Documentation

Workflows

Workflows let you automate tasks and processes by combining multiple steps. Instead of doing things manually, you can create workflows that run automatically or on demand.

What Are Workflows?

A workflow is a series of steps that execute in sequence. Each step can:

  • Ask an AI agent to do something
  • Call external tools through MCP servers
  • Process data and transform it
  • Make decisions based on conditions
  • Wait for human approval

Workflows help you save time by automating repetitive tasks and ensuring consistent processes.

Creating a Workflow

  1. Go to Workflows in the main menu
  2. Click Create workflow
  3. Give your workflow a name and description
  4. Define your workflow steps
  5. Set up triggers (manual, scheduled, or event-based)
  6. Save and test your workflow

Workflow Steps

Each step in a workflow performs a specific action. Common step types include:

  • Agent Steps: Use AI agents to process information, answer questions, or generate content
  • Tool Steps: Call external services and tools through MCP servers
  • Human Approval: Pause the workflow and wait for someone to approve before continuing
  • Subworkflow: Run another saved workflow as a step and reuse its results
  • Loop: Run a child workflow once per list item or a fixed number of times and collect aggregated results
  • Conditional Steps: Make decisions based on data from previous steps

Triggers

Workflows can be triggered in different ways:

  • Manual: Run the workflow whenever you want
  • Scheduled: Run automatically on a schedule (using cron expressions)
  • Event-based: Run when something happens in your system
  • Webhook: Run when an external system sends a request

Using Workflow Outputs

After a workflow runs, you can review:

  • Per-step outputs — everything each step produced (expand a step on the execution detail page).
  • Workflow outputs — the small set of named results you configured at the bottom of the workflow editor (workflow_outputs). At termination, each configured output is copied from the matching step field when that value is available (including early exits). Outputs whose source step never ran or did not produce the field are omitted.

Workflow outputs are useful when you want a stable “answer” from a run — for example the ID of a document that was created, or a workspace that was updated. In execution history, values like document_id and workspace_id are shown as links you can open in the app.

You can view execution history to understand what happened and debug issues.

Error Handling

Workflows include error handling so they can recover from problems:

  • Retry: Automatically retry a step if it fails
  • Skip: Continue to the next step even if one fails
  • Stop: Stop the entire workflow if a critical step fails

You can configure how each workflow handles errors based on your needs.

Workflow Expressions

Workflows support expressions to work with data dynamically. You can:

  • Reference outputs from previous steps
  • Access workflow inputs
  • Use conditional logic
  • Transform and combine data

This lets you build flexible workflows that adapt to different situations.

Next: Workflow authoring guide

If you’re building real workflows (templates, step configuration, chaining outputs into later steps, MCP tool arguments, conditions, approvals), start here:

Best Practices

  • Start simple: Create basic workflows first, then add complexity
  • Test thoroughly: Run workflows manually before scheduling them
  • Monitor execution: Check the execution history to ensure workflows are working correctly
  • Use meaningful names: Give steps and workflows clear, descriptive names
  • Document your workflows: Add descriptions explaining what each workflow does and why