What Hold to Talk does with your data

Effective date: March 8, 2026

This privacy policy covers the Hold to Talk macOS app and the Hold to Talk website at holdtotalk.ai.

Hold to Talk App

What the app does with your data

Hold to Talk records microphone audio only while you hold the dictation hotkey. By default, speech recognition runs locally on your Mac using on-device models. Audio and transcribed text are not sent to Hold to Talk servers.

Cloud features (bring your own key)

Hold to Talk requires Apple Silicon and uses local transcription by default. You can optionally enable cloud transcription or cleanup with your own OpenAI-compatible key. When cloud features are enabled:

• Audio or transcription text is sent directly from your Mac to the provider (e.g., api.openai.com or api.anthropic.com). Hold to Talk does not operate any proxy or relay server.
• Your API key is stored in the macOS Keychain for this Mac only, not in plain text or iCloud Keychain.
• Cloud base URLs must use HTTPS and cannot include embedded credentials, query strings, or fragments.
• Cloud requests use an ephemeral network session with no cookies and no disk cache for audio or transcript traffic.
• Hold to Talk never sees, collects, or stores your API key, audio, or transcription text.
• The provider's own privacy policy and data retention rules apply.

Cloud features are off by default.

Audio

• Microphone audio is captured only during an active dictation session.
• When using on-device transcription, audio is processed locally in memory and is not sent over the network.
• When using cloud transcription, audio is sent directly to the provider you configured. It is not sent to Hold to Talk.
• Audio is not stored as recordings by Hold to Talk.

Transcriptions

• Transcribed text is inserted into the app you are using.
• Hold to Talk does not keep a cloud transcription history.
• If optional Apple Intelligence cleanup is enabled, cleanup is performed by on-device system features provided by macOS.
• If cloud text cleanup is enabled, transcription text is sent directly to the provider you configured. It is not sent to Hold to Talk.
• If you explicitly enable local diagnostic logging for troubleshooting, transcript text is redacted in those logs by default.

Local storage

Hold to Talk stores app data on your Mac, including preferences such as hotkey choice, selected model, cleanup settings, downloaded speech models, temporary app state needed for onboarding and operation, and optional local diagnostic logs only if you enable them for troubleshooting.

Hold to Talk does not intentionally store audio recordings or a server-side transcript history.

Network activity from the app

The app may make limited network requests for product delivery features, including downloading speech models from supported model hosts such as Hugging Face and checking for direct-release app updates through Sparkle when using the non-App-Store version.

If you enable cloud transcription or cleanup, the app also makes requests to the provider's API (e.g., api.openai.com, api.anthropic.com) using your own API key. These requests go directly to the provider -- Hold to Talk does not relay or log them. Provider error response bodies are not shown or written to diagnostics because they can contain request details.

App analytics and tracking

Hold to Talk does not include in-app advertising, third-party analytics SDKs, or app telemetry that tracks what you dictate.

Website

Website analytics

The Hold to Talk website uses Google Analytics to measure site traffic and usage. When you visit the website, Google Analytics may collect information such as pages viewed, approximate geographic region, browser and device information, referral source, and basic interaction or session data.

This information helps understand website usage and improve the site. Website analytics do not include your dictation audio or transcription text from the app.

Website hosting and downloads

The website and release assets may be served through third-party infrastructure such as GitHub Pages, GitHub Releases, and other download providers. Those services may receive technical information such as your IP address, user agent, and request logs as part of normal web delivery.

Permissions

The app may request these macOS permissions:

• Microphone for voice input
• Accessibility for text insertion and interaction support
• Hold to Talk does not explicitly request Input Monitoring. Regular hold shortcuts are registered with macOS; modifier-only shortcuts use Accessibility-trusted modifier-state events.

These permissions are managed by macOS and can be revoked at any time in System Settings.

Third Parties

Depending on how you use Hold to Talk, third-party services may be involved:

• Google Analytics for website traffic analytics
• GitHub for website hosting, source code, and release downloads
• Hugging Face for speech model downloads
• Sparkle for direct-release update delivery
• OpenAI for cloud transcription and/or text cleanup (opt-in, bring your own key)
• Anthropic for cloud text cleanup (opt-in, bring your own key)

Their handling of technical request data is governed by their own policies.

Children

Hold to Talk is not directed to children, and Hold to Talk does not knowingly collect personal information from children through the app.

Changes to this policy

If this policy changes, the updated version will be posted on the website and in this repository with a new effective date.

Contact

If you have privacy questions, open an issue at github.com/jxucoder/hold-to-talk/issues.