Browse / Second brain / LLM wiki / second-brain
second-brain
trinhvandat
A personal knowledge base built on the LLM Wiki pattern (Karpathy): no vector database, no app, no cloud.
Open trinhvandat/second-brain → Checked against the four rules.
Works with any agent
ships AGENTS.md, the cross-tool instruction file.
In its own words
A local-first Markdown vault that Claude reads, answers from, and maintains — your notes turned into a wiki that talks back.
A personal knowledge base built on the LLM Wiki pattern (Karpathy): no vector database, no app, no cloud. Knowledge lives in plain .md files you own; Claude reasons directly over them and keeps them tidy. Every answer traces back to a specific file.
Why this exists
- You own everything. Plain-text Markdown on your disk. No lock-in, no silo, fully
grep-able and git-versioned. - Answers you can trust. Claude answers only from the vault, with
[[file]]citations. Nothing in the vault → it says "Not in the second brain" instead of hallucinating. - It maintains itself. A capture → compile → lint loop turns raw clippings into atomic, cross-linked notes, and flags broken links, orphans, stale claims, and superseded knowledge.
How it works
.md files are the source of truth. The operating rules live in AGENTS.md and are read by both Claude Code and Claude Desktop, so the two surfaces behave identically.
Requirements
- Claude Code (primary surface), and/or Claude Desktop for a chat-style surface
bash3.2+ (macOS system bash works),git- Optional:
pandocfor richer URL clipping, Node.js (npx) for the Claude Desktop MCP server
Install
That's it — no build, no dependencies. The Claude Code skill auto-activates when you start Claude in this directory.
Quoting 38% of the README. Read the rest on GitHub ↗
Read from the repository's git tree on 2026-08-17.