Built for the copilot Windows promised and never shipped.
Pluma is what happens when you get tired of waiting for an AI that actually works on your desktop.
Why this exists
Microsoft announced Copilot. It was supposed to be everywhere — in your taskbar, in your apps, understanding your screen. What shipped was a chatbox that opens a browser tab.
Meanwhile, the real work still happens across dozens of apps, clipboards, terminals, and browser tabs. Nobody built a tool that actually lives between those things — not inside any one of them, but above all of them.
Pluma is that tool. One keystroke. Transparent overlay. Sees your screen, reads your clipboard, runs commands, browses the web, and remembers what you've said. Then gets out of the way.
Design principles
Zero chrome
No persistent window. No dock icon. No notification badges. Press Alt+P — it's there. Let go — it's gone.
Local first
Memory lives in SQLite on your machine. API key in Windows Credential Manager. Nothing phoned home except your AI request.
Actually autonomous
Not a chatbot that tells you what to click. An agent that takes the screenshot, finds the button, and clicks it for you.
Writer's instinct
Ghost-writing is a first-class feature, not an afterthought. Pluma understands that most AI use is editing, not creating from scratch.
The stack
Pluma is built on Tauri v2 — a Rust + WebView framework that produces tiny, fast, native Windows apps. The UI is React + TypeScript. No Electron, no 200MB runtime.
What's next
Pluma is v0.1. The agent is capable but rough. The memory is keyword-based, not semantic. The UI is minimal by design but has room to breathe.
What's planned: voice input, scheduled tasks ("remind me at 9am"), vector embeddings for smarter memory, and a plugin system so you can add your own slash commands and tools.
The repo will be public soon. Until then — use it, break it, tell me what's wrong.