Jul 29, 2026 · by Christian Mathiesen · View source

Yap

Open-source voice dictation for Mac, fully on-device

Editorial analysis

The real bottleneck in your content pipeline is text entry, not creativity

Every social media operator I know has the same hidden tax: the hour spent re-typing the same caption into five platforms, turning a voice note into a tweet thread, or fixing a DM that autocorrected into nonsense. The tools that promised to solve this with voice dictation have usually been one of three things: a subscription, a privacy leak, or an Electron app that eats hundreds of megabytes to do what a hotkey should. That’s why Yap stopped me. It’s a free, MIT-licensed, fully on-device voice dictation app for Mac built by Frigade. Set a hotkey, talk, press the hotkey again, and your words land in whatever field you were typing in. No model download. No network calls. No server-side transcription. In a creator economy where most tools are racing to monetize your attention and your drafts, a tool that treats your words as yours is almost radical. And if Yap works the way its maker describes, it says something bigger about where the creator tool stack is heading: the next wave of tools won’t win on AI model size; they’ll win on how quietly they fit into the workflow you already use.

The dictation problem is not accuracy — it’s trust and bloat

I’ve tested enough dictation tools to know the pitch by heart: “Speak 3x faster than you type.” The accuracy is usually fine for short bursts. The problem is everything surrounding the speech-to-text engine. Cloud transcription services like Otter.ai are built for meetings, not for the fast, fragmentary way creators actually talk when they’re brainstorming hooks. Desktop assistants like Wispr Flow are better at the “dictate anywhere” promise, but they bring baggage. In his launch comment, Frigade’s CTO Christian Mathiesen says the team tried existing dictation tools and kept hitting the same walls: subscriptions, giant local models, and apps that “shipped a whole browser engine to run a menu bar icon.” He also raised the issue that most tools send transcripts to remote APIs — and claims that even Apple’s built-in dictation does this.

That last point matters more than most creators realize. When you dictate a caption into a notes app, it might be a half-formed idea about a client campaign, an unreleased product, or a personal rant you never plan to publish. You don’t want that on a server somewhere as a training artifact. Privacy in social media isn’t just about protecting your own drafts; it’s about protecting the brands and people who trust you with their voice. In my own tests of similar tools, the first question I ask is not “How accurate is it?” but “Where does this audio go before it becomes text?” Yap’s answer — nowhere — is the cleanest I’ve seen in this category.

Why TikTok creators should care more than LinkedIn ones

LinkedIn rewards structure. A thoughtful post needs a clear argument, a narrative arc, and a reasonable amount of editing before it goes out. Voice dictation is a terrible final editor for that kind of writing. TikTok, Instagram, and YouTube Shorts reward conversational velocity. A hook is a spoken sentence before it’s a written one, and the algorithm doesn’t care about your grammar; it cares about whether viewers stay past the first two seconds. That means TikTok creators need to generate and test 10 hooks before filming, and dictating those hooks into a notes app is the fastest way to feel their rhythm. If you’ve ever tried to write short-form captions with a keyboard, you know the struggle: typed language sounds stiff, over-corrected, like a press release. Spoken language sounds like a person. A hotkey dictation app gets you from the raw idea to the text field without the translation layer that makes captions sound robotic.

What Yap actually gets right

The first thing Yap gets right is respecting the clipboard. That sounds like a small detail, but if you’ve ever had a dictation app paste a sentence over the URL you just copied, you know it’s a dealbreaker. On the launch thread, a commenter named Asad M. asked exactly that: if Yap puts the transcript on the clipboard to insert it, it’s stomping whatever was copied, and that’s how a dictation app gets deleted on day two. The maker’s answer was that they don’t mess with the clipboard. In my book, that’s a trust signal worth more than a feature list.

Second, Yap is native. A commenter named Rabnoor Singh gave the clearest explanation of why that matters: most apps in this category ship a whole browser engine — an Electron runtime — just to draw a menu bar icon. That’s where the hundreds of megabytes go, and it’s why those apps feel heavy from day one and only get worse. Yap is around 3,000 lines of native Swift, delivered as a 4 MB app that idles in the tens of megabytes. It calls an API that’s already part of the operating system, so there’s no runtime to maintain and no language model to download. For a creator who runs a half-dozen apps at once while editing video, the difference between a 40 MB utility and a 400 MB Electron app is the difference between being able to breathe and watching your fans spin up.

Third, it’s simple by design. The hotkey pattern — press, talk, press, done — is exactly how dictation should work. You don’t open a window, select a microphone, click a button, wait for a waveform, and then copy the result. You press a key, speak, and the text is already in your caption field. When I’m batch-producing content, I don’t want a tool that gives me a dashboard. I want a tool that stays out of my way and lets me get the words out while they’re still connected to the idea.

Finally, it’s open source. The fact that Frigade open-sourced it under the MIT license is a statement in a market where every other dictation app is racing to lock you into a subscription. The company says it built Yap because they wanted it for themselves. That is the exact opposite of the venture-backed “magnify creator productivity” playbook, and for social media managers who have been burned by tools that raise prices or pivot after a funding round, it’s a refreshing change.

What creators and social teams can borrow from this

Even if you never install Yap, the thinking behind it is worth stealing for your own workflows.

