// the take
Lumos is a Chrome extension that adds a RAG chat sidebar to any web page, using Ollama for both local embeddings and local LLM inference so no page content or prompts leave the machine. It's aimed at developers who want to ask questions about long articles, docs, or threads without paying for or trusting a hosted LLM API.
Fully local pipeline (embeddings + inference via Ollama) is the real selling point, not just a thin wrapper around OpenAI. Per-domain content parser config (CSS selectors, chunk size/overlap) is a genuinely useful detail most RAG-over-webpages tools skip, since a generic body-text extractor is often wrong for Reddit, Wikipedia, Stack Overflow, etc. Multimodal support (binds images on the page to the model) and file attachment parsing (CSV, JSON, PDF, plaintext) extend it past pure web-page QA, and the vector store TTL cache avoids re-embedding the same page on every prompt.
Setup friction is real: users need a separately running Ollama server with OLLAMA_ORIGINS set correctly or every request silently fails on CORS, which is a support burden for non-technical users. Last push was January 2025, so it's been quiet for a while and newer Ollama/LangChain.js releases may have drifted from what it was built against. The 'Tools' (agent) feature is explicitly marked experimental, and local LLM quality/latency on consumer hardware will be a hard ceiling for anything beyond short-document summarization. It's also Chrome-only by design (Firefox/Edge get separate doc pages as workarounds, not native support), so it's a bigger lift to use outside the Chrome ecosystem.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →