Introducing Tolaria: A Markdown Knowledge Base App Built Entirely with AI

Introducing Tolaria 💧

TLDR: Luca Rossi, the author behind Refactoring.fm, has released Tolaria — a free, open-source Mac desktop app for managing markdown knowledge bases. It's built on files, Git, and opinions about how knowledge should be organized. The kicker: 100K lines of code, and Luca wrote none of them.

Summary:

I've been watching the "second brain" software space for years, and honestly most of it has been variations on the same theme. Then something like Tolaria shows up and makes me think differently about what we're actually building toward.

The premise here is straightforward but the execution is interesting. Luca decided to migrate his entire decade-scale note-taking operation out of Notion into plain markdown files, and then realized no existing app fit how his brain worked. So he built one. With AI. All 100K lines of it, in Rust and React, with 3000+ tests and 70 architecture decision records — and he claims he wrote zero of those lines himself. Not "most were generated," not "I scaffolded it." Zero.

That's the real story running underneath this product launch. Tolaria itself is genuinely well-considered — it has a sidebar-based filtering system, WYSIWYG and raw markdown modes, wikilinks with autocomplete, first-class Git integration including an AutoGit mode for people who just want sync without thinking about commits, and a keyboard-first command palette. These aren't novel features on their own, but the combination is coherent. The type system is particularly clever: every note has a type (Project, Topic, etc.) stored as a markdown file, with icons and colors, and you can filter the entire vault by type from the sidebar. That's the kind of opinionated design that actually makes tools usable when your vault has 10,000 notes.

What Luca is really arguing, though, is that context is the competitive advantage in the AI era. His thesis: the people who get the most out of AI are the ones who have organized their knowledge well enough to give AI something useful to work with. Markdown plus Git is his answer because it's portable, version-controlled, and AI-friendly in a way that proprietary formats simply aren't. You're not locked to one AI platform; you can have Claude Code writing research findings into the same vault where your OpenAI assistant is pulling meeting notes, all while you're journaling manually in the same folder structure.

The open-source angle matters too. If your entire professional and personal knowledge base lives in an app, you need to know that app isn't going to disappear or pivot or lock you in. Tolaria's code is public, the format is plain markdown, and Git history is your safety net when you delete something by accident. Luca commits to keeping it free forever and running his own life and business on it — which is about as strong a signal of intent as you're going to get from an indie developer.

Key takeaways:

• Tolaria is a free, open-source Mac desktop app for markdown knowledge bases — think Notion offline, with actual files
• Built in Rust + React, 100K lines of code, 1900+ commits, 3000+ tests — all generated by AI, zero lines written by the author
• Core principles: files, markdown, Git, open source — prioritizing portability and avoiding vendor lock-in
• The type system lets you assign types (Project, Topic, Note, etc.) to notes with custom icons and colors, enabling sidebar filtering
• AutoGit mode handles commits and pushes automatically when the app is idle or you switch focus
• Wikilinks create a knowledge graph with autocomplete across the entire vault
• The repo includes architecture docs, ADRs, and a vision doc — used as context for AI to make better decisions during development
• Available now with a "getting started vault" you can clone on install

Why do I care: From a frontend architecture perspective, the part that actually gets my attention is the ADR collection and the architecture documentation being used as live AI context. That's not a product feature — that's a workflow pattern. If AI-assisted development works best when the AI has rich context about past decisions, then maintaining good architectural documentation stops being a nice-to-have and becomes load-bearing infrastructure. Most teams treat ADRs as something you write once and forget; Luca is using them as a continuous context feed for his AI coding assistant. That's a pattern I'd seriously consider stealing. The fact that it apparently gives the AI >90% accuracy on non-trivial technical decisions is worth paying attention to.

Introducing Tolaria 💧