Lumos is a Chrome extension that acts as a Retrieval-Augmented Generation (RAG) co-pilot for web browsing, leveraging local Large Language Models (LLMs) via Ollama. It enables users to summarize web content, ask questions about articles or documentation, and interact with LLMs directly on their machine without relying on remote servers, offering enhanced privacy and control.

How It Works

Lumos integrates with a local Ollama server for LLM inference and embedding. It fetches web page content, chunks it, and uses an embedding model to create vector representations for efficient retrieval. When a user queries Lumos, it retrieves relevant chunks from the vector store and feeds them, along with the query, to the local LLM to generate a contextually aware response. This RAG approach allows LLMs to access and process information from the web.

Quick Start & Requirements

• Install Ollama: Download and install Ollama from ollama.ai.
• Pull a model: ollama pull llama2
• Start Ollama server: OLLAMA_ORIGINS=chrome-extension://* ollama serve
• Install Lumos: Load the unpacked extension from the dist folder in the project directory or download a pre-built package from the Releases page.
• Prerequisites: Ollama, a compatible LLM model (e.g., Llama2), Chrome browser.
• Setup Time: Estimated 10-20 minutes, depending on Ollama installation and model download speed.

Highlighted Details

• Supports multimodal LLMs for image analysis on web pages.
• Allows custom content parsing rules per domain/URL path for optimized text extraction.
• Features a chat history for saving and recalling conversations.
• Includes experimental tool invocation for enhanced LLM capabilities.

Maintenance & Community

The project is inspired by Web LLM and LangChain examples. Further details on community or maintenance are not explicitly provided in the README.

Licensing & Compatibility

The README does not specify a license. Compatibility for commercial use or closed-source linking is not detailed.

Limitations & Caveats

Lumos requires a local Ollama server to be running for functionality. While it aims for local-only inference, it relies on a local server for LLM operations due to Chrome extension security constraints. Content highlighted by the user is not cached in the vector store, requiring re-embedding for each subsequent prompt.