Frigade's new menu-bar app does system-wide voice-to-text on-device with Apple's macOS 26 Speech framework — so there's no cloud, no account, and nothing multi-gigabyte to download.
2026-07-28 · by Moe Ameen
Frigade (FrigadeHQ) has released Yap, a free, open-source voice dictation app for macOS, 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." It's a small menu-bar app that does system-wide speech-to-text: press a global shortcut (default ⌘⇧D), speak, press again, and Yap types the transcript into whatever field you had focused. Everything runs locally — no account, no API key, and no audio sent to a server.
The distinguishing design choice is that Yap ships no speech model of its own. Rather than bundling Whisper or downloading a multi-gigabyte model, it calls Apple's native SpeechAnalyzer and SpeechTranscriber APIs introduced in macOS 26 (with the older SFSpeechRecognizer as a fallback), so it transcribes with the models the operating system already carries. That keeps it unusually small: about 4 MB, roughly 3,000 lines of native Swift with no Electron or browser engine, idling near 60 MB of memory and never touching the network. It's MIT-licensed and installs via a Homebrew cask or a direct .dmg. The maker frames Yap as a reaction to the state of dictation tools — paid apps, bundled memory-heavy models, Electron wrappers, and cloud dependencies — and reports Apple's on-device model landing slightly ahead of Whisper Small on accuracy and roughly three times faster, though that's a developer claim, not an independent benchmark.
There are honest limits. Because it leans on macOS 26 Speech APIs, Yap requires macOS 26 (Tahoe) or later and runs on Mac only, so Windows, Linux, and older-Mac users can't use it. Dictation language currently follows the system locale, with a language picker noted as a likely next addition. And "Yap" is a crowded name — several unrelated open-source projects share it, so this launch specifically refers to Frigade's menu-bar dictation app.
The reason a launch like this is worth a creator's attention isn't the dictation itself — it's that frictionless voice capture is the missing front end of a content pipeline. Yap makes the "get the idea out of your head" step effortless and private. The catch is that it ends at a transcript. The value is in what happens next, and that's where Kompozy turns a spoken thought into a week of published content.
Here's the pipeline you can build today. Away from your desk, on a walk or between meetings, hit Yap's shortcut and talk out three or four content angles as they come to you — no typing, no cloud. Back at your machine, paste those transcripts into Kompozy as sources. From each spoken draft Kompozy generates the finished set — Persona Shorts and HeyGen avatar video that speak your words in a face-locked recurring identity, Clipped Shorts, brand-exact Carousels via HyperFrames, Photo Posts, Quote Graphics, a Blog Article, an Email Newsletter, and native Text Posts — all held to your cadence by a Persona Brief, since you literally spoke the draft. 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 made capture free; Kompozy makes everything after the transcript — the generation and the publishing — the part you never have to hand-assemble.
Yap is a free, open-source (MIT) voice dictation app for macOS from Frigade. You press a global shortcut, speak, and it types the transcript into any field. It costs nothing — there's no account, API key, or subscription — and it installs via a Homebrew cask or a direct .dmg download.
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 OS already ships. That's why the app is about 4 MB and runs entirely on-device with no cloud call.
Yap requires macOS 26 (Tahoe) or later because it depends on the newer Speech APIs, and it runs on Mac only — there's no Windows or Linux version. Dictation language currently follows your system locale, with an in-app language picker noted as a likely future addition.
Not on its own — Yap only transcribes speech into text at your cursor. To turn a dictated draft into video, carousels, images, a blog, and posts and publish them across platforms, pair Yap's capture with a content engine like Kompozy, which generates and schedules the finished content from your transcript.