The workflow editor is where you build multi-step agent pipelines visually. It's a canvas with nodes, not a code editor. Here's how to use it.
The canvas
When you create a workflow or open an existing one's editor, you see a vertical pipeline. Start node at the top, end node at the bottom. Steps in between.
Each step is a node representing an agent. Click + between nodes to insert a new step. Click a node to select it and open its configuration panel on the right.
Adding steps
Click any + button to add a step. Pick from your existing agents or create one inline.
Steps execute in order, top to bottom. Each step receives the output from all previous steps as context.
Task overrides
The right panel for each step shows the agent's name and a Task Override field. This is powerful: the same agent can do different things in different workflows.
A general-purpose research agent might have the task override "Research the top 10 companies in the fintech space" in one workflow and "Research recent M&A activity in healthcare" in another. Same agent, different job.
Leave the task override empty to use the agent's default task.
Validation
The editor validates your workflow as you build. A banner at the top shows errors (no steps, deleted agents) and warnings (consecutive duplicate agents). Fix errors before running.
Reordering steps
Drag nodes to reorder them, or use the reorder controls. The execution order always matches the visual top-to-bottom sequence.
Running from the editor
Hit Run in the editor header. The workflow executes and you can watch each step's logs by clicking on its node. Status badges update in real time: pending, running, completed, failed.
Shared volumes
For workflows that pass files between steps, every step has access to
/home/daytona/shared/. Step 1 writes a file there, step 2 reads it.
No configuration needed; the shared volume is automatically mounted
for all workflow runs.
Tips
Start small. Two steps is a workflow. You don't need five steps to get value. Research then report. Gather then analyze. Two focused agents beat one overloaded one.
Use task overrides. They let you reuse agents across workflows without creating duplicates for every variation.
Check the cost. Each step's credit cost shows up in the run detail. If a step is burning more credits than expected, check if it's iterating too many times and adjust max iterations in the agent settings.
Try it
Go to Workflows, create a new one, and build a 2-step pipeline: research then summarize. Run it and watch both steps execute in sequence.