agentic-wiki

A Markdown second brain for humans, operated by AI Agents. Clone and start—no dependency install or build step. No database. No vector store. No server. Just files, links, Git history, and an operating protocol.

Quick Start • Daily Use • Obsidian • Structure • How It Works • Inspiration • 中文说明

Inspiration

Agentic Wiki is an opinionated starter vault inspired by Andrej Karpathy's LLM Wiki idea file: instead of asking an LLM to re-derive answers from raw documents every time, let it incrementally build and maintain a persistent, interlinked Markdown wiki.

Karpathy's gist is intentionally abstract: it communicates the pattern and expects you and your LLM agent to instantiate the directory structure, schema conventions, page formats, and tooling for your own domain. This repository is one such instantiation. It keeps the core ideas — raw sources, wiki pages, schema, ingest/query/lint, indexes, logs, Obsidian-friendly Markdown — and makes a specific set of choices for a multi-domain personal second brain.

The Problem

Most personal knowledge systems fail in one of two ways:

• Saved links pile up. Articles, highlights, and notes accumulate, but nothing gets synthesized.
• RAG starts from scratch. Every answer requires the model to retrieve fragments and reconstruct context again.

Agentic Wiki turns each useful input into something your future self can reuse. When you give the agent an article, paper, project note, or conversation, the protocol asks it to: