🦫 BEAVER — Local Voice Assistant

A fully offline, voice-controlled assistant for Windows with an over-the-top dark HUD. Say a wake word + command and it shows maps, videos, weather, and search right inside the app. Speech recognition and text-to-speech run entirely on your machine — no accounts, no API keys, completely free.

Setup (Windows)

You need Python 3.9+ (python.org — tick Add Python to PATH during install).

1. Unzip this folder anywhere.
2. Double-click setup.bat (creates a virtual environment and installs dependencies). Run once.
3. Double-click start_beaver.bat.
4. On first launch, click INSTALL VOICE MODEL — it downloads and installs the offline model inside the app, with a progress bar. Then click ENTER BEAVER.
5. Click WAKE BEAVER, allow microphone access, and start talking.

Say "beaver" + a command

Say	Result
beaver what apps are open	Lists running applications
beaver open windows	Lists open window titles
beaver current tab	Names your active browser tab
beaver recent files	Recent files from Downloads/Documents/Desktop
beaver research electric cars	Searches, compares sources, writes a report (saved + shown)
beaver go to github	Navigates to a site in-app
beaver open maps of Vancouver	Map in the viewport
beaver play lofi beats	Plays a YouTube video in-app
beaver weather in London	Live weather panel
beaver search for SpaceX	Search results
beaver tell me about beavers	Wikipedia
beaver what time is it	Time card
beaver open calculator	Launches calculator
beaver how do you feel	Reports your detected mood
beaver clear	Clears the viewport

Beaver also detects your emotional tone from what you say (shown as a chip in the top bar) and adapts its replies — e.g. if you sound frustrated, it keeps things brief and reassuring. Reports are saved as Markdown in the reports/ folder and displayed in-app.

Build websites & apps by voice

Say things like "make me a simple website for a bakery", "build a calculator app", or "create a portfolio page". Beaver generates a complete, working site and shows it live in the viewport. Test it right there, then just say what to change ("make the background green", "add a contact form") and it updates the same site, remembering context. Finished sites are saved in the generated_sites/ folder. This uses the Anthropic API — create a file named api_key.txt in the Beaver folder with your key inside (from console.anthropic.com). Everything else works without it.

Interrupt

While Beaver is speaking, just start talking — it stops and listens. There's also a 🔇 button to silence it instantly.

Mini mode (always-on-top)

Click 📌 to shrink Beaver into a compact panel pinned to the bottom-right corner, always on top. Click again to restore.

There's also a text box if you'd rather type.

Notes on a couple of features

• Browser tabs: Windows only exposes the active tab's title to other apps, so Beaver reports the focused tab — not a full list of every open tab.
• Files: Beaver scans Downloads, Documents, and Desktop by default. Change SCAN_FOLDERS in core/config.py to adjust.
• Research/reports & site building need internet at runtime; everything else works offline.

Stack (all free / open source)

Python · Vosk (offline speech-to-text) · pyttsx3 (offline TTS) · Eel · sounddevice

Troubleshooting

• requirements.txt not found — setup.bat now runs from its own folder automatically; just keep all the files together.
• Model install fails — check your internet connection and click retry. The app installs the model on first run; you don't need to download anything by hand.
• No mic input — Settings → Privacy → Microphone, and confirm your input device is the default.
• Embedded video/maps blank — some sites block iframes; the command still opens them in your real browser as a fallback.