Filtering and finding runs

Every agent run is tracked with full step-level observability. The Runs tab on the agent detail page (and the global History → Agents view) lets you search, filter, and drill into individual runs.

Search runs by content

1. Open the agent’s Runs tab.

2. Type into the search box. Search runs full-text across the asked summary, did summary, input payload, and final output.

3. Combine search with verdict and metadata filters — they stack.

Filter by verdict

1. Use the Verdict filter to show only thumbs-up, thumbs-down, or unreviewed runs.

2. The Tune workbench and Review flipbook both default to verdict=down — this filter is the global view of the same set.

3. Clear the filter to see every run regardless of review state.

Filter by metadata

1. Add metadata filters to scope by trigger_type (manual, webhook, schedule, workflow, rerun, delegation), status, model, or caller.

2. Combine with verdict and search to drill into a specific scenario — e.g., “all webhook-triggered runs that were flagged in the last 7 days.”

3. Filters persist in the URL so you can share a filtered view.

Drill into a delegation chain

1. Click any run to open its detail page.

2. Expand Timeline to see the step-by-step LLM calls and tool executions.

3. Delegation steps render as expandable rows — click to inline the child run’s own steps without leaving the page. Child runs are filtered out of the top-level list, so the parent is the only entry point.

Rerun a finished run

1. On the run detail page, click Rerun. This works on completed, failed, and cancelled runs.

2. Bifrost creates a new run with the same input, trigger_type="rerun", and links back to the original. You’re navigated to the new run.

3. Use rerun to validate that a tuning change actually fixes a flagged scenario.

Cancel a running run

1. While a run is in running status, the run detail page exposes a Cancel button.

2. The run transitions running → cancelling → cancelled. Cancellation is checked before each iteration, during tool execution, and via a background watcher.

3. Cancelled runs can be rerun like any other terminal state.

Regenerate a run summary (admin)

1. If a run’s asked / did summary failed to generate, the run detail sidebar shows a Regenerate card. Platform admins also see this card on healthy runs.

2. Click Regenerate to re-queue the summarizer. The page polls and updates when the new summary lands.

3. Summaries cost a small amount per run — the Spend (7d) card on the fleet dashboard reflects regen cost.

Note

For bulk summary backfill (across many runs at once), use the Backfill summaries button on the fleet dashboard. It shows a cost estimate up front before queuing.

Open the global runs view

1. Click All runs in the agent fleet header (or open /history?type=agents).

2. The global view spans every agent and supports the same search/verdict/metadata filters plus an agent picker.

3. Use it for fleet-wide audits, ROI analysis, and finding runs across agents you don’t remember the name of.

Next steps

• Reviewing agent runs
• Tuning an agent’s behavior