A workflow is a visual diagram: nodes do work, arrows connect them, you save it, and it runs. This walkthrough builds a simple workflow with four nodes, a trigger, an AI step, an approval, and an action.
#What a workflow looks like
Every workflow starts with a trigger node and reaches a final node along the way. Common shape:
- Trigger: what starts the run (manual click, webhook, schedule, or an integration event)
- AI step: does something that needs judgment (classify, summarize, draft)
- Approval: pauses the run and shows you the AI’s proposal before the next step
- Action: runs the approved step (send email, create record, hit an API)
You can add conditions, loops, and delays in between as needed.
#Walk through it
#1. Start a new workflow
From the dashboard, click New Workflow. The builder opens to a blank canvas.
#2. Name it
Give it a clear name like “Triage support email” or “Follow up on new lead.” You’ll thank yourself later.
#3. Add a trigger
The first node is always a trigger. Pick one:
- Manual Trigger: you click “Run” yourself. Good for testing.
- Schedule: runs on a cron schedule (every hour, daily at 9am, etc.). Timezone-aware.
- Webhook: runs when an external system POSTs to a unique URL. Used for HTTP integrations not handled by the catalog.
- Integration Event: runs when a connected integration emits an event (Gmail message received, Stripe charge succeeded). Pick the integration and the event.
- Data Change: runs when a row is added to a workspace table or a bound form is submitted.
- Inbound Email: runs when an email arrives at your workflow’s inbound address.
You can attach more than one trigger to the same workflow if you want.
#4. Add an AI step
Click + after the trigger and pick AI. AI steps run a prompt against a model and return text or structured data.
Configure:
- Prompt: what you want the model to do, e.g. “Classify this email as
urgent,normal, orspambased on subject and sender.” - Model: Claude or GPT models from the dropdown. Cheaper models for classification, more expensive models for drafting.
- Output: choose Text for free-form output or JSON with a schema if you want structured fields downstream.
The AI step’s output is available to every node after it.
#5. Add an approval
Click + and pick Human Review. This pauses the run and queues a proposal for you to review.
Configure:
- Prompt: what the reviewer (you) sees, e.g. “Send this reply?”
- Approval type: Approve / Reject, Multi-choice (custom buttons), or Free Text (open-ended input)
- Priority: Low, Normal, High, or Urgent. Affects queue ordering and notification urgency.
- Timeout (optional): what happens if you don’t respond in N hours
When the run hits this node, it pauses. Paused runs cost nothing: you’re never charged for an approval sitting in your queue.
#6. Add an action
Click + and pick an action. Built-in actions:
- HTTP Request: call any REST API
- Send Email: send from Rills (with optional attachments and templates)
- Custom Code: JavaScript or Python in a sandbox
- Transform Data: reshape data between steps
Plus every action exposed by the integrations you’ve connected: send Slack message, create Notion page, refund a Stripe charge, and so on.
Reference data from earlier nodes using {{variable}} syntax. The builder shows what’s available.
#7. Save
Click Save. The workflow is live. If you used a manual trigger, click Run to test it.
#If a blank canvas feels intimidating
Use a template instead. Click Templates in the workflow builder to browse pre-built workflows for email triage, lead follow-up, meeting scheduling, invoice processing, and more. Install with one click and customize.
If you can’t find a template, open the chat and ask Ford for help. He’ll be able to suggest templates or build a custom workflow for you after asking a few questions.
#Next
Review and approve the proposals your workflow creates.