Show HN: IngotDB – SQL-based memory for LLM agents

Category: ai-ml

Tags: llm, memory, sql, vector-search, duckdb, postgres, rag, agent

Score: 7.0/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 6)

IngotDB is an open-source memory layer for LLM agents that stores tool results and documents as queryable tables, combining Postgres for hot writes and Parquet for cold storage. It uses DuckDB to unify SQL, full-text, and vector search, eliminating the need for a separate vector store and enabling hybrid queries. The project is notable for its LSM-tree architecture, per-format chunking, and a benchmark that argues for SQL-based retrieval over top-k RAG.

Target audience: AI engineers, backend developers, data engineers

Repository: https://github.com/tjbroodryk/ingot · TypeScript · MIT · 5 stars

View on Hacker News