Pati

v0.4.4.2 is out

A desktop companion that lives on your screen. It walks along the taskbar and the top edges of your windows, falls off the edge and flies when you throw it. It talks with a local AI on your own computer, remembers you and gets things done when you ask. The old desktop assistants, rebuilt for today: no ads, no tracking, and the AI runs on your machine.

Windows · Electron · TypeScript · MIT

Pati's character Tungarix in three poses on a night-blue desktop: standing on the top edge of the left window, falling through the air in the middle, playing guitar on top of the right window.
Tungarix on top of the windows: standing, thrown, playing guitar.

Lives on your screen, talks with you.

  • Lives on your desktop

    A small character above every window. It walks along the taskbar and window edges and falls off them; pick it up and throw it, and it dusts itself off after landing.

  • Local AI

    Connects to llama.cpp, Ollama or LM Studio on your computer and looks on the usual ports by itself. Each character has its own personality.

  • Remembers you

    It keeps a brain: your daily chats, a summary of each day and the facts you shared, as plain Markdown in a folder. Open it in Obsidian and edit what it knows.

  • Gets things done

    "Open Spotify", "turn the volume down", "take a note", "how's my battery?". It asks before acting, and you can say "always".

  • Talk by voice

    Press a shortcut and speak: whisper.cpp turns it into text on your computer, and the answer is read aloud.

  • Looks at your screen

    Say "explain this error": it asks first, takes one screenshot and shows it to a vision model. The image is never saved.

  • Speaks up when it matters

    A short morning briefing, a nudge after 90 minutes without a break, a word when it is very late. Quiet in focus mode.

  • Make it yours

    Add your own character with a folder of PNG frames and one JSON file. Turkish and English, sized from 50% to 200%.

Technical summary.

Version
v0.4.4.2
Platform
Windows 10 and 11
Licence
MIT — open source
Stack
Electron · TypeScript
AI
Local: llama.cpp, Ollama, LM Studio
Price
Free

What we deliberately left out.

A project without an out-of-scope list is a project with infinite scope. Keeping this list short is a decision — quietly growing it is not.

  • Cloud AI and accounts
  • Telemetry and ads
  • Shipping an AI model of our own (you pick and install the model)
  • Mac and Linux versions (for now)

Installing

The installer is not code-signed, so Windows SmartScreen asks "Run anyway" the first time. The code is fully open — you can read what it does before downloading. Chat needs a local AI server on your computer (a model of about 8-9B parameters is a good start on an 8 GB GPU); voice needs whisper.cpp. Tools that reach outside the app (clipboard, weather, Wikipedia) stay off until you turn them on.

Install steps and source code →