LLM-Wiki

A personal knowledge base template maintained collaboratively with an AI coding agent, inspired by Andrej Karpathy's LLM Wiki concept. Drop raw sources in, let the agent compile them into a structured, interlinked wiki — then query, synthesize, and explore.

Model-agnostic: works with any agent that reads AGENTS.md (Claude Code, Codex, Gemini CLI, etc.).

How it works

• Drop raw content (articles, papers, repos, datasets) into raw/
• The agent compiles each source into a summary, extracts concepts and entities, and links them together
• The source is consumed — moved from raw/ to archive/, so raw/ always shows exactly what's still unprocessed
• Query the wiki with natural language; the agent answers from the structured content
• Health-check with python tools/lint.py to catch dead links, orphans, missing index entries, and gaps

The wiki lives in plain Markdown, works great in Obsidian (with Dataview for live queries), and is fully readable in any editor.

Getting started

Prerequisites

• Python 3.10+ (for the tools)
• Obsidian (optional, for Dataview dashboard)

Setup

No dependencies to install — the Python tools use only the standard library.

Quick start

Detects installed AI agents on your system (Claude Code, Codex, Gemini CLI, Cursor, Aider, OpenCode) and launches your pick in the project directory.