Skip to main content
The Traces page is a firehose. Search and filters are how you turn it into targeted results.

​
Quick filters

Top of the Traces page:
  • Time range — last hour, day, week, custom.
  • Status — success / errored / stopped / running.
  • Feedback — thumbs-up / thumbs-down / no feedback.
  • Verification — passed / warned / blocked.
  • Cost bucket — cheap / normal / expensive.
Combine filters with AND semantics. The search bar accepts a query language:
Supported fields:
  • cost (in USD)
  • duration (with unit: ms, s, m)
  • status, feedback, verification
  • flow, version (Flow slug / version ID)
  • user_id, tenant, session (scope fields from your Trigger)
  • has_tool (any tool the run called)
  • tool_error (any tool that errored)
  • model (model name)
  • answer_length (chars)
  • citations (count)
  • groundedness_score
  • Any custom tag or feature you added
Operators: =, !=, >, <, >=, <=, AND, OR, NOT, IN [...].

​
Full-text search

Wrap a phrase in quotes to search prompts, tool arguments, tool responses, and answers:
Text search runs against the full Trace payload. Respects your Flow’s payload retention setting — traces past the payload-retention window are metadata-only and won’t match text queries.

​
Saved views

Save any filter combination as a view. Views show up in the sidebar and can be shared with teammates. Common saved views:
  • “Yesterday’s downs” — filter by feedback:down, last 24h.
  • “Expensive support runs” — cost:>0.5, flow:support-agent.
  • “Verification blocks this week” — verification:blocked, last 7d.

​
Exporting

Any filter result can be exported as:
  • CSV — one row per trace, includes summary metrics.
  • JSONL — full trace payloads, one per line.
  • Dataset — creates a Dataset in Reliable → Datasets (see Datasets).
Exports are async; large exports (>10k traces) are emailed as a download link.

​
Programmatic access

The API GET /traces?query=... accepts the same query language. Useful for building custom dashboards or offline analysis. Rate limits apply — see the API reference.

​
Insights view

For patterns across many traces (not just individual runs), the Insights page has pre-built charts:
  • Cost over time.
  • Feedback rate over time.
  • Top failing intents.
  • Retrieval hit rate.
  • Model cost breakdown.
  • Slowest tools.
Every chart is filterable with the same query language.