← All cases

Open Source

OSW

Open Source Whisper: local, privacy-first voice-to-text

AI Whisper Python Open Source Privacy

About OSW

HAVEN Intelligence built OSW to demonstrate that AI tools do not need to be expensive cloud services. OSW runs locally with OpenAI's Whisper models and reflects our focus on practical, accessible AI solutions.

The concept is simple: press a hotkey, speak, and text appears at the cursor. All inference happens locally on the user's machine. No data is sent to the cloud, no subscription is required, and no telemetry is collected.

How It Works

OSW uses faster-whisper with int8 quantization to run OpenAI's Whisper models efficiently on standard hardware. The result is voice-to-text that works without an internet connection.

Step 1

Activate with hotkey

OSW runs in the system tray and listens for a global hotkey. A single press starts recording, regardless of which application is in the foreground.

Step 2

Local inference

The audio recording is processed locally with faster-whisper. No data leaves the machine. Users choose between five quality tiers, from the smallest model at 336 MB to the largest at 3 GB.

Step 3

Text at cursor

The transcribed text is automatically inserted at the cursor position in the active application. No copying, no switching between windows.

Key Features

5 tiers
from 336 MB to 3 GB
100% local
no data leaves the machine
  • Local inference with faster-whisper and int8 quantization
  • Five quality tiers: tiny, base, small, medium, and large
  • System tray application with global hotkey
  • Privacy: audio and text always stay on the user's machine
  • Open source under MIT license

Source Code

OSW is open source and available on GitHub. Download it, use it, or contribute to the project.

View on GitHub

Need practical AI solutions?

Book a free AI Analysis and find out how AI can make your business more efficient.

Free AI Analysis

No commitment. Response within 24 hours.