Skip to content

Pipelines

A pipeline connects steps into one job that runs on its own: for example, every morning, summarize yesterday's agent failures, score a sample with an evaluation, and alert if quality dropped.

Pipelines live under Pipelines in the console:

Page What it's for
Pipelines Your pipelines, their status and run history
Pipeline Builder Build or edit a pipeline visually
Pipeline Incidents Runs that failed, timed out or got stuck

Pipeline List with scheduled and paused pipelines

What you can put in a pipeline

Step Does
Process unit Runs one of your process units
Eval run Scores data with an evaluation
Trigger Checks an alert and notifies
Function or algorithm Runs custom logic or a built-in algorithm
Condition Continues only if a logical condition holds
Timeout Fails the run if a step takes too long
Shared pipeline Runs another pipeline as a step

Steps can run in sequence or in parallel.

%%{init: {"flowchart": {"curve": "linear"}}}%%
graph LR
    A["Process unit:<br/>yesterday's failed runs"] --> B["Eval run:<br/>score a sample"]
    B --> C{"Score below<br/>threshold?"}
    C -->|yes| D["Trigger:<br/>Slack alert"]

Build a pipeline

  1. Open Pipelines → Pipeline Builder and click Create new pipeline.
  2. Give it a name and description.
  3. Drag steps from Available Units onto the canvas and connect them.
  4. Add conditions where the flow should branch.
  5. Save, then run it once with Run configured pipeline to test.
  6. Set it active.

Use Clone pipeline to start from an existing one.

When something goes wrong

Pipeline Incidents lists failed, timed-out and stale runs. From there you can stop the jobs of a stuck pipeline and run it again.