OpenWhispr
Open-source dictation that keeps your voice on the devicePress a hotkey, speak, and text lands in any app
Highlights
What you can do with it
- Reply to a colleague in Slack by holding the hotkey and speaking, without switching windows
- Dictate meeting notes in Notion and let the agent turn the recording into structured minutes
- Write blog paragraphs inside VS Code on a flight, with no network available
- Dictate code comments, documentation and commit messages in an IDE or terminal
How to use it
- 1 Download the installer for macOS, Windows or Linux from openwhispr.com, or build it from the GitHub source
- 2 On first launch choose a processing mode: local for full privacy and offline use, or your own API key for cloud models
- 3 Set a global hotkey and pick hold-to-talk or toggle-to-talk recording
- 4 Download a local model size that fits your hardware, from Tiny up to Turbo
- 5 Press the hotkey in any text field, speak, and release or press again to insert the text at the cursor
- 6 Add custom terms and correction rules, and use agent mode for instructions such as polishing a paragraph
Who it is for
Pros
- Core features are free and open source under the MIT license
- Local models keep audio on the device and work with no internet connection
- Works across macOS, Windows, Linux and iOS with the same hotkey workflow
- Covers more than 100 languages and improves on your own vocabulary over time
Limitations
- Local accuracy and speed depend on your hardware, and larger models are noticeably slower
- The weekly cloud quota of 2,000 characters is small, so heavier cloud use needs a paid plan or your own API key
- Recognition still makes mistakes in noisy rooms or with heavy accents, so text needs a read-through
- Free quotas and licence terms can change, so check the official page before relying on it daily
What OpenWhispr is
OpenWhispr is an open-source voice input tool that turns speech into text on your own computer or phone. It is built on speech models such as OpenAI Whisper and NVIDIA Parakeet, and by default the recognition runs locally, so the audio does not leave the device. The code is published on GitHub under the MIT license and the core features are free to use. Beyond local models, you can connect your own API keys from providers such as OpenAI, Anthropic or Google Gemini, or use OpenWhispr Cloud, which includes a weekly free quota of 2,000 characters. The app runs on macOS, Windows, Linux and iOS, and it types into any text field instead of locking you into one editor.
What you can do with it
- Dictate anywhere: hold a global hotkey, speak, and the text appears at the cursor in Slack, VS Code, Chrome, Word, Notion or a terminal.
- Transcribe offline: load a local model in a size from Tiny to Turbo and keep working with no network connection.
- Give spoken instructions in agent mode, such as tidying up rough notes or drafting an email to a colleague.
- Work across more than 100 languages with automatic detection and switching.
- Teach it your vocabulary so it remembers names, jargon and technical terms such as Kubernetes or PostgreSQL.
Who it is for
- Writers and journalists who want to capture ideas at speaking speed and keep unpublished drafts private.
- Developers dictating code comments, documentation and commit messages in an IDE or terminal.
- Managers who record meeting points and turn them into structured notes or email drafts.
- People in medical, legal or finance work where client data must stay on the device.
- Students, educators and travellers who need dictation without a reliable internet connection.
What to watch out for
Local quality and speed depend on the model size you pick and on your hardware. Larger models are more accurate but slower, and the smallest ones can struggle with accents or specialized vocabulary. The cloud path is convenient, but the weekly free quota of 2,000 characters is small, so regular cloud use means bringing your own API key or paying. Recognition is still imperfect in noisy rooms, so read the text before you send it, and expect to spend a few sessions tuning custom terms and agent instructions. Free quotas and licence terms can change over time, so check the official site before you build a daily habit around it.
FAQ
Is OpenWhispr free to use?
The core features are free and open source under the MIT license. OpenWhispr Cloud adds a weekly free quota of 2,000 characters, and you can also connect your own API keys.
Does my voice leave my device?
Not by default. The local model processes audio on the machine. Cloud models are only used if you choose that mode or supply your own API key.
Which platforms are supported?
macOS, Windows and Linux desktop clients are available, along with iOS support.
Do I need an internet connection?
No, as long as you have downloaded a local model. An internet connection is only required for cloud models or for syncing options.
How many languages does it recognise?
More than 100, including English, Spanish, French, German, Chinese, Japanese and Arabic, with automatic detection.
Why it is free
What the official page says： 免费额度完全免费免费
Free limits, trial length and features are decided by the official site and can change at any time.
