pg_index_search is an open-source tool that automates PostgreSQL index tuning by combining workload analysis, hypothetical index testing via HypoPG, and optional LLM-assisted suggestions. It reads slow queries from pg_stat_statements, parses them to identify indexable columns, generates candidate single and composite indexes, and tests them safely in memory without touching production data. A greedy search algorithm finds the best index combination using a cost function that weights query frequency and latency. For edge cases like partial indexes, expression indexes, or column ordering, an LLM (Claude or GPT) iteratively suggests alternatives, each validated through HypoPG. The tool outputs ready-to-run CREATE INDEX statements with standalone and stacked improvement metrics, leaving the final decision to the developer.
