Queries are the flexible layer beneath dashboards, audits, and investigations. Use ad-hoc SQL to test an idea, then save the query when it becomes part of a recurring workflow.
Create and run a query
- Go to Analyze → Queries and select new query.
- Open the schema browser and choose fields from the event, session, or evaluation data.
- Write SQL, add parameters when needed, and run the query.
- Save it with a clear name and description, then use add to dashboard when the result should be monitored.
The query editor combines the event schema, SQL, parameters, and a result preview so you can validate the question before saving it.
Saved queries then appear in the shared library, where teammates can rerun them or add their results to dashboards.
Use a clear name and description so the result remains understandable without reopening its SQL.Use fp query list to find IDs and fp query delete <query-id> to remove a saved query. Common uses
- Find sessions with repeated calls to the same tool.
- Compare evaluation scores across models or environments.
- Measure time between a human wait and resume.
- Identify policy denials followed by a successful alternative.
- Build a cohort for an audit.
Open Queries → Schema before writing against unfamiliar fields. Prefer explicit time and environment filters, and keep result limits during exploration.
A query can identify a suspicious pattern, but it does not establish the failure mode by itself. Open representative traces or run an audit before turning the result into a policy.
