I built Mudlark because I wanted an app for dictating prompts to LLMs and AI agents, and none of the existing apps worked quite the way I wanted. I like to see what's being dictated as I speak — it helps me know whether I was understood, and catch anything I need to go back and correct.
One thing that's always annoyed me about dictation software is the delay of your voice being sent off to a server. On top of that, using on-device AI to clean up what you said — with context from the app you're working in — makes a real difference for technical jargon, file names, and so on. But the last thing you want is screenshots being sent to third-party servers, particularly when you're working on company code.
When the Apple Speech and Foundation Model frameworks were released in mid-2025, the prospect of a fully local and private dictation app became very real.
The first time you launch Mudlark, if you don't have the speech models downloaded, the app prompts you to download them from Apple. After that, it never touches the network. There's simply no code in it to ship data elsewhere.
Mudlark is always going to have one focus: short-form dictation. You start a recording, talk, and paste into the active app. That's it — by design. If that doesn't suit you, there are plenty of other great tools out there.
I use Mudlark hundreds of times a day — it's become essential to how I work. I'm going to keep improving the app and fixing bugs, but that core experience isn't going anywhere. It's something I've come to depend on, and if you use it, you'll know what I mean.
I've put real effort into making the app fast and responsive, while keeping it as simple as possible.
Download a trial via TestFlight first to see if it works for you. If it does, it's a one-time purchase on the Mac App Store — for the app and all future updates.
Trial
Try Mudlark free via TestFlight. It's the full app with no limitations. No auto-charges or commitment.
Real-time transcription
Press a shortcut, start talking. Mudlark transcribes your speech as you go, showing live results in a floating panel. When you stop, the refined text is pasted into your active app.
On-device AI refinement
On-device AI refines your transcription before it's pasted. Minimal mode fixes punctuation and stutters. Balanced removes filler words and tightens sentences. Rewrite restructures your speech into clear, structured text. You can also provide your own refinement prompt.
Context-aware formatting
Mudlark can screenshot the active app when you start recording to understand what you're working on. It uses that context to format technical terms correctly, match the tone of your document, and handle domain-specific vocabulary.
Smart vocabulary
Add your own terms and Mudlark will recognise them when you speak. It also tracks your corrections over time and suggests words you frequently fix.