Little Prince — Desktop Companion

A tiny Little Prince who lives on your desktop: he walks along window edges, sits down and dangles his legs, climbs and teleports between surfaces, falls asleep when you leave him alone, and can talk to you with a built-in small language model.

Runs on Windows and Linux (X11).

Two editions are shipped:

Edition	Windows	Linux	What's inside
Classic	LittlePrince.exe	LittlePrince	Everything except AI
AI	LittlePrinceAI.exe	LittlePrinceAI	Classic + built-in model, Ollama & API support

Character	Companions

Features

• Lives on your desktop — a transparent, frameless, always-on-top window that never takes focus and never shows up in the taskbar.
• Walks on real windows — reads live window rectangles via Win32 on Windows and X11/EWMH on Linux, and treats their top edges as platforms. He climbs, falls, lands and sits on them.
• Sits & dangles — sits on the edge with his legs hanging over it, minimum 10 seconds so he doesn't get up instantly.
• Teleports — if there is another surface nearby, he occasionally vanishes in a puff of sparks and reappears there (default 30 %, at most every 20 s).
• Sleeps — after 90 seconds without interaction he dozes off with little "z z z" above his head. A click, a drag or simply moving the mouse near him wakes him up.
• Reacts — flails while dragged, panics while falling, bounces and says "Ouch!" after a hard landing.
• Companions — when he sits, one of his friends may join: the fox, the rose (under a glass dome), the lamplighter, the snake, or a baobab tree. Each arrival has two kinds of lines — the companion's own words in the companion's bubble ("Tame me!") and the Prince's remark about it in his own bubble ("The snake is here. Let us not move.").
• Speaks 20 languages — automatic from your system language, or pick one in the settings.
• Says goodbye properly — quotes come from the book, and every line is available in all 20 languages.

Download & run

Grab the build for your platform from Releases (or build it yourself, see below). No Python and no installation required.

Windows — LittlePrince.exe / LittlePrinceAI.exe: double-click.

Linux — LittlePrince-linux-x86_64.tar.gz (classic) or LittlePrinceAI-linux-x86_64.tar.gz (AI):

tar -xzf LittlePrinceAI-linux-x86_64.tar.gz
cd LittlePrinceAI-linux-x86_64
./LittlePrinceAI          # run directly
./install.sh              # optional: add to ~/.local/bin and the app menu

• Settings are stored in %APPDATA%\LittlePrince\settings.json on Windows and ~/.config/LittlePrince/settings.json on Linux, so the app keeps them wherever you move the binary (never contains anything harmful, but can contain your API key — it is not in the repository).
• Portable mode: put an empty settings.json next to the binary and settings will live there instead.
• Start with Windows: press Win+R, type shell:startup, and drop a shortcut to the .exe in that folder.
• Quit: right-click the Prince → Quit.

Linux notes

• X11 is recommended. Window geometry is read through X11/EWMH, which is what lets the Prince walk on top of your open windows.
• Wayland: for security reasons Wayland does not expose other applications' windows. The Prince still runs, but he can only walk along the screen edges.
• A compositor is required for the transparent background (KDE, GNOME, XFCE with compositing, Cinnamon, ...).
• If Qt fails to start with an xcb plugin error, install libxcb-cursor0 (Debian/Ubuntu: sudo apt install libxcb-cursor0).
• The release binaries were built on Ubuntu 24.04 (glibc 2.39), so they run on Ubuntu 24.04+, Fedora 40+, Arch and similar distributions.

Controls

Right-click the character:

Menu item	What it does
Talk (AI)…	Opens the chat window (AI edition only)
Sit down / Get up	Sits until you tell him to get up
Wake up	Wakes him immediately
Say a line from the book	Random quote
Start wandering	Puts him back into wander mode
Teleport	Jumps to another surface right now
Settings…	Character + AI options
Bring to screen	Recovers him if he gets lost
Quit	Closes the app

Mouse: left-click turns him toward the cursor, drag to carry him, double-click for a quote.

Languages

English, Türkçe, Español, Français, Deutsch, Italiano, Português, Nederlands, Polski, Русский, Українська, 中文, 日本語, 한국어, हिन्दी, বাংলা, العربية, فارسی, Bahasa Indonesia, Tiếng Việt

• Default is Automatic (your system language).
• Font fallback is handled per script (CJK, Devanagari, Bengali, Arabic, Persian and so on), so no boxes appear instead of letters.
• The interface (settings and chat) is in English; only the character's speech and the right-click menu are localised.

AI edition

Open the chat with right-click → Talk (AI)…. The model is prepared when the window opens and unloaded when you close it, so nothing runs in the background while you are not chatting.

Three providers (Settings → AI):

Provider	What you need	Notes
Built-in model	One click to download	Runs inside the app with llama.cpp, works offline
Ollama	ollama.com + ollama pull llama3.2:1b	Free, local, offline
OpenAI-compatible	An API key	Works on any machine
Off	Nothing	Quotes-only offline replies

Built-in model

The engine (llama.cpp) is bundled, the model file is downloaded once into %APPDATA%\LittlePrince\models\ on Windows or ~/.local/share/LittlePrince/models/ on Linux, so the binary stays small:

• Qwen2.5 0.5B — ~470 MB, small and fast
• Qwen2.5 1.5B — ~1 GB, noticeably better Turkish and other languages

Downloads resume automatically if the connection drops. Everything runs on the CPU, so it works on any machine — no GPU required.

