Workflows
What it is
A Workflow is a multi-step pipeline. Where an Agent is a single prompt-and-model call, a Workflow chains several steps in order — each step's output becomes the next step's input.
Step types
| Type | What it does |
|---|---|
render | Render a template, filling (and ) |
model | Call a model provider with the current text |
transform | Apply a text transform: upper, lower or trim |
The special variable in a render step holds the output of the previous step, which is how steps are chained together.
Why it matters
Real tasks are rarely one call. For example: render a summarisation prompt → call a model → trim the result. A Workflow makes that sequence a single, saved, runnable unit instead of glue code scattered across your app.
Using it
- Open the Workflows page and click New Workflow.
- Give it a key and name.
- Click Add step for each step. For each one, pick a type and configure it:
render— write the templatemodel— pick a provider and modeltransform— pick the operation
- Reorder steps with the ↑ / ↓ buttons.
- Click Save, then fill in any variables and click Run Workflow.
Reading the result
After a run, the result panel shows:
- Step trace — the output of every step, in order, so you can see exactly where the text changed.
- Final output — the result of the last step.
Behind the scenes
Running a workflow calls POST /api/workflows/:id/run, which returns the full step trace. Each run is recorded for Observability.
Next steps
- Agents — the single-step counterpart.
- Playground — try individual prompts before wiring them in.