Use dictation as a capture layer, not a publishing layer. The best way to use a tool like Yap is to speak a draft, then edit it on a second pass. That’s the same repurposing workflow that works for video: shoot messy, edit tight, distribute everywhere. The tool doesn’t need to format your text, fix your grammar, or make it sound polished on the first try. It just needs to get the raw material out of your head before you lose it. In my experience, creators who treat dictation as a final editorial tool are always disappointed. Creators who treat it as a draft generator are always faster.

Audit your tools the way you audit your content. Rabnoor Singh’s comment about browser engines is a perfect heuristic for any app in your creator stack: Is it native or Electron? Does it download a model on first run? Does it have an offline mode? Those three answers predict almost everything about how heavy a tool will feel six months in. I’d extend that to scheduling and analytics tools too. If a Buffer-style scheduler is making your browser crawl, it’s not because scheduling is inherently heavy; it’s because the tool’s architecture is doing more work than it needs to.

Look for native OS capabilities before adding another SaaS. The deeper lesson of Yap is that Apple has quietly made on-device speech recognition a commodity. With macOS 26, Apple introduced an on-device speech API called SpeechAnalyzer, and Frigade’s team claims it matches frontier API models in benchmarks. Whether or not that benchmark claim survives independent scrutiny, the direction is obvious: the platform is absorbing the hard AI problem, so third-party tools have to compete on workflow, not raw transcription. The same thing is happening with CapCut and AI video editing, with Canva and AI design. The winning tools are not the ones with the best model; they’re the ones that wrap the model in the right human workflow. Yap understands that. Most dictation apps don’t.

Where the math breaks

The source itself is a little inconsistent on memory. The product description says Yap idles near 60 MB of memory. The maker’s launch comment says the whole thing uses just a few MB. A reviewer who tested it measured 39.7 MB. My take: none of those numbers are bad, but the discrepancy tells you this is a young project and the team hasn’t standardized its performance measurements. It also tells you not to trust the first number on a Product Hunt page. Test the tool yourself, under load, with your normal number of open browser tabs and editing apps, and see if the hotkey still fires instantly. For a utility that sits in your menu bar, responsiveness under memory pressure is the only performance metric that actually matters.

Where I’d pump the brakes

I like what Yap is doing, but I’m not ready to crown it the default dictation tool for every social media operator — and you shouldn’t either.

Language support is Apple’s support. A reviewer asked what languages Yap supports, and the maker said 30+. But another commenter gave the more useful answer: because Yap runs on Apple’s on-device speech APIs, the languages it supports and how well each one works are determined by Apple, not by Yap. That cuts both ways. It means no model download and no data leaving your machine, but it also means if Apple’s on-device dictation is weak for your language, no amount of work on Yap’s side will fix it. Before you build a workflow around this, the practical advice is to check System Settings > Keyboard > Dictation and see whether your language is available for on-device dictation. If it only works via the server option, expect weaker results and more latency. That’s a limitation you have to live with, not a bug Yap can patch.

Secure fields remain an open question. One commenter asked a genuinely sharp security question: macOS deliberately blocks accessibility APIs from reading or writing into secure text fields — password fields, password manager vaults, sudo prompts in the terminal. If Yap “works anywhere you can type,” how does it behave in those fields? Does it detect that it’s in a secure field and decline to paste, or does it just fail silently? The visible launch thread doesn’t contain a clear answer from the maker. So I’d assume “anywhere you can type” means regular text fields, not every protected input on your Mac. If your workflow involves dictating into a password manager, test it before you trust it.

Grammar cleanup is a moving target. An early reviewer said Yap was quick and accurate, but didn’t clean up verbal stumbles and grammar issues, which made it ideal for short messages rather than polished copy. The maker responded that a fix was coming, and later said it was fixed in the latest version. Good. But “fixed” is a strong claim, and dictation cleanup either works with your particular speech patterns or it doesn’t. There is no universal grammar cleaner that works for everyone. If you plan to dictate client-facing emails or long LinkedIn posts, spend 15 minutes testing the cleaned output before you rely on it.

Who this is not for: if you need meeting summaries, searchable transcripts, speaker diarization, or action items, this is not your product. That’s what tools like Otter.ai are for. If you need team collaboration, shared libraries, or approval workflows, a hotkey dictation app is the wrong layer. And if you’re not on a Mac, Yap is simply not available to you. The source does not disclose any Windows version, and given that it’s built on macOS 26’s speech APIs, I wouldn’t expect one soon.

What I’d watch / test next

This week, if you’re on a Mac, the test is simple. Install Yap from the download page, set a hotkey, copy a URL to your clipboard, then dictate a real caption draft and see whether the URL survives. Then dictate the same caption into Buffer, Later, or wherever you do your scheduling, and see where the workflow breaks. If it holds up, add it to your capture stack: voice notes in the morning, dictation into drafts, editing before scheduling. Also watch the open-source forks. Because Yap is MIT-licensed, someone will build a cleanup pass on top of it, and that combination — on-device dictation plus local LLM-assisted editing — is the true threat to subscription dictation apps. My longer-term bet is that Apple’s on-device speech API will commoditize basic transcription within a year, and the tools that win your attention will be the ones that connect a hotkey to your calendar, your CMS, and your team’s approval flow. Yap is a small step in that direction. That’s why it matters.

Ready to Create Your Own?

Join thousands of brands creating high-performing video ads with FLOWNIB. No editing skills required.

Start Creating for Free