WritHer FAQ

Yes. WritHer is free, open-source software under the MIT license. There is no subscription, paid dictation tier, word quota or WritHer account. You can inspect, modify and share the source code.

Dictation works offline after the speech model has downloaded. The voice assistant also works offline when its language model and server run on your PC. Download the models and install dependencies before going offline. If you configure a remote assistant server, assistant requests go to that server.

A Windows 10 or Windows 11 PC, a working microphone and an internet connection for the initial downloads. Dictation can run on the CPU; a GPU is optional. Larger speech and assistant models need more memory and processing power. Python 3.11 or later is needed only when running from source.

No. Dictation uses faster-whisper and works without a language-model server. Assistant mode requires Ollama or an OpenAI-compatible local server, such as llama.cpp or LM Studio, with a compatible model. The two modes can be used independently.

Download the Windows ZIP from GitHub Releases, extract it and run WritHer.exe. Wait for the speech model to finish downloading and loading. Focus a text field, hold AltGr, speak and release. Follow the setup guide below for the assistant and troubleshooting.

Recognition uses Whisper multilingual models. Auto detects the language of each recording; Settings also lets you select English, Italian or German explicitly. The interface is available in English, Italian and German. Recognition quality varies with the language, model, recording and microphone.

WritHer pastes the transcript into the focused Windows application, so it works with most text fields that accept pasted text. Focus the intended field before recording. Applications that block paste or run at a different privilege level may need additional attention. The default clipboard behavior saves and restores your previous content.

Hold AltGr for dictation. Hold Ctrl + Alt + R for the voice assistant. Release to finish. You can change the shortcuts in Settings and choose between hold-to-record and toggle recording. Your saved settings take precedence over defaults.

Yes. Enable Symbol & spelling mode for spoken English symbol names such as forward slash or semicolon. Spoken letters and numbers can be combined into identifiers. Custom vocabulary maps your spoken phrases to written text, while priming terms help the model recognize names and jargon. This is dictation, not automatic code generation.

Use a clear microphone signal, reduce background noise and choose the correct recognition language. Try a larger speech model if your PC can run it comfortably. Add useful names and specialist terms to the vocabulary and priming list. Speed and accuracy depend on your hardware, model and recording.

Speech recognition runs on your PC. Notes, reminders and settings are stored locally. The app has no WritHer account or telemetry service. Assistant requests are sent to the provider URL you configure: keep it local for an entirely local setup. Initial model downloads and installation require network access.

The current application supports Windows 10 and 11. Contributions towards macOS and Linux support are welcome. The project uses Windows-specific integrations, so a port needs implementation and testing rather than just packaging.

Absolutely. Share an idea, report a reproducible issue, improve a translation, test a release or help another person get started. Developers and designers can also improve the code, widget and documentation. Everyone is welcome; start with the contribution guide or open an issue.

WritHer focuses on local speech recognition, a configurable local assistant and source code you can change. Wispr Flow offers a managed product across desktop and mobile platforms, with free and paid plans. The comparison page links to current official information instead of relying on old pricing or word limits.