A local-first, privacy-respecting AI voice assistant for Windows.
Bilingual · Profile-based memory · Modern HUD interface · Standalone .exe

---

📸 Interface Preview

_{JARVIS running in fullscreen mode with animated waveform HUD · Windows 11}

---

✨ What is JARVIS?

JARVIS is a local-first Windows voice assistant designed for real everyday use. It listens continuously, wakes on a trigger word, executes desktop and system commands, and remembers who you are — all without sending your voice to the cloud.

Wake word  ──▶  Command recognized  ──▶  Local execution
                                    └──▶  Online fallback (optional)

Core philosophy:

• 🔒 Voice input/output runs 100% locally
• 🧠 Per-profile persistent memory & preferences
• 🌐 Online only when you need: weather, news, Wikipedia, AI chat
• 🇫🇷 🇬🇧 Full French & English command support
• 📦 Ships as a single standalone .exe

---

🚀 Highlights

🎨 Interface Modern fullscreen PySide6 HUD Animated neural waveform FR / EN / AUTO language buttons Windows startup toggle Live mic level & transcript feed	🧠 Intelligence Wake words: Jarvis, reveille-toi, wake up Context-aware follow-up handling Dual-model speech recognition support Optional AI layer via g4f Auto-recovery from Vosk waveform errors
💾 Memory & Profiles Per-profile notes, tasks, agenda Learnable voice macros Relation memory (Amar est mon ami) Onboarding flow with confirmation Legacy data migration	🖥️ Desktop Power Open / close / install applications Volume, brightness, screenshots Dark/light mode toggle Lock, sleep, restart, scheduled shutdown Xbox Game Bar recording trigger

---

🗂️ Project Structure

MyJarvis/
├── 📄 main.py          ← PySide6 UI, wake flow, main loop
├── 🎙️ listener.py      ← Vosk recognition, mic input
├── 🔊 speaker.py       ← Windows SAPI voice output
├── ⚙️ actions.py       ← Command router & local execution
├── 🧠 brain.py         ← Memory, profiles, reminders, helpers
│
├── model/              ← Vosk speech model (you provide this)
│
└── profiles/
    ├── active_profile.json
    └── <profile_id>/
        ├── profile.json
        ├── memory.json
        ├── notes.txt
        └── tasks.json

---

⚙️ Installation

Requirements

Requirement	Details
OS	Windows 10 or Windows 11
Python	3.10 or higher
Microphone	Any working input device
SAPI Voice	At least one Windows voice installed
Vosk Model	One model folder (FR, EN, or multilingual)

Install dependencies

pip install PySide6 vosk sounddevice numpy comtypes pywin32 pycaw pyautogui psutil pygetwindow wikipedia screen_brightness_control pyperclip reportlab

Optional AI chat layer:

pip install g4f

JARVIS works fully without g4f for all local commands.

---

🎙️ Speech Model Setup

Download a Vosk model and place it in:

model/
├── am/
├── conf/
├── graph/
├── ivector/
└── README

Goal	Recommended model
Best French recognition	French Vosk model
Best English recognition	English Vosk model
Mixed usage	Multilingual Vosk model
Both languages	Set primary + secondary via env vars

---

▶️ Launch

python main.py

On startup, JARVIS:

1. Initializes the Qt HUD interface
2. Selects the best Windows microphone
3. Loads Windows SAPI speaker
4. Loads the active profile
5. Optionally runs onboarding
6. Enters standby mode — waiting for your wake word

---

🖥️ Desktop Controls

Button	Action
FR	Lock replies to French
EN	Lock replies to English
AUTO	Follow the language of the current request
STARTUP	Toggle Windows auto-launch at login
GitHub icon	Open the project GitHub page

---

📦 Standalone EXE Build

powershell -ExecutionPolicy Bypass -File .\build_exe.ps1

Output:

dist/Jarvis.exe

The build bundles the full app, Vosk model, native DLLs, and icon into a single portable binary — no Python installation required on the target machine.

---

🧬 First Run & Onboarding

On a fresh profile, JARVIS starts with a bilingual voice introduction (EN + FR) and asks you to choose your preferred language. That choice persists in your profile.

Onboarding collects:

• Your name
• Personality mode (serieux / drole)
• Startup briefing preference
• Auto-news preference
• PC surveillance preference

Every answer goes through a confirmation step — JARVIS repeats what it heard, and you confirm with oui or reject with non.

