GitHub Repository

A framework that replace traditional RAG pipelines. Ingest any number of documents in multiple workspaces (channels, departments, etc.), index it with BM25, and let the agent search, fetch, and reason over it, exactly like searching the web, but entirely on your machine. No vector store, no embedding needed.

8 starsPython

Local Search Agent – RAG replacement, no embeddings, free tier

Name: Local Search Agent – RAG replacement, no embeddings, free tier
Availability: InStock
Author: Wissam-Metawee

by Wissam-Metawee·Jul 17, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●●BangerBig BrainDark Horse

Swaps fragile vector embeddings for deterministic BM25 search to fix RAG reliability.

Strengths

•BM25 keyword search provides transparent, debuggable retrieval unlike black-box vectors.
•Eliminates vector database overhead and prevents index staleness from silent document updates.
•LangGraph agent loop fetches and reasons over local files with full source citations.

Weaknesses

•Keyword matching may miss semantic connections that embedding-based retrieval captures easily.
•Requires running a local Meilisearch instance, adding a dependency to the local stack.