Voice to clean text for Mac, still in your own words.
Whisperly is a Mac menu-bar app for dictating text you can send as-is.
Press a hotkey anywhere, talk, and the result lands at your cursor in whatever app you were already in. **Clean-up mode** drops the fillers, false starts and repetitions and fixes the punctuation, but keeps your wording, so it still reads like you wrote it and not like something rewrote you.
I built it because I think in Russian and work in English, and raw dictation was never usable: either a wall of *ums*, or a polished paragraph that no longer sounded like me.
Signup is open and there is no waitlist. We just haven't announced anywhere yet.
112 people supporting this launch
4.8
6 ratings
AI video with native sound, paid per clip. Nothing stored.
A calmer Mac menu bar, one click away.
Enterprise Software. Intelligent Automation. Global Impact.
Comments
79Hi everyone. It's always easier to formulate a thought in your native language, and when you work in another one that slows you down every day, so I built Whisperly. You hold a hotkey, say it however it comes out, and the text lands where your cursor already is: a straight transcript, or run through cleanup that drops the ums without rewriting how you sound, or translated into the language you need to send. Wiring up the speech model took an afternoon. The rest of it is where the time went: the cleanup, the hotkey handling, custom vocabulary for the names every recognizer mangles, and the paste step, which I didn't expect to be the hard part. macOS gives you no reliable way to confirm that an arbitrary app accepted inserted text, so when Whisperly can't confirm the text landed, it puts the result on your clipboard and tells you so. It's a menu bar app, direct download, no Mac App Store. Free covers 2,000 words a week, and new accounts get 14 days of Pro with no card, so translation and the tones are open while you're trying it. The thing I'd like feedback on is the tone setting. In cleanup and translation you pick one of four: Original, Casual, Neutral, Formal. Original is the default and leaves your wording alone, the other three make the text more casual, neutral or formal. I still don't know whether people want that choice or would rather have one mode that just works. If you try it, tell me which one you stayed in.
Great idea, fully support it! Tone choices are useful, I think
The 'keeps your wording' constraint is what makes this interesting — every other dictation cleanup I've tried quietly paraphrases you into corporate-speak. The bilingual origin story explains it: when you think in one language and work in another, a tool that 'improves' your phrasing is actively unhelpful. Is the cleanup pass running locally on the Mac, or going to an API? Menu-bar dictation tools live or die on latency.
Cleanup goes to an API. Recording and the voice detection are local, then the audio goes to our EU server and out from there to model providers: OpenAI's gpt-4o-mini-transcribe for speech to text, Fireworks GPT-OSS 120B for the cleanup pass, both with fallback chains behind them. We don't self-host either leg. Audio isn't stored on our side, it's dropped the moment the text exists. Latency, with the one number I actually measure: the cleanup pass is p50 1.4s, p95 4.3s over the last month (n=503 clean dictations, small user base, so treat it as directional). That's LLM wall-clock on our side, not button-to-paste. Add about a second of transcription plus two round-trips and you land near three seconds on a short take, though that's arithmetic, not a measured end-to-end number. Server sits in Frankfurt, so add a bit if you're far from it. Past 30 seconds the app cuts the take at pauses and uploads chunks while you keep talking, so the earlier chunks are usually transcribed before you release the key. There's also a Raw mode that skips the LLM pass entirely. The prompt takes out fillers, false starts and repetitions, fixes grammar and punctuation, and formats a bit: bullets when you dictated a list, letter layout when it sounds like an email. It's also allowed light rephrasing when the spoken structure reads badly in writing, and that permission is where it can overreach. Default tone is Original and it still tidies more than it should sometimes. When it does, I fix the prompt.
Co-founder of imgtovideoai.net. Math, sports, and building products.
Huge congrats on launching Whisperly – what a brilliant concept! I'm particularly impressed by your 'Clean-up mode' that masterfully retains one's unique voice and wording while cleverly stripping out all the noise and refining punctuation. That's truly a game-changer! So many dictation apps force a choice between raw, unusable transcriptions or overly polished prose that just doesn't sound like you. Your solution beautifully bridges that gap, especially for those of us who might think in one language and write in another. I'm genuinely curious, what kind of feedback have you been getting from early users about how perfectly it captures their authentic 'voice'? Also, any thoughts on offering even more nuanced control over the clean-up process in future iterations?
Thank you very much for the kind words and the question. Yes, we’re actually thinking about giving users more control over the final text. Some people prefer exclamation points instead of periods, some always want to add emojis at the end of sentences, and some like using short dashes instead of M-dashes. So we’re considering making the settings more flexible and customizable for each user.
I build & lead the engineering behind AI agents · Co-Founder & VP R&D @ Dial
Transcription tools usually force you into their ecosystem - you dictate, the audio lives in their app, and you export later. Whisperly flips that by keeping your words natural while capturing them directly as text on your Mac. The "still in your own words" angle is key - no weird AI paraphrasing, just clean capture of what you actually said. For writers, researchers, or anyone doing voice notes throughout the day, this removes the friction of switching between voice recording app and text editor. The Mac-first approach means tight OS integration without the bloat of cross-platform solutions.
Yep, thats totally right! Thanks!
nice to have product may consider it in our future projects
Thanks! Hope you'll like it!
The clean-up mode that drops fillers but keeps your actual wording is the part most dictation tools get wrong. Does the Mac app work offline, or does audio get sent to the cloud for processing?
Cleanup goes to an API. Recording and the voice detection are local, then the audio goes to our EU server and out from there to model providers: OpenAI's gpt-4o-mini-transcribe for speech to text, Fireworks GPT-OSS 120B for the cleanup pass, both with fallback chains behind them. We don't self-host either leg. Audio isn't stored on our side, it's dropped the moment the text exists.
looking great and i think its way better than whisprflow helps dev really very much its very crucial part of workflow tbh
Thanks! Hope you'll like it!
AI Builder.
The custom-vocabulary support across scripts is a thoughtful detail; names and product terms are where dictation usually breaks. In Translate mode, can one entry define both the spoken source form and the preferred spelling in the target language?
Actually, those are two separate issues. Your vocabulary is always exactly how you define it. So if a word is added in Latin script, the Translate mode will always insert it in Latin script. If you entered it, say, in Chinese characters, it will always be inserted in Chinese characters.
Great work! Hard to find these type of products now
Thank you very much, that's really nice of you!
https://www.partyvillas.in/
Great, Hard to find these type of products now
Thanks! Hope you'll like it!
QA
Awesome app, it's a real time-saver! Personally, I feel like tone choices are useful, so don't remove them.
Thanks! Sure, we definitely will keep them :)
Building LancePay, a simple invoicing and payment collection tool for freelancers.
Clean positioning — turning voice into polished text while preserving the writer's voice is a useful Mac workflow. A before/after example on the page would make the value even clearer.
Thank you very much, that's really nice of you!
Ops for Namefi!
Such a great product!
Thanks!
Transcriptions is literally how I am constantly interacting with my workflows. I literally have just stopped typing and switched fully to writing emails, prompting AI, and everything with my voice. So, this tool literally helps a lot.
Thanks! I’m almost forgot what is keyboard :)
Managing Director of Mini Power
Such a great product!
Thanks!
I like how easy this tool looks.
thanks!
Hello Guys my name is usama and i am saas developer
one of the graete product i was work with video to audio to text but i face issues on voice to text which cover by this
Thanks, means a lot!
Such a great product!
Thanks!
Full Stack Developer
such a good product.
Thanks!
Such a great product!
Thanks!