Skip to content
ABxAGPublic

About

Resources

Stars

2 stars

Watchers

2 watching

Forks

Latest commit

 

History

189 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ABxAG logo

ABxAG

Current version: v1.1.29 · changelog

Your private 3D AI desktop companion.
Talks with you by voice, sees your screen when you allow it, controls apps and files — powered by your own Gemini API key. No cloud account. No subscription.

CI status Latest release Windows 10/11 Linux experimental Bring your own Gemini key

⬇ Download · Roadmap · Website · All releases


✨ Features

Era What you get
🎙 Voice companion Real-time conversation via Gemini Live, “Hey AB” wake word, reminders & alarms, quiet hours
👁 Eyes Screen vision + OCR; screen text is always treated as untrusted data
🖐 Hands 96 local PC tools (apps, files, input, browser, clipboard) gated by confirmations, an emergency stop and a permission board
🧠 Memory & mind Local cognition runtime: episodic memory, goals, learned skills, curiosity and presence
🧍 Body three.js character engine — import any .pmx / .vrm / .glb / .fbx / .obj model, .zip or folder (Character Studio included)
🖱 Reliable hands She drives your physical mouse with exact text labels, pauses the moment you touch the PC, and every risky action goes through confirmations + emergency stop
🔄 Auto-update Hybrid updater: fast differential installs for the current version, one-click switch to any older version, mirror-feed fallback

⬇ Download

Prefer one click? Get everything at https://abxag.absup.dev/DownLoad/ — it auto-detects your OS.

Platform Build Status
Windows 10/11 64-bit Setup (installer + differential auto-updates) ✅ Recommended
Windows 10/11 64-bit Portable (single file, guided updates) ✅ Available
Linux x64 AppImage (chat/voice/memory UI; no PC control) 🧪 Experimental
macOS — 🔜 Planned

🚀 Quick start

Just use it: install the Setup .exe → paste your free Gemini key (from https://aistudio.google.com) → Import my character → say “Hey AB”.

Run from source:

npm install
pip install -r desktop_agent/requirements.txt
npm run dev        # → http://localhost:3000
npm run app        # desktop app (Electron + companion)

The first screen asks for your Gemini API key — stored encrypted with Windows DPAPI in your user-data folder, never in the project. Optional: copy .env.example to .env for dev settings (never put a real key in a file you share).

Build the public installer:

pip install pyinstaller
pyinstaller desktop_agent.spec --distpath agent_dist --workpath agent_build
npm run dist:public   # → release-public/ABxAG-Setup-<version>.exe

npm run dist expects a character in assets/characters/ (not included); use dist:public.

🛠 Tech stack

  • UI: React 19 · Vite 6 · Tailwind 4 · Three.js (MMD/PMX)
  • Backend: Node + Express + WebSocket (server.ts), Google GenAI Live bridge
  • Shell: Electron 43 (desktop companion, tray, crash recovery)
  • Agent: Python FastAPI desktop tool server (UI Automation, input, files, apps)
  • Brain: local event-driven cognition runtime (cognition/: memory, goals, skills, presence)
  • Site: dependency-free static pages in website/ (Cloudflare Pages)

🗺 Project map

Folder What it is
src/ React UI, 3D character engine, companion renderer
electron/ Desktop shell: windows, tray, companion brain, updater, crash recovery
server.ts, runtime/ Local backend (Express + WebSocket), Gemini Live bridge
agent/ PC-control planner, actions, safety checks
desktop_agent/ Python tool server (UI Automation, input, files, apps)
models/ Model router (Gemini and others)
cognition/, memory/, presence/ Memory, initiative, presence
settings/, permissions/, secrets/ Settings, permission engine, DPAPI key store
character_import/, voice/ PMX importer + capability report, live voice tools
website/ Public site (Home, DownLoad, RoadMap, release feed)
conductor/, PRD/ Architecture tracks (spec + plan) and per-feature PRDs
tests/, desktop_agent/test_*.py Agent/character/Python test suites

🤝 Contributing

  1. Work on the DEV branch (never commit straight to it — open a PR).
  2. Branch naming: feat/<x>, fix/<x> or chore/<x>; merge only when CI is green.
  3. Definition of done: npm run lint clean, relevant tests green (npm run test:agent, test:characters, test:python), website changes verified with desktop + mobile screenshots and zero console errors.
  4. Bigger features get a short PRD in PRD/ + a conductor track in conductor/tracks/<id>/ (spec.md, plan.md, registered in conductor/tracks.md).
  5. Never commit secrets (.pfx/.p12/keys are gitignored) or licensed character models.

🔒 Privacy & safety

  • Your Gemini key never reaches the web page; all AI calls go through the local backend.
  • High-risk actions (sending messages, deleting, installing, purchases) ask for confirmation.
  • Websites, documents and screen text are treated as untrusted data.
  • Stop everything any time with the emergency shortcut (default Ctrl+Alt+Shift+S).
  • The repo ships no API key, no character model, no personal data — see “What is NOT included” below.

📦 What is NOT included (on purpose)

  • No API key. You use your own free Gemini key from https://aistudio.google.com.
  • No 3D character model. The models used during development have licences that forbid redistribution. Import your own MMD model (.pmx, or a .zip / folder containing one) in the app: the first screen offers Import my character, or use Settings → Character → Character Studio. Only use models whose licence allows your use.
  • No personal data, logs, memories or settings.

🔗 Links

About

Resources

Stars

2 stars

Watchers

2 watching

Forks

Releases

Packages

Used by

Contributors

Languages