> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cyberwave.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Workflow assistant

> Describe a recurring task and the workflow assistant drafts or edits the workflow: triggers, models, conditions and robot commands. You review every change.

The **workflow assistant** turns a recurring task into automation. You describe the behavior, for example *"inspect every pallet and alert me if one is damaged"*, and it drafts a [workflow](/feature-reference/workflows): the trigger, the models, the decisions and the robot commands. It can also change an existing workflow node by node. Every change is shown as a proposal before it is applied, and applying a draft doesn't activate the workflow.

Use it when the job should keep running on its own. For a one-off action ("move to waypoint A now"), use the [control agent](/ai/control-agent) instead.

## Use it in the workbench

* **Create or edit a workflow:** select the workflow in the workbench, then use **Prompt** or the right-hand **Assistant** tab. Click the assistant's title to switch between the **Environment assistant** and the **Workflow assistant**. The workflow option needs the workbench open.
* **Configure one node with AI:** in an editable environment workflow, **Configure with AI** opens the workflow assistant for the selected node. Edit the request, choose **Plan edit**, review the proposal, then **Review and apply**.

## Use it from Python

The workflow assistant works in steps, so you can inspect each one before anything changes:

| Step                      | Call                                                                            | Changes anything?                                                                         |
| ------------------------- | ------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- |
| Preview the strategy      | `cw.agents.workflow.preview(env_uuid, prompt)`                                  | No                                                                                        |
| Plan the workflow         | `cw.agents.workflow.plan(env_uuid, prompt)`                                     | No                                                                                        |
| Set up and draft          | `cw.agents.workflow.setup_and_draft(env_uuid, prompt, confirmed_actions=[...])` | Yes: applies the setup actions you confirmed and drafts the workflow (private by default) |
| Edit an existing workflow | `cw.agents.workflow.apply_plan(env_uuid, workflow_uuid, prompt)`                | Yes: a constrained edit to that workflow                                                  |

```python theme={null}
from cyberwave import Cyberwave

cw = Cyberwave()
prompt = "Inspect every pallet and alert me if one is damaged"

plan = cw.agents.workflow.plan(env_uuid, prompt)       # read-only
print(plan)                                           # review it first
draft = cw.agents.workflow.setup_and_draft(env_uuid, prompt, agent_plan=plan)
```

Then open the draft in the workbench, test it against simulation, and activate it when you are happy. See [Run a workflow against simulation](/feature-reference/workflows/simulation-execution-target).

## From your AI tools (MCP)

With the [MCP server](/overview/tools/mcp-server), Claude Code, Cursor or any MCP client can use `cw_create_workflow_from_prompt` to preview or draft a workflow, then `cw_list_workflows`, `cw_get_workflow` and `cw_trigger_workflow` to find and run it.

## Next steps

<CardGroup cols={2}>
  <Card title="Workflows" icon="diagram-project" href="/feature-reference/workflows">
    Triggers, nodes, execution and monitoring.
  </Card>

  <Card title="Workflow node reference" icon="list" href="/overview/features/workflow-nodes">
    Every node the assistant can use.
  </Card>
</CardGroup>