Most SQL performance tooling requires either a $400/month monitoring agent or asking models and hoping the advice applies to your database.
The interesting architecture decisions:
- Heuristic engine runs first (<200ms), LLM is optional and additive — if LLM fails, you still get structured findings
- All dialect logic lives in one dialect_config.py — DDL templates, optimizer syntax, LLM system prompts, maintenance commands for all 5 DBs
- Schema-aware mode: paste DDL, get confirmed recommendations with real table names instead of placeholders
- Every analysis gets a permanent shareable URL
Source: https://github.com/AutoShiftOps/querytuner
Live: https://querytuner.com