Cloud endpoints

• OpenAI: https://api.openai.com/v1 — e.g. gpt-4o-mini
• Groq: https://api.groq.com/openai/v1 — e.g. llama-3.1-8b-instant
• OpenRouter: https://openrouter.ai/api/v1
• LM Studio (local): http://localhost:1234/v1

The API key is stored in plain text in your local settings file. Never commit that file — it is already in .gitignore.

Settings

Setting	Description
Language	One of 20, or automatic
Size	0.7× – 3.5×
Walk speed	Step speed
Companion chance	Probability a friend joins when he sits
Teleport chance	Probability of jumping to another surface
Minimum time between teleports	Default 20 s
Minimum sit time	Default 10 s
Sleep / delay	Whether and when he falls asleep
Always on top	Keep the window above everything
AI provider / model / address / key	AI edition only

How it works

• Transparent window: FramelessWindowHint + WindowStaysOnTopHint + Tool with WA_TranslucentBackground — no taskbar entry, no focus stealing.
• Art: every character is drawn with vector code (prince.py, companion.py). No image assets are needed.
• Physics: gravity plus platform checks. The character's support point (feet, or hips when sitting/sleeping) is aligned to the platform's edge.
• Windows: EnumWindows + DwmGetWindowAttribute give the real frame rectangles of open windows; their top edges become platforms.
• Linux: X11/EWMH (_NET_CLIENT_LIST, _NET_WM_WINDOW_TYPE, _NET_WM_DESKTOP) is read through a small ctypes layer; the window manager's frame geometry is used, and panels/desktops/menus are ignored.
• Always on top: on Windows the taskbar also lives in the topmost band, so the window silently re-asserts SetWindowPos(HWND_TOPMOST) a few times per second — but only when something is actually above it. On X11 the same check uses _NET_WM_STATE_ABOVE plus XRaiseWindow.
• DPI: Win32 physical coordinates are converted to Qt logical coordinates, so alignment stays correct at 125 % / 150 % scaling. X11 reports physical pixels and the same conversion is applied.

Building from source

Windows:

python -m pip install -r requirements-dev.txt
python scripts\make_icon.py     # generates assets\app.ico + app.png
scripts\build.bat               # -> dist\LittlePrince.exe + dist\LittlePrinceAI.exe

Linux:

python3 -m venv .venv && . .venv/bin/activate
pip install -r requirements-dev.txt
python scripts/make_icon.py
scripts/build.sh                # -> dist/LittlePrince + dist/LittlePrinceAI (+ .tar.gz)

llama-cpp-python is compiled from source on Linux, so cmake, g++ and make are needed for the AI edition. The chat still works with Ollama/API if you skip it.

Or just one edition: scripts\build.bat classic / scripts\build.bat ai (scripts/build.sh classic / scripts/build.sh ai on Linux).

Smoke test without opening the UI:

$env:LP_SELFTEST="1"; dist\LittlePrinceAI.exe; Get-Content "$env:TEMP\lp_selftest.txt"

LP_SELFTEST=1 dist/LittlePrinceAI; cat /tmp/lp_selftest.txt

Run from source:

python -m pip install -r requirements.txt
python main.py

Use LP_NO_AI=1 python main.py to try the classic behaviour from source.

Project layout

main.py                     entry point (+ self test)
littleprince/
  pet.py                    window, physics, sleep, teleport, reactions, menu
  prince.py                 Little Prince vector art
  companion.py              fox / rose / lamplighter / snake / baobab art + windows
  i18n.py                   20-language strings (quotes + menu)
  features.py               AI on/off detection for the two editions
  ui_font.py                font stack for CJK / Indic / Arabic scripts
  llm.py                    AI backends (built-in llama.cpp, Ollama, OpenAI-compatible)
  chat.py                   chat window + model downloader
  config.py                 settings load/save (Windows / Linux / macOS paths)
  settings_dialog.py        settings UI
  platforms/
    __init__.py             backend selection
    win32.py                Win32 helpers (window rects, z-order, topmost)
    x11.py                  X11 / EWMH helpers for Linux
    fallback.py             screen edges only (Wayland and other platforms)
assets/                     app.ico, app.png, ai_enabled.flag
scripts/
  build.py                  PyInstaller command builder (both platforms)
  build.bat / build.sh      Windows / Linux wrappers
  make_icon.py              generates the icons from the vector art
tests/                      behaviour and screenshot tests
tools/                      art previews, diagnostics, screenshots
docs/                       README preview images
packaging/                  Linux install script and Linux readme
LICENSE                     MIT for the code (story/character rights excluded)

Credits

• Story and characters: Antoine de Saint-Exupéry, Le Petit Prince.
• Quotes are based on the original French text; the English wording follows the classic Katherine Woods translation where well known. Other languages are faithful translations — published editions may differ slightly.
• Built-in model: Qwen2.5 (Apache-2.0), running on llama.cpp (MIT).
• UI framework: Qt for Python (PySide6).

This is a fan project. The Little Prince is not public domain in every country, so please respect the rights of the estate and of the translators.

License

The source code is released under the MIT License.

The license covers only the code. It does not cover the book, its characters, the quotations or the translations — those remain the property of their respective rights holders and are included here as a non-commercial fan tribute. See the scope section at the end of the LICENSE file.

Legal notice

This is an unofficial, non-commercial fan project. It is not affiliated with, endorsed by, or sponsored by the estate of Antoine de Saint-Exupéry, any publisher, or any translator.

The Little Prince (Le Petit Prince) and its characters are still protected by copyright in many countries. The quotes are reproduced as a fan tribute; all rights to the story, characters and published translations belong to their respective rights holders. The English wording follows the classic Katherine Woods translation, which is separately copyrighted.

No commercial use is intended. If a rights holder requests removal, the project will be taken down or renamed.