Overview
AI Workflows let you chain AI actions and external API calls into reusable automations — without code. Use them for things that are too complex for Autopilot but happen too often to do by hand.
Examples:
- "When a 5-star review comes in, post it to our team Slack and tag the cleaner who turned the unit."
- "When a maintenance task gets a 'plumbing' label, create a Trello card for the contractor."
- "On every check-out, generate a summary of the stay and email the property owner."
Anatomy of a workflow
A workflow has three pieces:
- Trigger — the event that starts it (new review, new task, message arriving, schedule, etc.)
- Steps — what to do next: ask the AI to extract something, send a message, call an external API, branch on a condition
- Approval gate (optional) — pause and wait for a teammate to approve before continuing
Each step's output is available to the next step, so you can chain reasoning across calls.
Building a workflow
Open AI Workflows from the side menu. Click + New Workflow, pick a trigger, then add steps. You can:
- Preview a workflow against historical data before turning it on (dry run)
- Test individual steps in isolation
- Pause / resume a workflow without losing state
External integrations
Workflows can call out to external services through the integrations layer. Anything you've connected — Slack, Gmail, Stripe, your CRM — is available as a step.
Approvals and notifications
If a step requires a human-in-the-loop, the assigned teammate gets a push notification (workflow_approval) and the workflow waits for their decision before continuing.
What's next?
- AI Memory - Memories workflows can read from and write to
- Setting Up AI Autopilot - The simpler automation path for guest messages