Skip to main content

​
Serving a Workflow

Once your nodes and workflows are defined, you can serve them using the built‑in FastAPI server.

​
Steps

  1. Define Your Workflow
    Use the @workflow() decorator to create your workflow.
  2. Start the Server
    Use the CLI command:     
    pui start
    This launches the FastAPI server and the React UI.
  3. Access APIs and UI
    • REST endpoints (e.g. /api/workflows) let you run workflows.
    • The web‑UI at http://localhost:8114 provides an interactive interface.
  4. Stream Execution Status
    The backend streams workflow progress via Server‑Sent Events at /api/workflows/sse/events.
Playground Using the playground UI, you can run your workflow by clicking the “Run” button. Or you can copy the curl command and run it from the terminal. 
curl -N -X POST \
  'http://localhost:8114/api/workflows/run/simple_workflow' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "input": {
    "model_id": "stable-diffusion-v1-5/stable-diffusion-v1-5",
    "prompt": "A beautiful image"
  }
}'
This outputs the following: You can also run the workflow using the REST API directly. 
{
  "name": "simple_workflow",
  "status": "completed",
  "func_type": "workflow",
  "duration": "36.70s",
  "artifact_id": "1d76ee91-dda0-428a-8386-e3755d3fcca8",
  "artifact_url": "localhost:8114/api/workflows/artifacts/1d76ee91-dda0-428a-8386-e3755d3fcca8"
}
As pipeline-ui is still in early development, the response format for the REST API is subject to change. In the future we want to align to replicate.com format as much as possible.

​
Performance

What is really cool is that the API does not introduce any additional latency to the workflow! Indeed, it is the same as running only your python code but exposed with a nice fastapi endpoint. We do have some additional latency when we send the requests to the UI, but this is completly optional. Do not visit the UI webpage and then zero extra work is done to display the UI. Btw, this took us a while to figure out. 😅