OpenWhispr

OpenWhispr tackles a simple arithmetic problem: most people type around 40 words per minute but speak closer to 150. The app listens through a configurable hotkey, streams your speech through a speech recognition model, and pastes the result into whatever text field is focused — ChatGPT, Claude, Cursor, Slack, Docs, Gmail, or an IDE. Because it operates at the operating-system level rather than inside one app, it becomes a universal input method rather than a single-purpose tool.

The distinguishing choice is where the audio goes. Local mode runs on-device models such as Whisper or NVIDIA Parakeet entirely on your hardware, so nothing leaves the machine. Cloud mode is optional and lets you plug in your own provider keys, with recordings deleted once transcription finishes. The project is published under the MIT license with a public GitHub repository, and it publishes compliance credentials including SOC 2 Type II, ISO 27001, HIPAA and GDPR alignment — unusual for a free utility, and aimed squarely at teams whose security reviewers would otherwise block a dictation tool.

Beyond raw transcription, OpenWhispr layers on voice commands and AI cleanup. You can say things like "clean this up" or "draft an email to Mike" and the model restructures the rough transcript accordingly. A custom vocabulary system learns names and domain jargon such as product names, internal services or database identifiers directly from corrections you make, which is where generic dictation tools usually fall apart. Language support spans 100+ languages with automatic detection, so switching languages mid-conversation keeps working.