OpenWhispr sits between the strictly-local apps and the cloud subscriptions: it runs Whisper or NVIDIA Parakeet on your machine when privacy or connectivity matters, and calls out to cloud APIs with your own keys when you want speed. It's MIT licensed, free, and works on all three desktop platforms. The project started in June 2025 and has picked up roughly 3,700 GitHub stars.
At a glance
Platforms
macOS, Windows, Linux
Pricing
Free, MIT licence. Cloud models cost whatever your provider charges
Free tier
The whole app
Processing
Local or cloud, switchable
Models
Whisper Tiny through Turbo, NVIDIA Parakeet, cloud APIs via BYOK
Languages
100+ with auto-detection
Offline
Yes, on local models
Integrations
ChatGPT, Claude, Cursor, Slack, Google Docs, Gmail, Teams, and other text fields
Local and cloud on the same install. Most tools pick a side. Running Parakeet locally for sensitive work and a cloud API for speed, without switching apps, is a practical advantage.
Auditable privacy. MIT licence with the code on GitHub, zero data retention on cloud calls, and a stated commitment not to train on user data without consent. You can verify the first claim yourself.
No telemetry. Transcription history stays local.
Custom dictionary that learns. Corrections feed back into recognition rather than needing manual vocabulary entry.
Voice commands. AI-driven cleanup and text generation beyond straight transcription.
Where it falls short
Independent coverage is thin. Most of what's written about OpenWhispr is either the project's own site or comparison posts published by competing commercial tools. The claims below the surface — the ~150 WPM figure, the accuracy positioning — are the project's own and not independently benchmarked.
Electron. Built on Electron 41, with the memory footprint that implies. Noticeably heavier than a native app like VoiceInk.
BYOK for the good models. Cloud speed means getting an API key and funding it. Convenient it is not.
Young project. Created mid-2025. Fewer sharp edges filed off than Handy, which has an order of magnitude more users finding bugs.
Local accuracy is model-bound. Tiny and Small will disappoint; Turbo and Parakeet need decent hardware.
Pricing
Free under the MIT licence, for personal and commercial use. If you use cloud models you pay the API provider directly.
Who it's for
People who want one dictation app that can be strictly private when the work demands it and fast when it doesn't, across Mac, Windows, and Linux, without paying a subscription. If you want the largest community and the most-tested build, Handy is the safer free pick; OpenWhispr's edge is the local/cloud switch.
Last verified: 9 August 2026.
Similar to OpenWhispr
OpenWhispr: Free MIT dictation that runs local or cloud, your choice per job – Dictation Database
