Agent Workflow
Early Alpha Version
AI Agents are in early alpha. Node interfaces, behaviors, and APIs are subject to change.
The Agent Workflow node lets you build step-based pipelines that chain agents and tools together. Steps can run sequentially, in parallel, or conditionally based on previous results.
Inputs
| Input | Type | Default | Description |
|---|---|---|---|
| Workflow Name | String | My Workflow | Human-readable name for the workflow |
| Description | String (multiline) | — | What this workflow does |
| Steps | Array | — | Array of step definitions (see below) |
| Execution Mode | Enum | Sequential | How steps run: Sequential, Parallel, or Conditional |
| Input Data | Object | — | Initial data passed to the first step |
| Execute | Signal | — | Trigger to start the workflow |
| Max Retries | Number | 1 | Number of retry attempts per step on failure |
| Timeout | Number | 60000 | Maximum time (ms) before a step times out |
Outputs
| Output | Type | Description |
|---|---|---|
| Result | Object | Final result from the last step (or combined results) |
| Step Results | Object | Results keyed by step name |
| Status | String | Current workflow status (idle, running, completed, failed) |
| Current Step | String | Name of the currently executing step |
| Completed | Signal | Fires when the entire workflow finishes |
| Step Completed | Signal | Fires after each individual step finishes |
| Error | String | Error message if the workflow failed |
| Progress | Number | Completion percentage (0–1) |
Execution Modes
Sequential (one after another)
Steps run in order. Each step receives the output of the previous step as input.
Step 1 → Step 2 → Step 3 → Result
Parallel (all at once)
All steps run simultaneously. Results are combined into a single object keyed by step name.
Step 1 ─┐
Step 2 ─┤→ Combined Result
Step 3 ─┘
Conditional (based on results)
Steps include conditions. Only steps whose conditions are met will execute.
Step Definitions
Each step in the Steps array is an object with:
| Property | Type | Description |
|---|---|---|
name | String | Unique name for this step |
agent | Object | The agent instance to use for this step |
message | String | The prompt/instruction for this step |
condition | String | (Conditional mode only) Expression that must be truthy to execute |
Usage Tips
- Use Sequential mode for multi-step reasoning: "Research → Analyze → Summarize"
- Use Parallel mode when steps are independent: "Translate to Spanish + French + German"
- The Progress output is useful for showing a progress bar during long workflows
- Connect Step Completed to update the UI after each step finishes
- Max Retries helps with transient API errors — the step will retry before failing