Does my audio leave my Mac?
Not for transcription: every speech model runs locally. It only leaves if you deliberately point the AI step at a cloud provider such as OpenAI or Anthropic, and the app says so per provider before you pick one.
Our own product
Your meetings, transcribed and summarised on your own Mac
YWhisper records both sides of a call, turns it into text, and writes the summary. Speech never leaves your Mac unless you decide it should. Dictate into any app with one shortcut, and transcribe the recordings you already have.
Private beta
Apple Silicon, macOS 26.4+
Why we built it
We are on calls all day, and the notes were always the part that slipped. The tools that promise to fix that want your meetings on their servers, and a recording of a client call is not ours to hand over. So we built the version we were willing to use ourselves.
The notes never get written
Everyone means to write the minutes up afterwards. An hour later the decisions are approximate, the owners are vague, and the numbers are gone.
The alternatives upload everything
Meeting transcription is almost always a cloud service. That means client conversations, salaries and roadmaps sitting on somebody else's infrastructure under somebody else's retention policy.
What YWhisper does
Three things, done properly, without sending your work to somebody else's server.
Capture
Both sides of the call
Your microphone and the audio coming out of your speakers, recorded together, so a Teams, Zoom, Meet or FaceTime call ends up as one recording.
Starts when the call does
Point it at the apps you take calls in and it notices them reaching for the microphone. Record automatically, or ask first, per app.
Dictate into any app
Press your shortcut anywhere, speak, press it again. The text lands where your cursor is, in Mail, Slack, Notion or your editor.
Turn it into something useful
Transcribed on your Mac
Six speech models, all running locally on the Neural Engine and the GPU. No connection needed to turn speech into text.
A summary you can act on
Decisions, owners and the numbers that were actually said, in the language the meeting was held in, and longer for a long meeting.
All of it searchable
Recordings, dictations and transcribed files in one Library, searchable by their contents, exportable to Word, PDF or Markdown.
See it work
Screenshots from the app as it is today, not a mock-up of where we are heading.
One button, and it is running
A big record button, a timer, and the level of whatever is actually being captured. Mute your own side for a moment without stopping the recording, switch microphone mid-call if you changed it in the call app, and see at a glance whether the other side is being recorded too. Close the window and the menu bar keeps the timer and the controls.
- Mute your own side without stopping the recording
- Switch microphone mid-call if you changed it in the call app
- Close the window and the menu bar keeps the timer and the controls
It notices the call started
Point YWhisper at the apps you take calls in, Teams, Zoom, Meet, WhatsApp, FaceTime or anything else, and it watches for them reaching for the microphone. Per app you decide: record automatically, or ask first. The recording stops by itself when the call ends, so there is nothing to remember at either end.
- Teams, Zoom, Meet, WhatsApp, FaceTime or anything else you add
- Record automatically, or ask first, per app
- Stops by itself when the call ends
Everything in one Library
Recordings, dictations, transcribed files and rewritten text all land in the same place, searchable by their contents. Play the audio back, read the transcript beside the summary, re-transcribe with a different engine, or export to Word, PDF or Markdown.
- Search across every transcript and summary
- Re-transcribe an old recording with a better engine
- Export to Word, PDF or Markdown
Teach it the words it keeps getting wrong
Speech models have no idea how your name, your products or your team's jargon are spelled, so they guess, and they guess the same way every time. List the words once and YWhisper writes them the way you want, in the transcript itself and in every summary. Export the list as CSV to share it with colleagues.
- Applied to the transcript itself, so it works even with no AI involved
- Whisper is told the words before it listens
- Import and export as CSV
Transcribe what you already have
Drop in an audio or video file you recorded elsewhere and get a transcript and a summary out of it. Old voice memos, a webinar recording, an interview, a lecture you were sent.
- Audio and video, whatever macOS can read
- Pick the engine and the language per file
- The result lands in the Library with everything else
What that adds up to
6
Speech models
Whisper, Parakeet and Canary, downloaded on demand and switchable per recording.
~100
Languages
With Whisper. Parakeet covers 25 European languages, Canary four.
0
Bytes uploaded
Speech to text is always local. Keep the AI local too, with Apple's built-in model, omlx, Ollama or LM Studio, and nothing you say or write leaves the machine at all.
