How Data Flows

Every automation runs the same loop: a trigger fires, then a sequence of actions runs, then the run finishes. Knowing this loop helps you reason about what Claude is building when you describe a workflow.

The trigger fires

An automation starts when one of its triggers fires. Most automations have one trigger, but an automation can have several — a webhook for instant updates plus a polling check as a backstop, or two webhooks for different events from the same app, or a manual fire button alongside a scheduled run. The trigger is the event that starts the run: a webhook arrives, a poll finds a new record, a schedule ticks, a form gets submitted, a human kicks it off manually.

Whatever data the trigger produces — webhook payload, polled record, form fields — becomes available to every step below it.

Actions run in order

Steps run one after another. A step that fails halts the run by default, but Claude can add alternative paths when you ask for them — "if step 3 fails, send the bad record to a Slack channel and continue".

Conditions and loops are first-class: you describe the branching or repetition in English, and Claude builds the structure.

Data flows between steps

Each step has access to whatever the steps above it produced. Claude sets up the data flow when it builds the automation. See Field Mapping.

State that has to outlive a single run

Some state has to persist across runs — the last record ID you synced, a token that's good for an hour, a customer-specific config value. APIANT has built-in lookup-table storage for this. You describe what state you want kept and Claude sets it up.

See Data Storage for the full picture.

After the run

Every run is recorded with full step-by-step input and output. You can ask Claude "what did the last execution do?" or "why did automation X fail this morning?" and it walks the recorded run for you.