3 ms·
Gemma 4 E4B running inside Obsidian via WebGPU and Karpathy-style LLM Wiki
- itsyuimorii 2mo agoI made Gemma4 E4B run directly inside Obsidian, it allows me to chat with any specific note or the whole vault of notes! Even better, it uses Andrej Karpathy's llm-wiki format for each note so it helps me connect ideas and organize them into a knowledge base . Everything it writes is plain Markdown, but your original notes stay untouched! Currently my creation supports: - Turning the notes you've already collected into a living wiki. - Summary cards + an index: See what’s in your vault at a glance. - Find orphan notes & reconnect ideas: Discover isolated notes and hidden relationships. - Related links + shared tags: Relink your knowledge and surface connections. - Concept pages: Bring related notes together around emerging topics. - Find gaps and contradictions: Surface missing pieces and conflicting claims. - Quizzes + flashcards: Turn saved notes into something you can revisit. LiteRTLM + WebGPU do the work, and the model runs inside Obsidian’s renderer process. So after a one-time ~3 GB model download, it all runs entirely locally — and offline with #WebAI. If you’re a fan of Obsidian Web Clipper and have ended up collecting too much — work notes, learning notes, ideas, articles, papers, and saved posts... this is for turning that growing pile into a connected, searchable, living knowledge base. Free, private, and local. No API key. No account. No Ollama. No LM Studio. No separate inference server. No subscription. Two days since launch — nearly 200 downloads. Download link: https://lnkd.in/gp8s5Zp9 https://lnkd.in/gp8s5Zp9 Let me know if there’s any feature requests or if you find any bugs. I’d really appreciate any feedback!