allowed to use. It plans an approach, calls connectors, reads the results, and loops until it’s done — choosing the next step instead of following branches you wired by hand.
And because every plan, tool call, and model response is recorded, you’re never guessing why it did what it did.
- TRIGGERWebhook
stripe.dispute.created
- AGENTPlan
Fetch dispute → summarize → notify
- TOOLhttp.request
GET /v1/disputes/du_1N4Hc…
- MODELllm.summarize
claude-sonnet-4.6 · 1,240 tok
- TOOLslack.postMessage
#risk-ops
Describe it. Actuant builds it.
Author in plain language, refine on the canvas, run on the model you choose.