Skip to main content
Every pipeline run — manual, scheduled, watched, or webhook — is recorded. The Run logs page lets you drill in from the top-level summary all the way to a single item’s journey through every stage.

​
The runs list

For a given pipeline, the runs list shows:
  • Started at — kickoff timestamp.
  • Duration — total wall clock.
  • Status — success / partial / failed / running.
  • Items in / out — how many source items came in, how many chunks were written out.
  • Cost — total spend (embeddings, LLM enrichers, model calls).
Filter by date range, status, or trigger type.

​
Drill into a run

Click a run to see its stage breakdown:
  • Trigger — what kicked it off (with the run payload).
  • Source — items pulled, size, cache hits.
  • Clean — dropped counts, redaction stats.
  • Normalize — chunks produced per source item.
  • Enrich — per-enricher time and cost.
  • Output — writes by destination.
Each stage shows per-block timing so you can spot the slow parts. Long stages don’t necessarily mean broken — a good pipeline can spend most of its time on embeddings.

​
Item-level view

Any item that made it into the pipeline can be inspected end-to-end:
  • The raw source (or a hash if it was too large).
  • What the Clean stage did to it (which redactions, which strippers).
  • What chunks Normalize produced.
  • What metadata Enrich added.
  • Where in the Output store it landed (or why it was skipped).
Great for debugging “why isn’t this document showing up in retrieval” questions.

​
Errors

Errors show up as red items in the stage view. Common ones: Click any error for the full stack and the exact input that triggered it.

​
Rerun

Two flavors:
  • Rerun (from cache) — only reprocesses items whose source content changed. Fast. Default.
  • Rerun (ignore cache) — reprocesses everything. Use when pipeline logic changed and you want a full rebuild.

​
Alerts

Configure per-pipeline alerts:
  • On failure — post to Slack / email / webhook.
  • On threshold — items dropped > X%, cost > $Y, duration > Z minutes.
  • On silence — expected schedule missed.
Alerts route through your workspace’s Notifications settings.

​
Retention

Run logs retain for 90 days by default. Metadata (durations, counts, cost) retains for 1 year. Raw source snapshots follow the pipeline’s payload retention setting.