Building a workflow
1. Open the Agent Builder
Go to Workspace → Workflows and start a new workflow. The right-hand panel is a chat with the Agent Builder — describe what you want automated.
2. Describe the automation
Include the trigger, the steps, and any conditions:
"When someone fills out our contact form (webhook), check if their message contains the word 'urgent'. If it does, post to the #support Slack channel. Otherwise, add a row to our Google Sheet."
The Builder will:
- Propose a trigger (a webhook, in this example).
- Add a condition node to check the message.
- Wire the two branches to a Slack connector call and a Google Sheets connector call.
- Ask you to connect the Slack/Google Sheets accounts if they aren't linked yet.
3. Watch the graph build live
The canvas (center panel) updates as the Builder adds nodes — you're always looking at the real graph that will run, not a description of it.
4. Adjust by chatting
Nothing is locked in after the first draft:
"Also send me an email if it's urgent"
"Change the Slack channel to #escalations"
Every change goes through the same validated tool-calls the Builder uses to construct the graph in the first place — dangling edges, unreachable nodes, or a condition missing a branch are all caught before you can save.
5. Dry-run it
Before saving, ask the Builder to dry-run the workflow, or use the dry-run button in the run panel. Every step with a real side effect (sending a message, writing a row, calling a paid API) is mocked, so you can see the full execution trace safely.
6. Save and set the trigger live
Once the dry-run looks right, save the workflow. If it's a webhook or schedule trigger, it starts listening/firing automatically — see Triggers.
7. Watch it run
Every real run appears in Workspace → Runs, with a step-by-step trace: each node's input, output, duration and status, in the order it actually executed (parallel branches are shown running concurrently, not squeezed into a single timeline).