Niki Desktop · v2.1.0

Install Niki Desktop

A small local helper that runs voice transcription on this PC instead of streaming audio to Google. POPIA-clean, GPU-accelerated, free, and your patient audio never leaves the room.

Download Niki Desktop v2.1.0 · single Python file, self-installs deps on first run

Or download the platform installer (zip with launcher)

Download v2.1.0 for Windows 10/11 Download v2.1.0 for macOS 12+ Download v2.1.0 for Linux (Ubuntu / Debian / Fedora)

Auto-detected: Windows 10/11 · pick another above if needed. ZIP bundles a click-to-launch script that runs the helper for you (no command line).

Unzip anywhere

Right-click the downloaded zip, "Extract All". A folder named niki-desktop appears.

Double-click the launcher

Run niki_desktop.bat. First run downloads ~750 MB of voice + GPU libs (3–5 min on broadband). Subsequent launches are instant.

Leave the window open

The helper listens on ws://localhost:9876. The window must stay open while you use the dashboard. Press Ctrl+C to stop.

Reload the dashboard

The mic pill in the topbar should turn green with "GPU" or yellow with "CPU". The orange "browser fallback" banner goes away.

What gets installed

✓ faster-whisper — local speech-to-text, runs on CPU int8 or NVIDIA GPU fp16.
✓ Piper TTS — neural text-to-speech voice (~60 MB model). Replaces robotic browser voice.
✓ NVIDIA CUDA libs — cuBLAS + cuDNN. ~250 MB. Skipped harmlessly on non-NVIDIA machines.
✓ Local WebSocket server — listens on localhost:9876 only. No outbound network calls.
✓ Log file — plain text at ~/.niki-desktop.log for troubleshooting.

Test connection

Once you've launched the helper, click below. The dashboard will probe ws://localhost:9876 and report green if Niki Desktop is responding.