01 The story
In 1997, Microsoft shipped Microsoft Agent — a framework for putting animated characters on the desktop that could speak, listen, and respond. Merlin was one of the originals. Genie, Robby, Peedy too. Clippy got famous; the wizards got forgotten.
The whole thing was retired in 2009. Speech recognition was bad. The characters didn't really understand anything. The dream of an actual companion on your desktop got quietly filed away with WordArt and Comic Sans.
"What if Merlin came back, but this time he could actually think?"
— The whole pitch, basicallyThat's what this is. The same wizard. The same sprite engine. The same desktop-floating wave-when-you-look-at-him energy. But this time the brain is a modern language model — yours, on your machine, or one you pay an API for. He can see what's on your screen. He can search the web. He remembers what you told him last time.
It's a tribute. It's a working desktop app. It's open source. Microsoft has not endorsed any of this.
02 A brief history of the wizard
03 How it works
Merlin is an Electron app. Standard desktop chrome, except the chrome is mostly transparent and the contents are a wizard. The sprite engine is clippyjs — a wonderful piece of preservation that ports the original character animation files to the browser. Anything that runs in clippyjs runs here, which means you can drop in Clippy, Bonzi, Rocky, Genie, or your own custom sprite sheet.
The brain is whatever you point it at. The Vercel ai SDK handles the streaming and tool-use plumbing, so swapping between Anthropic, Groq, OpenRouter, MiniMax, Ollama, and the self-hosted Hermes Agent is a matter of changing one dropdown.
04 Author
05 License & credits
MIT License
Free to use, fork, modify, redistribute. Attribution appreciated but not required.
Microsoft Agent and the Merlin character are trademarks of Microsoft Corporation. This project is a community tribute and is not affiliated with, endorsed by, or sponsored by Microsoft.
See the full license →