With AI, Markdown has become my format for everything: product specs, blog posts, notes... Until now, I read it in preview mode in my IDE. But it lacked reading comfort, especially for long specs. And when I wanted to continue reading on the subway, there was no simple solution.
I looked for a beautiful Markdown reader that works on all my devices and just displays my .md files well. I couldn't find what I was looking for. So I built it.
I am launching Fude.md, a Markdown reader designed for reading comfort, synchronization across devices, and made for the AI agent era.
Codex: from brief to prototype in record time
After my experience with SIZR where I created a MacOS app in a few minutes, I once again used OpenAI's Codex to bootstrap Fude.md.
My workflow is now well-established: I discuss with AIs to write and refine a clear brief of what I want (in Markdown of course). Then Codex takes care of the project structure, architecture, and code, assisted by Claude Opus for review and design. In a few hours, I had a functional prototype with Markdown rendering, file navigation, and the app's foundations in place.
What would have taken several days of setup was done in one session. Vibe coding continues to blow me away with every new project.
I then spent a lot of time adding all the desired features and ensuring code quality, with the help of AIs. But we're talking days, whereas without AI it would have taken me weeks or months.
My stack: Nuxt, Electron, and Convex
Nuxt is my classic stack, the one I use for all my web projects (this site, Begonia.pro, my internal tools). I know the framework well, which allows me to move quickly on the application part; official skills exist so AIs can code cleanly.
But most importantly, Fude.md allowed me to test two technologies I had been wanting to explore for a while:
- Electron: to offer a native desktop application. This is my first Electron app, and being able to reuse my Nuxt code to create an installable app for Mac and Windows is a real asset. The experience of a Markdown reader as a native app is much smoother than in a browser.
- Convex: Supabase's challenger, a serverless backend oriented towards real-time connections. This is what allows Fude to effortlessly sync your documents between desktop and mobile versions, while keeping your files on your drive by default. No more tinkering with WebSockets or managing a server.
This is exactly what I love about side projects: they serve as a testing ground for technologies that I can then offer to my clients in my Product Engineering missions.
Illustrations: Claude Opus and Gemini for animated SVGs
For Fude.md's homepage, I wanted animated SVG illustrations: lightweight, elegant, and consistent with the project's visual identity.
I used Claude Opus and Gemini to generate these visuals. The process was iterative: I described the concept in natural language, the AIs proposed SVG code with CSS animations, and I refined it until I got a smooth and harmonious result. It's not perfect yet, but for a first test, I'm happy with how it looks.
The result: animated vector illustrations, lightweight (a few KBs). Just like for the creation of ANVL's Pop-art logo with Nano Banana, AI has become a real lever for creating visual assets, in addition to code and writing assistance.
Fude.md features
Fude is a reader, not a Markdown file editor. You keep your favorite tool for writing (VS Code, Cursor, Obsidian), and you use Fude to read comfortably. Here is what the application offers:
- Elegant Markdown rendering: typography and layout designed for long-form reading, with customizable rendering.
- Connection to your sources: local files, GitHub, or Google Drive, grouped into projects without moving your documents.
- Multi-platform: desktop (Mac, Windows) and mobile (iOS, Android) applications planned.
- Seamless synchronization: your documents and preferences synced across all your devices, with encryption. Local-first by default.
- Mermaid support: native rendering of Mermaid diagrams directly in your files.
- Built-in MCP server: described below, to connect your notes to your AI agents.
- Custom themes: to adapt to your reading preferences.
A reader ready for the AI agent era
From launch, I integrated an MCP server (Model Context Protocol) into Fude. An MCP server is an open standard that allows AI agents to connect to external data sources. It's a bit like an API between your tools and your AIs, with pre-defined prompts.
Practically, this means you can connect your Fude notes to any compatible AI agent (Claude, ChatGPT, Cursor, and many others), and it will be able to read your documents directly.
The use cases are numerous:
- Analyze your writing style so the AI writes new content in the same tone.
- Leverage your existing notes as a context base to generate articles, emails, or specs.
- Query your documents in natural language: "summarize my meeting notes from this week" or "what are the open points in my product specs?".
Everything runs locally on your machine. No data passes through external servers. For someone handling product specs or confidential notes, this is a non-negotiable point.
Conclusion
Fude.md is a project that combines my desire to solve a real everyday problem with the joy of exploring new technologies. Electron for the desktop, Convex for the real-time database, and AI at every step: Codex and Claude Code for the code, Claude Opus and Gemini for the design.
The application is currently in private beta. If you work with Markdown on a daily basis and are looking for a true reader for your notes, specs, and documentation, I invite you to sign up.
👉 Sign up for the Fude.md private beta
And if you have a web or desktop application project and want guidance from idea to product, check out my Product Engineering services.