andrewnguonly/Lumos
A local LLM sidekick that actually reads the page
Lumos exists to turn any web page into a private, locally-run RAG session so you can query articles, threads, and docs without shipping your browsing history to a cloud LLM.
Not currently ranked — collecting fresh signals.
star history
What it does
This Chrome extension treats the current web page as a knowledge base for a local LLM. It parses content—by default everything inside the <body> tag, or via custom CSS selectors configured per domain—chunks it into a vector store cache, and answers questions through a retrieval-augmented generation pipeline using an Ollama instance running on your own machine. You can summarize articles, extract details from reviews, or query technical documentation without copying text into a separate chat window.
The interesting bit
Instead of re-embedding the same page on every prompt, the extension keeps an in-memory vector store with a configurable TTL, reusing embeddings for repeated queries. It also supports multimodal models by binding on-page images to the prompt, and can ingest file attachments—PDFs, CSVs, JSON, plain text, and images—running them through the same local RAG workflow.
Key highlights
- Runs entirely against local Ollama models; no remote inference involved.
- Per-domain content parsing rules via CSS selectors and adjustable chunk sizes.
- Optional highlighted-text mode: select a specific passage to query instead of the full document.
- Multimodal support for on-page images and uploaded image files.
- Experimental tool-calling that auto-invokes LangChain tools based on prompt classification.
- Chat history, keyboard shortcuts, and clipboard-to-attachment workflow for quick context injection.
Caveats
- Requires a local Ollama server running alongside the browser; it will not work standalone.
- The “Tools” feature is explicitly marked experimental.
- The README teases an “Enable/Disable Dark Arts” option with no explanation beyond a smiley face, so its capabilities are genuinely unclear.
Verdict
Developers and researchers who want LLM assistance on web content without shipping browsing data to a third-party API should take a look. If you are unwilling to run a local Ollama server or just want a simple cloud-based chat overlay, this is not your extension.
Frequently asked
- What is andrewnguonly/Lumos?
- Lumos exists to turn any web page into a private, locally-run RAG session so you can query articles, threads, and docs without shipping your browsing history to a cloud LLM.
- Is Lumos open source?
- Yes — andrewnguonly/Lumos is open source, released under the MIT license.
- What language is Lumos written in?
- andrewnguonly/Lumos is primarily written in TypeScript.
- How popular is Lumos?
- andrewnguonly/Lumos has 1.5k stars on GitHub.
- Where can I find Lumos?
- andrewnguonly/Lumos is on GitHub at https://github.com/andrewnguonly/Lumos.