Core concepts
You build a workflow on the canvas. Dagweave turns it into Argo Workflows YAML, and it can import Argo YAML back onto the canvas.
The canvas
Section titled “The canvas”The canvas is a graph of steps. An edge means the second step waits for the first. You add steps from a searchable list and fill in a form for each one. No YAML typing.
Steps have typed inputs and outputs, called ports. An edge from an output to an input passes data as well as order: a value (a parameter) or a file (an artifact). The canvas won’t let you connect ports that don’t match, or draw a loop. See Wiring data and nesting.
A workflow can contain another workflow, and you edit its steps the same way.
A step that calls a WorkflowTemplate shows Open ↗. It opens the WorkflowTemplate
read-only so you can see its steps, and any it calls in turn. A workflow imported
with workflowTemplateRef opens the same way.
You can also:
- Give a step a
when:condition so it runs only when that’s true. - Repeat a step over a list expression (
withParam), a fixed list (withItems) or a range of numbers (withSequence). The current item is{{item}}. See Parameters and loops. - Set a step’s resources, environment variables, node selector, retry policy
and timeout. Anything else in the pod spec goes in
podSpecPatch. - Undo with Ctrl+Z (Cmd+Z on a Mac). Redo with Shift+Ctrl+Z (Shift+Cmd+Z) or Ctrl+Y. Deleting a step also shows an Undo button. Text fields and the YAML editor have their own undo.
Dagweave checks the graph before it compiles, so a broken graph gives you an error, not bad YAML.
Saving and closing
Section titled “Saving and closing”Dagweave saves your changes a second after you stop editing. Just opening a workflow never saves it.
If someone changes the workflow through the API or MCP while you have it open, Dagweave won’t save over their changes. You’ll see a banner: Reload to get their version, or Save mine over it to keep yours.
To stop editing, click Close next to the workflow name, or choose Close workflow from the workflow menu. You get an empty canvas, and nothing is saved. If you have edits that haven’t saved yet, Dagweave asks first.
How it compiles
Section titled “How it compiles”The output is plain Argo Workflows YAML. Dagweave adds one annotation,
dagweave/step-types, so the next import reads each step the same way. Argo
ignores it, and you don’t install anything extra on the cluster.
Tasks and templates take their step’s name. fetch rows becomes fetch-rows.
An unnamed Script step becomes script, and the next one script-2. Imported
steps keep their names.
Pick the kind under the gear menu, Workflow settings, Output kind:
- Workflow is a single run.
- WorkflowTemplate is a named definition you submit or call from other workflows. New canvases start as this.
- ClusterWorkflowTemplate is a WorkflowTemplate for the whole cluster.
- CronWorkflow runs on a schedule.
Steps of the same type and config shape share one template in the output, so the YAML doesn’t repeat itself.
To see it, click the caret beside Run and choose Get YAML. Copy or
download it, then kubectl apply or argo submit it like any Argo manifest.
Import and round-trip
Section titled “Import and round-trip”To start from your own YAML, click Import it on an empty canvas, or choose Import from the menu beside the logo.
If you import a manifest, change one setting and export it, the Git diff shows
that one change. Comments and formatting are lost, key order and quoting can
change, and a steps workflow comes back as a dag.
Import and round-trip has the details.
Where runs happen
Section titled “Where runs happen”The Dagweave app doesn’t run your steps. They run:
- On your cluster’s Argo Workflows, when you apply the YAML yourself.
- On your cluster’s Argo Workflows, through the connector. Dagweave submits the run and collects status, logs and artifacts. You can start runs with triggers and follow them in the Runs panel. Needs Production or Business.
- On your laptop, with Dagweave Local. Local is free on its own. Sending runs to it from the app needs Developer or above.