Flows

Creating and Editing Flows

Build AI-powered workflows by creating Flows that process data, call AI models, and take actions automatically. If you are new to the feature, start with What are Flows? for a quick overview.

Create a Flow

  1. In the sidebar, click Flows.
  2. Click Create Flow.
  3. In the Create New Flow panel, choose one of the available creation options.
  4. The Flow editor opens. Click the Flow name at the top to rename it.

You can also create a Flow from a Product when adding a Capability. It opens the same editor.

Add steps

Steps are the building blocks of your Flow. Each step performs a specific action, such as calling an AI model, fetching data, making API requests, or running code. For a broader map of what is available, see Flow step types overview.

  1. Click the + button between steps, or at the top for your first step.
  2. Browse or search step types in the dialog that appears.
  3. Configure the step in the step card that appears.

Use the search bar in the Add Step dialog to quickly find the step type you need. If your search narrows to a single result, press Enter to add it.

How steps work together

Steps execute sequentially from top to bottom. To pass data between steps, reference an earlier step’s output variable with double curly braces, such as {{summary_result}} or {{ticket_analysis.category}}.

System and Record variables like {{_record.metadata.field}} are also available.

Reorder steps

Drag steps in the left sidebar to change execution order. The Flow runs from top to bottom in the order shown.

Enable or disable steps

Use the enable or disable button in the sidebar to turn individual steps on or off. Disabled steps are skipped during execution, which is useful for testing or temporarily removing a step without deleting it.

Edit a step

Click any step in the sidebar or the main editor to select it, then update its configuration in the step card. Click Save in the toolbar when you’re done.

Delete a step

Click the trash icon on a step in the sidebar to remove it. If other steps reference the deleted step’s output variable, update those references.

Test your Flow

  1. Click the Run button in the toolbar.
  2. The Run Flow sheet opens with several tabs.
  3. Enter your test input and run the Flow.

Results show step-by-step execution details, including input, output, and timing for each step. Click any step in the results to see its full details.

You can test a Flow before saving it. The editor uses your current unsaved steps, which is helpful for rapid iteration.

Save and publish

The toolbar shows Create for new Flows or Save for existing ones. Save your work frequently as you build.

To make a Flow the live version used by your Products:

  1. Click the chevron next to Save to open the dropdown.
  2. Click Publish, or Save & Publish if you have unsaved changes.

From the same dropdown, you can also view past versions, open Conversations for Flows with Agent steps, or view Activity to see who changed the Flow and when.

Published versions are locked. Any further edits create a new draft automatically.

Generate a Flow with AI

Click the Generate button in the toolbar to have AI help you create or modify your Flow’s steps. Describe what you want your Flow to do, and Runtype generates the steps for you.

Error handling

Each step has configurable error handling. Click the error handling icon on a step card to choose what happens if that step fails.

  • Continue on error skips the failed step and continues running the rest of the Flow.
  • Stop on error halts the entire Flow immediately.
  • Use fallbacks tries alternative approaches before giving up. You build a chain of fallbacks that run in order until one succeeds.

Each fallback in the chain is one of three types:

  • Retry re-runs the same model.
  • Different model switches to a backup model, with an optional temperature and max tokens override.
  • Fixed message returns a pre-written reply without calling a model. Use this as the last fallback so a surface never renders an empty reply.

You also choose what triggers the chain. The step errors runs fallbacks when the step fails outright. The reply is empty runs them when a model finishes successfully but returns no visible text — useful for reasoning models that can spend their whole token budget thinking and return a blank answer. You can enable both triggers at once.

Fallbacks are especially useful for AI steps. You can set a backup model so your Flow keeps running if a specific model is unavailable, and end the chain with a fixed message so the step always produces a reply.

Next steps