I’ve kept 330+ notes for years. Two years ago, I migrated from Evernote to Obsidian to build my own wiki/ knowledge base from multi-year saved documents. Not only technical ones, but any personal interests — how to fly DCS, riding motorcycles in mountains, photograph techniques — in both English and Chinese.
I can query and chat with my wiki, pulling related topics across all my notes. Amazing, isn’t it?
Found Karpathy LLM Wiki and got interested. It’s simpler than your expectation:
There are tons of tutorials about how it works and its architecture/ workflow. I recommend:
I won’t repeat them. There are complex variants, but I tried several and burned tokens for days — this is the simplest approach.
I have a GPD Win4 gamepad with internal and external AMD GPUs. I wanted to test its limits running a local LLM — speed and summarization quality.
Select a vault to start with. Start with a new vault and a few documents — LLM processing really takes time and costs money if using a public API.
Install Karpathy LLM Wiki plugin through community plugin.
Don’t install any other plugin yet — you won’t know which is which.
Connecting to a public LLM is simpler, but I burned $10/ day for 3 days and switched to local.
As long as ollama service is up, download a model. For instance:
ollama run qwen2.5:7b
Model size depends on your hardware’s vRAM capacity. The one I’m using requires around 6~ 7 GB vRAM. Google’s Gemma 4 would work too.
Obsidian settings > Community plugins > Options (gear icon) of Karpathy LLM Wiki >
Once set up, ingest (copy/ move) your files into the raw/ folder. Press Ctrl + P > Karpathy LLM Wiki ingest from folder > select raw/ folder.
You should see GPU1 usage spike like the bottom-right screenshot above. The top-right is Graph View of my wiki built from 330+ documents — looks impressive.
I’ve used Arch Linux for years, only kept Windows for gaming. Ditched it completely when Proton made Windows games run on Linux without performance penalty. Arch Linux can do almost everything I need — it’s now my only OS for work + leisure.
btw, I use Arch
Build Your Own AI-Powered Wiki with Obsidian + Karpathy LLM Wiki + Ollama was originally published in Dev Genius on Medium, where people are continuing the conversation by highlighting and responding to this story.