Security & Query Safety
- Ensured only single statements are passed to
EXPLAIN
(#224), with improved parsing to block multiple commands. - Renamed
explainQuery
tounsafeExplainQuery
to highlight potential risks where edge cases might escape safety checks. - Introduced
safeExplainQuery
, which fetches SQL directly frompg_stat_statements
byqueryId
. - Switched unsafe explain to use
generic_plan
for consistency and safety (#229).
AI Model Updates
-
Migrated OpenAI integration to GPT-5 only (#211):
- GPT-5 is now the default (chat + scheduling).
- Removed support for GPT-4 and other deprecated OpenAI models.
- Migration ensures schedules using GPT-4 are upgraded to GPT-5.
- Pending LiteLLM and judge/eval testing.