Skip to content

Start here

Create a workflow

Start a blank workflow and build a complete source-to-output process.

Start with a blank workflow when you need a process that does not fit an existing template. Define the smallest useful path first, then expand it after examining real output.

Before you begin

Identify the input, the expected output, and any external actions the process should perform. Create the necessary connections and confirm access to the documents or records you want to use. For an app workflow, establish which settings should come from app parameters instead of being fixed in the definition.

Build the workflow

  1. Open Workflows and start a new workflow.
  2. Choose the blank-workflow option.
  3. Give the workflow a descriptive name, such as “Prepare supplier documents for review.”
  4. Choose a trigger. Use Manual while developing the process.
  5. Select Add job, choose a source, and complete its configuration.
  6. Add processing steps and connect each to the data it needs.
  7. Add a destination or an appropriate terminal step.
  8. Validate, save, and run a small example.

The picker separates Sources, Processing, and Destinations. Processing choices can remain disabled until a source exists. Test Data is useful when you want to design record transformations before connecting a live source.

Check completion

Every step needs a unique job ID, a handler, valid settings, and valid dependencies. A canvas that looks connected can still have a type mismatch or a missing required setting. Validation checks the definition; a real run checks the behavior with your inputs and permissions.

Name steps for their role rather than their position: read_documents, extract_fields, and save_results remain understandable when the graph changes. Moving a card does not change its dependencies.

Use Save to persist the definition. Save & Run saves and starts execution, so inspect destination and model settings before choosing it. A new workflow without a successful baseline runs in full mode even if incremental processing is its intended default.

For the next steps, read configuring steps, validation, and monitoring runs.