Skip to main content
What this builds. A deploy-and-run-in-background script using pipeline.start(...) to fire-and-forget an LLM pipeline, then a RunHandler to track it. You’ll end up with. A task_id / trace_id you can store, plus the eventual result once the run completes — and the pipeline cleaned up afterward. 
from vectorshift.pipeline import Pipeline

# Create a standalone LLM pipeline
PIPELINE_NAME = "bg_run_example"
try:
    pipeline = Pipeline.fetch(name=PIPELINE_NAME)
    print(f"Pipeline fetched: id={pipeline.id}, branch_id={pipeline.branch_id}")
except Exception as e:
    print(f"Error fetching pipeline: {e}")
    pipeline = Pipeline.new(name=PIPELINE_NAME)
    print(f"Pipeline created: id={pipeline.id}, branch_id={pipeline.branch_id}")
inp = pipeline.add(name="input_0", id="input_0").input(input_type="string")
llm = pipeline.add(name="llm", id="llm").llm(
    provider="openai", model="gpt-4o", prompt=inp.text
)
out = pipeline.add(name="output_0", id="output_0").output(
    output_type="string", value=llm.response
)
pipeline.save(deploy=True)

# Start the pipeline in the background
handler = pipeline.start(
    inputs={"input_0": "Tell me a fun fact about space in 50 words."}
)
print("Pipeline started in background")
print(f"  Task ID:  {handler.task_id}")
print(f"  Trace ID: {handler.trace_id}")

# Check status manually
status = handler.run_status()
print(f"  Status:   {status['status']}")

# Block until the run completes and outputs are available
result = handler.result(poll_interval=2.0, timeout=120.0)
print(f"  Final status: {result['status']}")
print(f"  Outputs:      {result.get('result')}")


print(f"  Pipeline ID (to be deleted): {pipeline.id}")
pipeline.delete()
print("  Pipeline deleted")

​
Expected output

Pipeline created: id=..., branch_id=...
Pipeline started in background
  Task ID:  ...
  Trace ID: ...
  Status:   running
  Final status: success
  Outputs:      {'output_0': '...'}
  Pipeline ID (to be deleted): ...
  Pipeline deleted
handler.run_status() is non-blocking; handler.result(...) polls until the run finishes (or the timeout trips).

​
See also

Terminate a run

Cancel a background run mid-flight.

Run pipeline (async)

Same idea, but with await instead of polling.

Pipeline reference

Full method surface.