mudler/LocalRecall

What it solves

LocalRecall provides a lightweight, offline way to manage knowledge bases and files for AI agents and chatbots. It removes the need for GPUs, internet connectivity, or cloud services to handle long-term and short-term memory retrieval, offering a generic abstraction layer for knowledge management.

How it works

It operates as a RESTful API that interfaces with vector databases to store and retrieve information. It supports two main storage engines: Chromem (a local file-based store) and PostgreSQL (which uses pgvector and pgvectorscale for production-grade hybrid search). The system can ingest Markdown, plain text, and PDF files, as well as external sources like web pages, sitemaps, and Git repositories, which it monitors and updates automatically.

Who it’s for

It is designed for developers building AI agents, chatbots, or RAG-based applications who want to keep their data fully local and avoid cloud dependencies.

Highlights

• Fully Local: Works offline without cloud services or GPU requirements.
• Hybrid Search: Supports combining BM25 keyword search with vector similarity search via PostgreSQL.
• Diverse Ingestion: Handles local files (PDF, TXT, MD) and external sources (Web, Git, Sitemaps).
• Flexible Storage: Choice between a simple file-based store (Chromem) or a scalable production database (PostgreSQL).
• MCP Integration: Compatible with the Model Context Protocol for easier control via AI tools.