Every new AI conversation starts blank. You re-explain who you are, what you're working on, and what you already tried — every single time — because the model has nothing to reason from but the current chat. This kit is the fix I use every day, with my life removed so you can put yours in.
What it is
A starter repository for an AI-maintained second brain: a persistent, cross-linked knowledge base in plain markdown that Claude Code reads from and writes to. Raw sources go in an inbox; the assistant processes them into wiki pages, links them, logs the work, and files the originals away as provenance. Ask it questions and the good answers get filed back in. The longer you run it, the more it compounds.
The architecture is the LLM-wiki pattern, plus the layers most do-it-yourself builds skip: an epistemic-quality layer that keeps the brain trustworthy rather than just large — load-bearing claims verified, evidence labeled, contradictions named instead of silently resolved — an assistant constitution that fixes how the AI behaves so you don't renegotiate it every session, and the maintenance loops that keep the whole thing from rotting the way most personal wikis do.
You own the artifact
Plain markdown files, versioned with git, readable without any tool. No platform can reprice, lock, or delete your brain.
That's the durable bet. Notes apps come and go; a folder of text files you can read in anything, with your assistant's operating rules stored next to the knowledge itself, survives every platform decision anyone else makes.
Where it comes from
This isn't a design exercise — it's the system behind everything else on this site, de-personalized. The wiki that runs my work has hundreds of pages, a daily journal, a task system, and an assistant that improves its own rules from its mistakes. The kit ships that machinery: the schema, the constitution, nine workflow skills, the export guides for getting your data out of every major service, and twenty-six seed pages so the graph isn't empty on day one.
Current state
Shipped and MIT-licensed — my first open-source release. Clone it, run the setup, and start feeding it your data; the README walks the whole path, including the honest part about which data exports take days to arrive and how not to miss their download windows. Get the kit →