Skip to content

Monitor pipeline runs

This guide shows you how to monitor and debug pipeline runs.

View runs

Navigate to Pipelines in the sidebar
The run list shows all runs with status, template name, start time, and cost

Filter by:

Status: running, completed, failed
Template: which template was used
Time range: recent, last 24h, last 7 days

Run detail

Click a run to see:

DAG visualization — the step graph with color-coded statuses
Step list — each step with status, duration, and cost
State — the shared state bag (input/output JSON)
Timeline — step execution timeline
Error details — if the run failed

Step inspection

Click any step to see:

Input: JSON data passed to the step
Output: JSON data the step produced
Error: error message if the step failed
Duration: wall-clock execution time
Attempt count: retries so far
Branch index: for parallel branches

For agent steps, the output includes the agent’s response text and any tool call summaries.

For bash script steps, the output includes stdout and stderr.

Debug failures

When a step fails:

Check the error message — it usually identifies the root cause
Check the input — was the data correct?
Check the attempt count — did retries run?
For agent steps: check the agent’s run log for detailed trace

Common failure reasons:

Agent unavailable: the agent was already running or not found
Timeout: the step exceeded its configured timeout
Script error: the bash command returned a non-zero exit code
Schema validation: the step’s output didn’t match the expected schema
State error: a referenced input key was missing from the state

Cost monitoring

The run detail shows cumulative cost across all steps:

Total cost: in cents
Total tokens: input + output across all agent steps

Use this to identify expensive pipeline templates and optimize them.

Run history

Each template shows its run history. Compare runs to track:

Whether the pipeline is getting faster or slower
Whether costs are increasing
Failure rates over time