A guide · ~6 min read
Obsidian vault → portable AI persona
Obsidian users already won the data-ownership argument. The vault is a folder of Markdown files on your machine. No export ceremony, no proprietary database, no shutdown risk. But the second-brain advantage stops at the AI boundary — ChatGPT can't read your vault, Claude can't follow your wiki-links, and the plugins that do help only work inside Obsidian itself.
Why dragging notes into ChatGPT doesn't scale
You can attach a Markdown file to a chat. You can paste a handful of notes. But a real vault is 500–5,000+ notes, and the model has no idea which 20 to weight for any given question. The signal-to-noise problem isn't theoretical — it's why "I just paste my notes into ChatGPT" usually underwhelms after the first session.
The fix is pre-processing. Run distillation across the whole vault, find the patterns, produce one compact persona file. Pasting one file is fast. The model gets the structured "who you are" without having to ingest the raw corpus on every question.
The portable answer
Konshus has an Obsidian importer on Founding and above. We respect the wiki-link graph, weight repeated themes, and produce a persona that fits in any AI prompt. The notes themselves stay searchable in your vault; nothing gets used to train anything external.
See also Notion as AI memory and Readwise → AI persona for the rest of the second-brain trio.