Chouse AI in Monitoring
Optimize query-log rows, fix system.errors rows and diagnose part-log rows without leaving the tab — read-only, before→after EXPLAIN.
- In the app
- Monitoring › Query logs, Errors and Parts
/monitoring - Needs
- ai:optimize
Chouse AI is built into three Monitoring tabs, so you can work on a problem without leaving the row you found it on. It needs ai:optimize and a configured AI model.
The three actions
Optimize (query logs)
On a query-log row:
- Click Optimize with Chouse AI
- Receive an optimized rewrite that produces the same result
- Review the before → after
EXPLAINestimate as proof - One click Open in Explorer to run the rewrite yourself
The AI never runs the rewritten query — you do, after review.
Fix (errors)
On a system.errors row, Fix returns cause / impact / ordered solutions for that specific error.
Diagnose (parts)
On a part-log row, Diagnose reads part health: merge pressure, too-many-parts symptoms, partition-key evaluation.
Why in-tab matters
The classic flow — copy an error into a chatbot, paste back a guess, hunt for the query, guess again — loses the context. Here the diagnosis and the fix live next to the problem, with the cluster's real telemetry one click away.
Safety model
| Guarantee | Mechanism |
|---|---|
| Read-only | readonly=1, single-SELECT tool surface, system.* only |
| Advisory only | Suggestions require human review before anything runs |
| Spend-gated | ai:optimize controls who can invoke it; provider budgets apply |
Provider setup
Uses the same models as AI Assist, configured in AI models. With no active deployment, the AI actions are hidden.
Quick tour
- Open query logs → pick your slowest pattern
- Optimize with Chouse AI → read the rewrite + EXPLAIN delta
- Open in Explorer → run the improved query
- Re-check the pattern's cost after a day — the Patterns sub-view shows the before/after in Total duration
Warning: The AI can be wrong — EXPLAIN estimates are estimates. The evidence chain (rewrite → EXPLAIN → your run) is designed so you can verify every step before trusting it.