Privacy
This is the part of Yappy we care about most, so it is written plainly rather than legally. If something here is unclear, mail [email protected] and we will fix the wording.
The short version
Your audio never leaves your Mac. Your screen never leaves your Mac. The parts of Yappy that can talk to a server are the ones you choose to switch on, and each of them is named below.
The Mac app
Dictation is local, always. Audio goes from the microphone into whisper.cpp running on your own machine. No bytes of audio are uploaded, in any configuration, including the paid ones. There is no "improve the product with your recordings" setting, because there are no recordings to improve it with.
Screen context is local, always. When Screen Context is enabled, Yappy reads the frontmost window with Apple's on-device OCR and discards the screenshot immediately. The image is never written to disk and never uploaded. Turn it off in Settings → Dictation and the feature stops running entirely.
Cleanup can be local. A small language model tidies the raw transcript before it is typed. Point llm.provider at ollama (or any OpenAI-compatible local server such as LM Studio, vLLM, or llama.cpp) and that step runs on your machine too, which makes the whole dictation pipeline work with the network off. Choose a hosted provider instead and the text of that one transcript is sent to it.
Agent Mode talks to Anthropic. Holding Right ⌘ starts an agent session that runs through the claude CLI, so the task text and the screen-derived context needed to carry it out are sent to Anthropic's models. This is the one part of Yappy that requires the network by design. It is off unless you use that key.
Usage analytics are optional. Yappy reports anonymous product analytics — which features are used, whether a dictation succeeded, app and OS version — to PostHog. Transcripts, audio, screen contents, filenames, and prompts are never included. Turn it off in Settings with Share usage analytics.
Crash and diagnostic logs stay local unless you send them yourself with Send Feedback, which shows you what it is attaching first.
This website
yappy.biz is a static site on Cloudflare Pages. It sets no advertising cookies and runs no third-party trackers beyond product analytics (PostHog), which records page views and download clicks. The public REST API and MCP server serve read-only public data — release metadata, docs, FAQ, pricing — and require no account, so there is nothing to identify you with.
What we never do
- Upload your audio.
- Upload your screen.
- Sell, rent, or share your data with data brokers.
- Train models on your dictation.
- Require an account to use dictation.
Your controls
| What | Where |
|---|---|
| Screen context off | Settings → Dictation → Screen Context |
| Cleanup fully local | Settings → Engine → provider ollama |
| Cleanup off entirely | Settings → Dictation → AI Cleanup |
| Agent Mode off | Settings → Agent → Enabled |
| Analytics off | Settings → Share usage analytics |
| Microphone, Accessibility, Screen Recording | macOS System Settings → Privacy & Security |
Revoking a macOS permission takes effect immediately; Yappy degrades to the features that permission is not needed for rather than nagging you.
Data requests
Because dictation is local and the site requires no account, there is usually no personal data of yours for us to hold. If you have written to support, mail [email protected] and we will delete that correspondence and any analytics identifier you name.