A free, open-source macOS voice dictation app that transcribes on-device with Apple's Speech framework — press a shortcut, talk, and your words land in whatever field you were typing in, with no cloud, no account, and no model to download.
Last verified · 2026-07-28 · by Moe Ameen
Yap is a free, open-source voice dictation app for macOS, built by the team at Frigade (FrigadeHQ) and surfaced in a "Show HN" post in late July 2026 under the line "OSS on-device voice dictation for macOS with no model to download." The workflow is the familiar dictation one: press a global shortcut (default `⌘⇧D`), speak, press it again, and Yap types the transcript into whatever text field you had focused — a document, a chat box, a code comment, a search bar. Nothing leaves your Mac. There is no account, no API key, and no cloud call anywhere in the app.
What sets Yap apart from most dictation tools is that it carries no speech model of its own. Instead of bundling Whisper or downloading a multi-gigabyte model on first launch, it calls Apple's native `SpeechAnalyzer` and `SpeechTranscriber` APIs introduced in macOS 26, with the older `SFSpeechRecognizer` as a fallback, so the transcription models it uses are the ones the operating system already ships. Audio is captured with `AVAudioEngine`; text is inserted by placing the transcript on the clipboard, issuing a `⌘V`, then restoring your previous clipboard contents. The result is unusually small and light: the app is around 4 MB, roughly 3,000 lines of native Swift with no Electron or browser engine, and it idles near 60 MB of memory while never touching the network. It's MIT-licensed and installs via a Homebrew cask (`brew install --cask frigadehq/tap/yap`) or a direct `.dmg` download.
Feature-wise it stays deliberately focused: a rebindable global shortcut, a live waveform while you speak, a local searchable transcript history (stored with SwiftData) you can copy or delete from, single-modifier trigger support, microphone auto-detection, and an optional launch-at-login toggle. The author reports Apple's on-device model landing slightly ahead of Whisper Small on accuracy and roughly three times faster, though that's a maker's claim rather than an independent benchmark.
Two honest caveats. First, because it leans on macOS 26 (Tahoe) speech APIs, Yap requires macOS 26 or later — it won't run on older systems. Language currently follows your system locale, with a language picker listed as the most likely next addition, so multi-language dictation on demand isn't there yet. Second, "Yap" is a crowded name: several unrelated open-source projects share it (including local-Whisper dictation tools and a Speech.framework CLI). This page is about Frigade's menu-bar dictation app specifically — the one whose whole pitch is that there's no model to download.
Dictation is the fastest way to get a thought out of your head, but Yap stops the instant your words hit the cursor — it hands you clean text and nothing more. It doesn't turn that text into a video, a carousel, or a scheduled week of posts. That's the exact seam Kompozy fills, and it makes Yap a natural front door for a voice-first content workflow.
The concrete play is talking your ideas into being. Open any text field, hit Yap's shortcut, and ramble a rough script, a hook, or a full content angle the way you'd say it to a friend — then drop that transcript into Kompozy as a source. From that one spoken draft Kompozy generates formats you could never dictate: Persona Shorts and HeyGen avatar video that read your words aloud in a face-locked recurring identity, brand-exact Carousel Posts via HyperFrames, Photo Posts, Quote Graphics, a full Blog Article, an Email Newsletter, and native Text Posts — all held to your voice by a Persona Brief and banned-word filters. Because you spoke the draft in your own cadence, the Persona Brief has authentic phrasing to anchor to, so the generated set sounds like you rather than like a template. Then Autopilot and a per-post review pipeline schedule and publish the batch across eight social platforms plus blog and email, reframed to 9:16, 1:1, and 16:9. Yap is how the idea gets out of your mouth; Kompozy is how that spoken idea becomes a published, on-brand content operation.
Yap is a free, open-source (MIT) voice dictation app for macOS from Frigade (FrigadeHQ). You press a global shortcut, speak, and it types the transcript into whatever field you were using. It runs entirely on-device with no account, no API key, and no cloud call, and it was shared as a Show HN post in late July 2026.
Instead of bundling its own speech model, Yap uses Apple's native SpeechAnalyzer and SpeechTranscriber APIs introduced in macOS 26 (with SFSpeechRecognizer as a fallback), so it transcribes with the models the operating system already ships. That keeps the app tiny — about 4 MB — and means nothing has to download or run in the cloud.
Yap requires macOS 26 (Tahoe) or later because it depends on the newer Speech APIs, so it won't run on older systems. Dictation language currently follows your system locale, with a language picker noted as a likely next addition. It also shares its name with several unrelated open-source projects, so make sure you're looking at Frigade's menu-bar app.
Use Yap to speak a rough script, hook, or angle into any text field, then paste that transcript into Kompozy as a source. Kompozy generates avatar video, clips, carousels, images, a blog, and a newsletter from that one spoken draft, holds your voice with a Persona Brief, and schedules and publishes across eight social platforms plus blog and email.