A new tool, Fidx, has been introduced that enables fast and private semantic search across local files and agent data, running entirely on CPU and utilizing a single SQLite file.
What Happened
The Fidx project has been released—a tool for local semantic search that combines hybrid search (BM25 and vector search via the sqlite-vec extension) within a single SQLite file. The system supports indexing Markdown, text files, code, and chat exports, providing millisecond response times even on large volumes of data without the need to call an LLM during the search query execution.
Context
Traditional RAG (Retrieval-Augmented Generation) methods often rely on cloud-based vector databases and require LLM calls to navigate data, which increases latency and costs. Fidx proposes an "Edge RAG" pattern, moving the computational load to the client side and using optimized SQLite extensions to perform vector search directly within the database.
Why It Matters for the Industry
This solution sets a new direction for implementing client-side RAG, demonstrating the possibility of creating high-performance search systems without relying on heavy cloud infrastructures. This contributes to the shift toward a "SQLite-first RAG" architecture and decentralized indexing methods within the AI agent ecosystem.
Why It Matters for Users
Users gain the ability to organize instant and fully private search across their local knowledge base, documents, or chat history. This eliminates the need to send sensitive data to the cloud and saves resources by not consuming power on heavy models for every search query.
Sources
Author
Look at AI, Editorial Team