To trigger onboarding later:

lance la configuration
configure mon profil
refais la configuration

---

💬 Command Reference

🔔 Wake & Control

Jarvis / reveille-toi / wake up
Jarvis ouvre Chrome
stop / silence / arrete
merci / au revoir / ferme-toi

🌐 Language

parle francais / speak french
parle anglais / speak english
langue auto / language auto

📂 Apps, Sites & Folders

ouvre Chrome / open Chrome
ouvre Telegram / ouvre Steam / ouvre Word
ouvre mes documents / ouvre mes telechargements
ferme Telegram
installe Discord / install Spotify
prepare mon travail

🔊 Audio, Display & Capture

augmente le volume / diminue le volume
volume a 35 / quel est le volume
augmente la luminosite / luminosite a 60
capture d'ecran / take a screenshot
enregistrement video / record video
capture vocale

💻 System

verrouille l'ecran
mets le pc en veille
redemarre le pc
eteins le pc dans 5 minutes / annule l'arret
mode sombre / mode clair
statut du systeme / system status
mon cpu tourne a combien
j'ai combien de ram dans mon pc

📝 Productivity & Files

note ca : acheter du pain
ajoute la tache appeler maman
affiche mes taches
rappelle-moi de boire de l'eau dans 10 minutes
ajoute rendez-vous demain 15h dentiste
agenda aujourd'hui
cree un pdf / cree un CV / cree une facture
cree un fichier Word / cree un dossier Travail sur le bureau
quand je dis mode travail, fais ouvre chrome puis ouvre vscode

🌍 Information & Utilities

quelle heure est-il / what time is it
quelle est la meteo / meteo a Montreal
recherche Albert Einstein / cherche la relativite
actualites / prix du bitcoin
donne-moi un resume de l'histoire de Napoleon
tell me about Alan Turing
choisis entre cafe et the

🖼️ Image Generation

genere une image de chat bleu
create an image of a blue cat
une image de crevettes qui mangent du chewing-gum
[follow-up] un animal / de style cartoon / with sunglasses

🎮 Modes & Automation

mode gaming / mode dev
mode assistant quotidien
dashboard / mise a jour
active le mode conversation continue

---

🌍 Language Support

Feature	Status
French commands	✅ Full support
English commands	✅ Full support
Mixed AUTO mode	✅ Language-aware routing
French voice recognition	✅ Best with French Vosk model
English voice recognition	✅ Best with English or multilingual model
Dual-model recognition pass	✅ Optional via JARVIS_SECONDARY_MODEL_PATH

Note: Command routing is fully bilingual. Recognition quality depends on the Vosk model you install.

---

🔧 Environment Variables

Variable	Purpose	Example
JARVIS_MIC_GAIN	Software mic gain multiplier	2.5
JARVIS_MIC_DEVICE	Force a specific mic by index or name	Realtek
JARVIS_MODEL_PATH	Primary Vosk model path	C:\Models\vosk-model-fr
JARVIS_SECONDARY_MODEL_PATH	Secondary Vosk model (optional dual-pass)	C:\Models\vosk-model-en

$env:JARVIS_MODEL_PATH="C:\Models\vosk-model-fr-0.22"
$env:JARVIS_SECONDARY_MODEL_PATH="C:\Models\vosk-model-small-en-us-0.15"
python main.py

---

🤖 Optional AI Layer

JARVIS uses local logic first. The AI layer activates only when a command is meaningful but unmapped locally.

Use case	Handled by
Common commands & system actions	Local router (always)
Open-ended questions & explanations	AI layer (optional)
Translation requests	AI layer (optional)
Partial command recovery	AI layer (optional)

Guardrails in place:

• JARVIS will not identify itself as ChatGPT or OpenAI
• Image requests prefer the local image path
• Short follow-ups use recent context before falling back
• Raw JSON technical dumps are suppressed in recovery mode

---

🛠️ Troubleshooting

Jarvis hears itself / reacts to its own voice

JARVIS clears the recognition buffer while SAPI is speaking and keeps a guard window after speech ends. If echo still triggers commands:

• Lower speaker volume
• Use a headset
• Move the mic away from speakers

Listener crashed with "Failed to process waveform"

JARVIS auto-resets the recognizer and restarts mic streaming. If this recurs:

• Lower JARVIS_MIC_GAIN
• Reduce Windows mic input gain
• Try another device via JARVIS_MIC_DEVICE
• Use a headset mic

Wrong reply language

Use the FR / EN / AUTO buttons in the UI, or say:

parle francais / parle anglais / langue auto

App or site does not open

JARVIS uses explicit aliases → shortcut indexing → fuzzy match → web fallback. Make sure the app has a Start Menu shortcut or a known URI handler. Built-in aliases include: Telegram, Steam, Discord, Spotify, Word, Excel, PowerPoint, VLC, OBS, Parsec, Rocket League.

Brightness does not change

External monitors typically block software brightness control. This works best on laptop displays.

g4f fails

All local commands remain available. g4f provider instability does not affect local routing.

---

🔮 Roadmap

• Microphone selector directly in the UI
• Persistent reminders across full restarts
• Cleaner command plugin/extension system
• Improved browser automation
• Richer document templates (CV, invoice, report)
• Better natural dialogue memory across longer sessions

---

⚠️ Known Limitations

• Windows only — no macOS or Linux support
• No automatic speaker recognition yet
• AI layer depends on third-party g4f provider availability
• Some app launches require a Start Menu shortcut or URI handler
• Game Bar recording depends on Windows configuration
• Software brightness control limited on external monitors

---

📜 Credits

Creator	benzoXdev
Speech Recognition	Vosk
Voice Synthesis	Windows SAPI
UI Framework	PySide6 / Qt
UI Inspiration	Siri · Iron Man JARVIS HUD
Optional AI	g4f

JARVIS — because your PC deserves a voice.