# Privacy & data flow

> Understand how Voice Key processes audio locally, stores history, and sends text only for optional cloud refinement and translation.

## Local speech recognition

Voice Key processes microphone audio on your device using downloaded models. Local recognition does not upload microphone audio to an ASR provider. Classic and Streaming modes both run locally.

## Optional cloud text processing

When you enable refinement or translation, the relevant transcript or selected text is sent to your configured LLM provider. Dictionary terms may accompany these requests as spelling references. The provider’s own data policies and fees apply. These features do not send raw microphone audio.

## History and model files

Transcript history and downloaded models remain in the application’s local user-data directory. History remains until you delete it. You can search, copy and delete records in the app. Local storage does not imply encrypted storage or protection from other users with access to your device.

## Other network connections

Local-first does not mean the app never connects to the internet. Model downloads, update checks, optional glossary retrieval and configured cloud services use the network. Before a Streaming model download, the app obtains a country code to choose a download source; it does not store or log the public IP address.

## Using dictation offline

Download the recognition model in advance and disable cloud refinement and translation. The core recognition and text-insertion workflow can then operate without an internet connection.

## Open to inspection

Voice Key is MIT licensed. The source repository documents the recognition pipeline and provider integrations. Review the code and report privacy concerns through the project’s GitHub issues without posting personal data or credentials.

## Links

- [Releases](https://github.com/BuildWithAIs/voicekey/releases/latest)
- [Source](https://github.com/BuildWithAIs/voicekey)