Overview
The Workflow Agent builds a workflow pipeline from a natural-language prompt. It opens in a panel at the bottom of the workflow design screen and translates a described data operation — load, merge, save — into connected nodes on the canvas.
The agent binds to data through @ datasource references. A datasource named in a prompt must be connected in Datasource Management beforehand; the agent resolves the reference and any relative instruction, such as saving to the same database, to that datasource.
Note: A Workflow in the UI corresponds to a Weave pipeline in product documentation. This reference uses “Workflow.”
When to use it
- Building a pipeline by description rather than placing nodes manually.
- Loading a table from a connected datasource and merging it with another.
- Producing a starting pipeline that can then be refined on the canvas.
Prompt references
| Element | Description |
| @datasource | References a datasource by name, for example @Hospital_records. The datasource must already be connected in Datasource Management. |
| Source table | The table to load from the referenced datasource. |
| Merge table | A further table to merge into the pipeline, each referenced with its @ datasource. |
| Save target | Where the result is written. “Same db” resolves to the datasource referenced in the prompt; another datasource can be named instead. |
Key behaviors
@ binds to connected datasources only. An @ reference resolves to a datasource connected in Datasource Management. A name that is not connected cannot be referenced.
Relative save targets resolve to the referenced datasource. When a prompt instructs the agent to save to the same database, the target is the datasource referenced earlier in that prompt rather than a new connection.