Jev as a retrieval layer for an Obsidian vault
A tutorial shows a three-minute setup where Jev only picks which note or file to use, Claude Code does the work, retrieval stops burning tokens, and each decision takes about a tenth of a second.
Kev puts Jev in front of an Obsidian vault and Claude Code answers a question about a CSV before he finishes typing it - the whole setup takes about three minutes. this is his follow-up tutorial, and the trick is that Jev never writes anything. it only decides which note, file or asset is the right one, then Claude Code does the work. community plugin on → MCP server enabled → API key handed to the agent → the vault becomes searchable memory → "we got it back before I even finished the sentence." "38% of medium rare steak eaters also do speed when they drive." - the answer that came out of a survey CSV he dumped into the vault the part that got me is what a vault actually is: a folder of markdown files. no index to build, no embeddings to maintain, and a logo search in the demo returns four files ready to drop into an editor. Jev is TypeSafe's System One model, so it costs a fraction of a cent per decision and answers in about a tenth of a second. the retrieval step stops burning tokens, because the model doing it isn't reading your whole repo. bookmark & watch today ↓
This post quotes another post: “Long read: how to build a Jev-powered agent stack”