LocalRecall is a 100% local, RESTful API for managing knowledge bases and files for AI agents, offering a simple abstraction layer for retrieval-augmented generation (RAG). It operates entirely offline, requiring no internet or cloud services, and is designed for seamless integration with AI frameworks like LocalAI and LocalAGI, providing a user-friendly web UI for file management.

How It Works

LocalRecall utilizes a local vector store, currently powered by Chromem, to index and retrieve information from various file types including Markdown, plain text, and PDFs. It supports RAG workflows by vectorizing content using specified embedding models, allowing AI agents to query and retrieve relevant data for context. The project is built with Go, emphasizing a lightweight and efficient design for local deployment.

Quick Start & Requirements

• Install: Clone the repository and build from source using go build -o localrecall.
• Run: Execute ./localrecall. The web UI is available at http://localhost:8080.
• Prerequisites: Go 1.16 or higher. Docker is optional for containerized deployment.
• Docker Deployment: Pre-built images are available on quay.io/mudler/localrecall.
• Configuration: Environment variables are used for customization (e.g., EMBEDDING_MODEL, OPENAI_BASE_URL).
• Docs: REST API documentation is provided within the README.

Highlighted Details

• Fully local operation, no internet or cloud dependencies.
• Supports RAG with multiple file types (Markdown, TXT, PDF) and plans for more.
• Integrates with LocalAI, LocalAGI, and other agent frameworks.
• RESTful API for managing collections, uploading files, and searching.
• Supports adding and automatically updating external sources like web pages and Git repositories.

Maintenance & Community

The project is part of the "Local Stack Family" alongside LocalAI and LocalAGI. Contributions are welcomed via issues and pull requests.

Licensing & Compatibility

Released under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

Currently supports Chromem as the vector engine, with plans to add Milvus and Qdrant. Embedding model selection and configuration are crucial for performance. Private Git repository access requires providing a base64-encoded SSH private key.