Yap – OSS on-device voice dictation for macOS with no model to…
Yap — open-source macOS menu-bar dictation that types into any focused app via Apple's on-device Speech framework: no model download, no API key, nothing leaves the machine.
What is the idea?
Hold a key, talk, and it types into whatever app has focus. It rides Apple's built-in speech stack, so it starts instantly, works offline, and the whole tool stays a thin, hackable Swift wrapper (101 HN points, 41 comments).
Who would use it, and what do they get?
Non-developer user: anyone who writes more than they type comfortably — a bilingual professional, someone with RSI, a person drafting messages while walking. Finished thing: text already in the app they were using, no cloud round-trip.
Why would anyone bother?
Choose Yap for its lightweight, fast, and private dictation experience, especially if you want to avoid the bloat and latency of other voice dictation tools that rely on heavy models or cloud services.
What is missing today?
It hits a real pain but solves it imperfectly, and the privacy promise has a hole: on Intel Macs it falls back to Apple's servers, which the README's 'nothing leaves your machine' framing does not make obvious. There is also no way to dictate in a language other than the system locale — the single most-requested thing in the tracker — which quietly excludes every bilingual user.
What makes this one different?
The wedge is not transcription, it is the bilingual person Apple ignores: one hotkey per language, mixed Chinese-English in a single sentence without switching anything, and cleanup that knows which words are proper nouns in the other language. Ship on-device only and say so honestly — no silent server fallback. Cold start (Chinese): 小红书 and V2EX threads on 「语音输入法 中英混说」, where bilingual users complain that every dictation tool forces one language; Cold start (English): the Yap tracker itself (issues #2 and #15 are the request), plus r/macapps.
What is the sharpest thing about it?
Zero-setup on-device dictation: no Whisper download, no API key, no cloud round-trip — the anti-bloat answer to Wispr Flow.
What could kill it?
This is a funded startup's side repo (Frigade), so the opening is theirs to close, and macOS ships dictation for free one settings toggle away. A thin wrapper over an Apple API is a weekend to clone and has no asset that compounds.
Who pays, and for what?
Honest read: thin on its own. The paid lane (Wispr Flow, ~$12/mo) is LLM cleanup, custom vocabulary, and cross-device sync; a free wrapper of Apple's free API can seed that funnel, but the wedge must be the polish layer, not the transcription.
What is the main risk?
Apple's built-in dictation is one settings toggle away and improves every OS release — and Yap is a side project of a funded startup (Frigade), so the repo is marketing, not an opening someone else can own.
What did real people actually say?
Verbatim from the public sources linked above, not paraphrased: “First off, thanks for Yap — the on-device approach with no model download is exactly what I was looking for.” — github-issue “I'd love to be able to dictate in 2 langu[ages]” — github-issue “Yep sends all audio data to Apple servers (on Intel Macs)” — github-issue
How much work is it?
about two weeks of work
Where does the evidence come from?
https://github.com/FrigadeHQ/yap https://news.ycombinator.com/item?id=49073834