1. Event contract
One row in agent_events, with the types used by the query made explicit.
- timestamp_utc
- Timestamp
- event_name
- Utf8
- data.tool_name
- Utf8
- data.status
- Utf8
- data.args.operation
- Utf8
- data.latency_ms
- Float64
How it works
A reviewable workflow from signal to decision
1
Give the question enough context
Name the table, metric definition, time range, grouping, and expected output. Ambiguous business language produces ambiguous SQL.
2
Review the generated assumptions
Check field types, joins, denominators, null handling, and time boundaries before trusting the chart.
3
Save only the queries that survive review
Use AI for acceleration, not authority. A saved query should be understandable by the next human or agent that maintains it.
Generating an editable query with AI
A real product capture showing a plain-language question becoming SQL that remains visible for review.
Boundaries
What this does not replace
- Generated SQL is a starting point, not an authoritative metric definition or security decision.
- AI assistance cannot see business context that is absent from the schema, prompt, or reviewed event contract.
- Queries should be checked for fields, joins, denominators, null handling, and time windows before they power a dashboard or alert.
Inspectable proof path
From event contract to a visible answer
This example uses a declared schema, read-only SQL, and deterministic synthetic results. It demonstrates the workflow without presenting sample data as a customer benchmark.
2. Read-only SQL
Which AI tools and arguments are associated with the most failed calls?
SELECT
"data.tool_name" AS tool_name,
"data.args.operation" AS operation,
COUNT(*) AS calls,
SUM(CASE
WHEN "data.status" = 'failed' THEN 1 ELSE 0
END) AS failures,
100.0 * SUM(CASE
WHEN "data.status" = 'failed' THEN 1 ELSE 0
END) / NULLIF(COUNT(*), 0) AS failure_rate_pct,
approx_percentile_cont("data.latency_ms", 0.95) AS p95_latency_ms
FROM agent_events
WHERE event_name = 'agent_tool_called'
AND timestamp_utc >= now() - INTERVAL '7 days'
GROUP BY "data.tool_name", "data.args.operation"
HAVING COUNT(*) >= 20
ORDER BY failure_rate_pct DESC, calls DESC;3. Synthetic result
CRM lookup is both the least reliable tool operation and the slowest at p95.
| tool_name | operation | calls |
|---|---|---|
| crm_lookup | search_contact | 842 |
| order_api | fetch_order | 2210 |
| knowledge_search | semantic_search | 4510 |
Capabilities
What is included
Schema-aware SQL generation
Prompt-to-visualization assistance
Generated SQL remains editable in the Query workspace
Works with the same tables, dashboards, and alerts as human-written analysis
Agent skill and prompt pack for repository-level instrumentation
Documentation
See the analysis
SQL recipes that use this capability
Query Nested AI Tool-Call Events
Filter dotted nested fields and rank failing tools without flattening the original event payload.
Open recipeMeasure Accepted AI Outputs per Dollar
Connect model spend to accepted, saved, or otherwise useful product outcomes.
Open recipeCalculate Weekly Cohort Retention
Group users by first activity week and measure the percentage returning in later weeks.
Open recipeCustomer evidence
How teams use this workflow
Related capabilities
Continue the event-to-decision workflow
Structured event ingestion
Turn application behavior into queryable event tables
Explore capabilitySQL workbench and Query API
Ask detailed questions with the same SQL humans and agents can inspect
Explore capabilityExplore
Filter, aggregate, and visualize a table without starting from blank SQL
Explore capabilityStart with one production workflow
Use a focused prompt, send synthetic events, and verify the first useful query before expanding coverage.