Semantic search on tiny databases just got buzzier. SQLite-RAG blends vector search with the SQLite ecosystem, using Reciprocal Rank Fusion to merge vector results with full-text results from the FTS5 extension [1]. It’s a hybrid approach tuned for embedded workloads where you want both semantic and exact-text recall in one store.
SQLite-RAG (hybrid vector + FTS) On top of SQLite, it wires vector similarity alongside traditional text search. The goal is better document retrieval by combining signals rather than choosing one path [1].
Lingo (on‑device semantics) Lingo is pitched as the on‑device linguistic database—designed to run locally with nanosecond‑level performance, a bold on‑device take for semantic search [2]. The discussion around memory‑mapping (mmap) flags a key architectural debate: mmap can blur the line between disk and RAM and spark correctness issues, though some teams lean into it in tightly controlled single‑writer setups [2].
Cellulite (geo‑aware in Meilisearch) For geo workloads, the LMDB‑based Cellulite store stores geojson inside Meilisearch, enabling geo‑centric search inside the Meilisearch ecosystem [3]. The approach highlights how geo data can live close to text and vector indexes in embedded or edge scenarios.
Closing thought: there isn’t a one‑size‑fits‑all path—hybrid vector+FTS, on‑device semantics, and geo‑first storage each map to different real‑world workloads.
References
Show HN: SQLite-RAG – A semantic search engine built on top of SQLite
Hybrid search on SQLite unites vector similarity with FTS5, using Reciprocal Rank Fusion to improve document retrieval.
View sourceShow HN: Lingo – A linguistic database in Rust with nanosecond-level performance
Rust linguistic database for on-device semantic search; open-source; mmap debate; feedback requested; critiques of claims and repos from community comments.
View sourceNew Geo Database
LMDB-based database to store and retrieve GeoJSON inside Meilisearch
View source