Execution History

Every pipeline trigger creates an execution recorded in the database. You can inspect it to the second, see what each node produced, and replay it against a different version of the pipeline.

Accessing the history

Designer → Pipelines → your pipeline → History.

The table is filterable by:

• Time period
• Status (success, error, awaiting human approval, abandoned)
• Input channel
• Cost (in credits)

Anatomy of an execution

Click a row to open the execution drawer. You see:

• The complete input payload (display-side anonymisable according to your policy)
• The traversed graph — each node visited, with its state (skipped, ok, error)
• For each node:
  • Its inputs at the time of execution
  • Its outputs produced
  • The execution duration
  • The cost (for metered nodes: agent, voice, etc.)
• The tools called by each agent, with their arguments and return values
• The complete LLM log (prompts sent, responses received)

Replay

From the drawer, two options:

• Replay as-is — on the same pipeline version. Useful for reproducing a bug.
• Replay on current version — switches to the latest published version. Useful for validating that a fix correctly resolves a past failure.

The replay is tracked as a new execution linked to the original. The history remains intact.

Advanced filtering

The table supports native filters by single value or date range. You can also export the filtered executions as Excel / PDF for sharing outside the platform.

Retention

After the retention period, executions are purged. The GDPR audit log (see Audit) remains accessible for longer, in line with the regulatory obligations of your sector.

Audit use case

If an auditor asks you to "show all calls where the agent refused a refund request":

1. Filter by pipeline.
2. Filter by status.
3. Filter by tool called (payment.refund_request).
4. Export the selection as Excel.

Everything is traceable, nothing is fabricated.