Private because of where it runs, not because of a promise
Speech to text happens on your Mac. The audio, the transcript and your recordings stay in a folder you own. There is no account and nothing to upload, and the AI step only reaches the internet if you deliberately pick a provider that lives there.
Transcription is always local
Every speech model runs on your Mac, using the Neural Engine and the GPU. No internet connection is needed to turn speech into text.
AI is your choice
Polishing and summarising can run on your Mac too, through Apple's built-in model, omlx, Ollama or LM Studio. Prefer a cloud model? Point YWhisper at OpenAI, Anthropic or any OpenAI-compatible endpoint. The app tells you plainly, per provider, whether it runs here or somewhere else.
It can stay entirely on your Mac
Transcription never leaves the machine, whatever you do. Neither does the polish and summary step, as long as you point it at a model running on your Mac. A cloud provider only sees your text if you chose one yourself. Beyond that the only traffic is the licence check and the update feed.
You decide where the AI runs
Transcription is local, always. The polish and summary step is the one place a cloud model is even an option, so the app names the provider and says plainly whether it runs here or somewhere else before you pick it. Apple's built-in model, omlx, Ollama and LM Studio all keep it on the machine.
Pick the speech model you trust
Different work needs different engines. YWhisper ships six, downloads them on demand, and lets you switch per recording.
Whisper Large v3 Turbo
1.6 GB · 100 languagesThe most accurate engine here. Large v3 tuned for speed, so it is far quicker than its size suggests.
Whisper Small
487 MB · 100 languagesWhat a new install gets. Good accuracy for its size, and it covers almost every language people actually speak.
Parakeet TDT 0.6B v3
670 MB · 25 European languagesNVIDIA's transducer. Very fast on Apple Silicon and strong on European languages.
Parakeet TDT 0.6B v2
661 MB · EnglishEnglish only, and the best speed to accuracy trade-off if that is all you need.
Canary 1B Flash
940 MB · 4 languagesHandles accented speech better than its size suggests. English, German, Spanish and French.
Canary 180M Flash
213 MB · 4 languagesThe smallest engine here, and the quickest to download.
Whisper works the spoken language out from the audio by itself. You can also set the language yourself, which is more reliable on short or noisy recordings.
Switch engine whenever you like
Every model is a download away, and the app shows you the size and what each one is good at before you commit. Try a recording with one, re-run it with another, and keep whichever got it right.
What you need
- A Mac with Apple Silicon
- M1 or newer. The speech models use the Neural Engine and the GPU, which is what makes running them locally practical. Intel Macs are not supported.
- macOS 26.4 or later
- YWhisper uses the current system frameworks for audio capture and the interface.
- Disk space for a model
- Between 213 MB and 1.6 GB, depending on which engine you pick. Downloaded once, then it is yours.
- Two permissions
- Microphone, always. Screen Recording only if you want the other side of a call recorded, which is how macOS gates system audio.
Request beta access
YWhisper is in private beta. Tell us a little about how you would use it and we will send you a licence key and a download link.
Questions people ask
What people ask before they try it.
Does it work on an Intel Mac?
No. YWhisper is Apple Silicon only. The models lean on the Neural Engine and the GPU, and on Intel the result would be too slow to be useful.
Can it record both sides of a Teams or Zoom call?
Yes. YWhisper records your microphone and the system audio, which is everything coming out of your speakers, and mixes them into one recording. macOS asks for Screen Recording permission for that, because that is how it gates system audio.
Which languages can it transcribe?
With Whisper, around 100, including Dutch, Turkish, German, French, Spanish, Polish and Arabic. Parakeet covers 25 European languages, Canary four. The app itself is available in English, Dutch and Turkish.
Can I transcribe a recording I already have?
Yes. Drop an audio or video file into the Transcribe tab and you get a transcript and, if you want, a summary.
Do I need an internet connection?
Only to download a model the first time, to check your licence, and to fetch updates. Recording, transcribing and summarising with a local model all work offline.
What does it cost?
The beta is free while it lasts. Pricing for the release has not been announced yet, and beta users will hear first.
Want one of your own?
YWhisper is our own product, built by the same people who build software for our clients. If your team needs something like it, that is a conversation we are happy to have.
Start the conversation