Tolaria: Karpathy's LLM Wiki Concept Is Now a Real Desktop App

By Prahlad Menon 3 min read

Andrej Karpathy proposed the idea of an “LLM Wiki” — a structured knowledge base that both humans and AI agents can read and write. That concept just became a production-ready desktop app.

Tolaria is a free, open-source desktop app for Mac and Linux that manages markdown knowledge bases. It writes like a modern block editor (think Notion), but saves everything as plain .md files with YAML frontmatter. No databases, no proprietary formats, no accounts required.

What Makes It Different

Plenty of note-taking apps exist. Tolaria’s angle is that it’s designed equally for humans and AI agents:

  • Files-first — Everything is plain markdown. Portable, editable with any tool, zero export steps.
  • Git-first — Every vault is a Git repo. Visual version history built into the app, use any remote.
  • Offline-first — No cloud dependencies, no subscriptions, no lock-in. Stop using Tolaria and you lose nothing.
  • MCP server built in — Claude Code, OpenClaw, and other AI tools can read, search, and edit your vault natively through the Model Context Protocol.

The MCP integration is the key differentiator. Your knowledge base isn’t just something you browse — it’s a live context source that any agent in your workflow can tap into.

The Engineering Behind It

The project is a showcase of AI-assisted development at scale:

  • Built with Tauri, React, and Rust
  • 100K+ lines of code shipped
  • 3,000+ tests at 85% coverage
  • 9.9/10 code health score (CodeScene)
  • 70+ architecture decision records

Created by Luca Rossi of Refactoring, who runs his own vault of 10,000+ notes on it daily.

Why This Matters

The LLM wiki pattern solves a real problem: AI agents need persistent, structured knowledge, but most note-taking tools store data in formats that are opaque to LLMs.

Tolaria’s approach — markdown files in a Git repo with an MCP server on top — gives you:

  1. Human editing through a polished block editor with keyboard-first design
  2. AI editing through MCP or direct file access (it’s just markdown)
  3. Version control with full Git history and rollback
  4. Portability — switch tools anytime, your data is just files

This is the pattern Karpathy described: a shared knowledge layer where humans curate and AI agents consume and contribute. Tolaria makes it concrete.

Getting Started

Download from the GitHub releases page. Create a vault (or point it at an existing markdown folder), and you’re running.

For AI integration, the built-in MCP server registers automatically — Claude Code and compatible tools discover it without configuration.

GitHub: refactoringhq/tolaria

License: AGPL-3.0-or-later

Walkthroughs: Workspace organization · Inbox workflow