A voice assistant for Windows with swappable AI personalities: GLaDOS, JARVIS, Wheatley, or one you invent. Say its name, talk to it, and it answers in character, controls your PC, writes your reports, reads your email, and has its own animated core, soundtrack and memory.
| Pick its brain | Claude Code by default, or Ollama (local or cloud models), OpenRouter, or any OpenAI-compatible API. Switch mid-conversation; the conversation comes along. |
| Or type to it | Double-click the orb for a themed, see-through terminal: chat, /sysinfo, /timer, /calc, music, settings, and !PowerShell. |
| Talk to it | Say "GLaDOS" (or "JARVIS", "Wheatley"…) to wake it, then just speak. Say "OK GLaDOS" to interrupt it mid-sentence. Say "you can go" to send it back to sleep. |
| Personalities | Each one has its own voice, wake word, lines, eye colour, sound effects, music, skills and memory. Switch by voice, from the tray, or from the orb. Create new ones by asking. |
| It actually does things | Opens apps, scrolls, types, moves windows, writes Word/PowerPoint/Excel/PDF files, codes in your editor, reads Gmail, Drive, Calendar and Zoom, and searches the web. |
| Safe by default | Risky actions (deleting, sending email, unknown commands) are confirmed out loud: "yes, no, or always?". Dangerous ones are blocked outright. Tokens live in Windows Credential Manager. |
| Fast and private | Wake word and speech recognition run offline (Vosk + Whisper, ~0.5 s per command), replies are spoken sentence by sentence as they're written, and the voices are generated on your PC. |
| Alive | A 60 fps animated core that floats on your desktop, mood-driven voice, idle chatter about its own lore, and a background score that changes with the situation. |
The orb, in eight states per personality: asleep · listening · hearing · working (with the Auto-mode hazard ring) · speaking (with the music badge) · waiting for yes/no · dictation · interrupted
Hover panel, themed per personality: what it heard you say, current model / effort / permissions, and a live step log of its work
Control panel: personality, brain, permissions, voice, audio, music, integrations and appearance; every change applies live
Right-click menu on the orb · tray icon states (Off, Asleep, Awake, Working, Auto)
| Voice | Personality | Soundtrack | |
|---|---|---|---|
| GLaDOS | Fan-made GLaDOS model (Piper) | Cold, passive-aggressive, darkly funny, and helpful anyway | Portal 2: GLaDOS Awakens, Want You Gone, test-chamber score, real game voice lines * |
| JARVIS | Piper British "Alan" | Calm, loyal, dry British butler | Original HUD effects + CC-BY music by Kevin MacLeod |
| Wheatley | Kokoro blend (+ licensed recordings for set lines) | Rambling, nervous, over-confident, means well | Portal 2: his chapters' score, real game voice lines * |
| Your own | Any Piper / Kokoro / Voicebox voice | Whatever you describe | Generated effects + any music you add |
* Portal 2 audio is never included. It's built from your own copy of the game by tools/extract_portal2_assets.py, for personal use.
⬇ Download the latest installer (PersonaCore-Setup-<version>.exe, Windows 10/11 64-bit). Run it, follow the steps, then say the personality's name. It installs for your user only (no admin), brings its own Python and Node, and downloads the voices and models during setup. Details: Getting started.
git clone https://github.com/YousefHlaly/persona-core.git
cd persona-core
pip install -r requirements.txt
python tools\setup.py # models, voices, node packages, settings
python tools\extract_portal2_assets.py # optional: if you own Portal 2
.\Jarvis.bat # tray icon + assistant; then say "GLaDOS"Requirements: Windows 10/11, Python 3.11+, Node.js 18+, Claude Code (signed in), a microphone. Optional brains: Ollama, pi, OpenCode, an OpenRouter key. A GPU is optional (only for Voicebox voices). Full guide: Getting started.
| Getting started | Install, first run, start with Windows |
| Using it | Every voice command: wake, interrupt, dictation, desktop control, music |
| Personalities | Profiles, profile.json, creating your own, memory and skills per personality |
| Brains | Claude Code, Ollama (local and cloud), OpenRouter or any OpenAI-compatible API |
| Voices | Piper, Kokoro, Voicebox, moods, voice effects, licensed recordings |
| Interface | Orb, hover panel, menu, tray, control panel, design system |
| Sound & music | Soundtrack events, theme-music moods, idle music, adding your own |
| Permissions & integrations | Confirmations, rules, Auto mode, Gmail/Drive/Canva, custom MCP servers |
| Skills | Reports, coding, briefings, email, research, PC help, personality creator |
| Measuring | Private usage stats (/stats), the brain benchmark, the speech benchmark |
| Architecture | Processes, threads, files and how they talk to each other |
| Configuration | Every config.json setting |
| Troubleshooting | Mic, voice, orb, Voicebox, GPU memory |
| Releasing | How the Windows installer is built and published |
| Credits & legal | Licences, third-party assets, what's not in this repo and why |
| For coding agents | Where the project stands, how it works, and how to work on it |
Claude Code, Ollama, OpenRouter, pi & OpenCode (brains) · Vosk (wake word) · faster-whisper (speech recognition) · Piper & Kokoro (voices) · Voicebox (optional voices) · GLaDOS voice model · @napi-rs/canvas (orb renderer) · Interface designed with Claude Design.
License: MIT for this project's own code and original assets. Third-party parts keep their own licences (Credits & legal).
Fan project. Not affiliated with or endorsed by Valve or Marvel. Portal, GLaDOS and Wheatley are trademarks of Valve Corporation; JARVIS and Iron Man are Marvel's. See Credits & legal.
0 comments
log in to comment.