Show HN: SQLite-RAG – A semantic search engine built on top of SQLite (github.com)

SQLite-RAG is a new open-source semantic search engine built on top of SQLite using the SQLite AI and SQLite Vector extensions. It provides a lightweight, local RAG-style search stack that combines vector similarity with full-text search (FTS5) and merges results via Reciprocal Rank Fusion (RRF) for better relevance. The project is aimed at developers who want a fast, embeddable retrieval layer with minimal ops: it’s preconfigured to use the Embedding Gemma model from Hugging Face, stores settings in the database, and exposes an interactive CLI (including a REPL) so you can add documents and run semantic queries with commands like sqlite-rag add-text and sqlite-rag search. Technically, SQLite-RAG builds dual indexes (vector + FTS), uses a token-aware recursive character splitter to create overlapping chunks, generates embeddings with user-selectable GGUF models, and lets you weight and configure embedding models, chunk sizes, and search fusion parameters. It accepts many formats (PDF, DOCX, Markdown, code, CSV/JSON, etc.), auto-creates a DB on first use, and supports custom database paths. The combination of SQLite’s reliability, local GGUF models, and hybrid retrieval with RRF makes this attractive for privacy-conscious, offline, or low-latency applications and for teams wanting an easy-to-run semantic layer without a separate vector database.