say it once.mochi writes it down.
Press one key and talk. Before you have finished the thought it is a note in your Obsidian vault, in your own words.
macOS 14.4 or later · Apple Silicon
How it works
01
press ⌥space
The pet wakes and starts listening. It reacts first, before anything touches the disk, so you know it heard you.
02
say it
Talk for as long as you like. Recognition is running on your Mac while you speak, so nothing is queued and nothing is uploaded.
03
it is in your vault
A markdown note appears in Obsidian: what you said, then what it means. Go back to what you were doing.
why you would want it
six reasons, in the order people ask them
01
it is written down before you lose it
One key, anywhere, in any app. No window to find, no button to hunt for, no file to name. The gap between having a thought and recording it is where most thoughts go, and this closes it.
02
your exact words are kept
The transcript goes in first, word for word, including the parts where you changed your mind. The tidied version is added underneath it. You can always see what you actually said, which is the only thing that makes the tidy version worth reading.
03
nothing you say leaves the machine
Recognition runs on your Mac. Not on a server with a privacy policy, on the computer in front of you. That is what makes it free to use as much as you like, and why there is no recording of your day sitting in someone else's account.
04
they are just files
Plain markdown, in a folder you picked, with names you chose. Readable by anything, searchable by everything, and yours whether or not Mochi is still here in five years.
05
no robot joins your call
Nobody watches a notetaker appear in the participant list and wonder what it is. Mochi records what your Mac is already playing, separates the voices on your Mac, and hands you one note at the end.
06
you will actually keep using it
A small drawn version of you sits at the edge of the screen and wakes up when you talk to it. That is not a feature. It is the reason this becomes a habit instead of another tool you installed once.
keep scrolling to move along the rowswipe to move along the row
pick how well it should hear you
All three run entirely on your Mac and none of them cost anything to use. The only trade is disk space against how well it copes with an accent, a noisy room or a long meeting.
Between 227 MB and 483 MB on disk, downloaded once. Change tier whenever you like; nothing you have already written is affected.
small
227 MB · english
Quick voice notes, older Macs, the fastest first run.
standard
336 MB · 25 languages
Most people, most meetings, more than one language.
recommended for meetings
full
483 MB · 25 languages
Strong accents, noisy rooms, long meetings.
Meeting mode needs one more download, 14 MB, once. You can swap tiers at any time in settings, and your old notes stay exactly as they are.
meeting mode
one note per call
Start it before a call. Mochi records what your Mac is playing and what your microphone hears as two separate tracks, tells up to four voices apart without sending any of it anywhere, and writes a single note when you stop.
It never starts itself and it never runs because something looked like a meeting. The recording indicator on the pet and in the menu bar cannot be switched off while it is on.
Recording a call records the other people on it, and in some places every one of them has to agree first. Mochi gives you a line to paste into the chat, and shows you a reminder the first time.
The line Mochi copies for you:
Heads up, I'm recording and transcribing this call for my own notes.
obsidian
it writes to your vault
No plugin, no API, no sync service in between. Mochi puts markdown files in a folder you pick, with the filename pattern, the frontmatter and the daily-note behaviour set the way you want them.
The verbatim transcript is written before any model is called. If your network is down, your key is wrong, or the provider is having a bad day, you still have exactly what you said. The tidied version is appended when it arrives, so a failure there costs you the summary and never the words.
---
created: 2026-07-28T09:14:03+03:00
source: mochi
duration: 00:00:42
---
## Transcript
so the migration is blocked on the schema change, I need to talk to
Priya about it before Thursday, and we should probably hold the
release until it lands
## Note
Migration is blocked on the schema change.
- Talk to Priya before Thursday
- Hold the release until the schema change landsyour own key
you pick the model
The tidying step uses a model you choose, with a key you paste in once. It is kept in your Keychain, it goes straight from your Mac to that provider, and it never passes through anything of ours.
That is also why there is one price instead of three: we are not reselling anyone's inference, so we have no reason to meter yours. Skip it entirely if you like and Mochi still transcribes, it just will not tidy up after you.
- OpenRouter
- Anthropic
- Google Gemini
- DeepSeek
- Moonshot Kimi
- opencode Zen
- opencode Go
it looks like you
Upload one photo and Mochi draws four pets from it. Pick the one you like, re-roll if none of them are right, give it a name.
The photo is used once and deleted straight after. No face recognition, no template, nothing kept to compare against later.
Would rather not send a photo at all? Pick a ready-made pet instead. Everything else works exactly the same.
three exceptions
Your audio, your transcripts and your notes never reach a Mochi server. Three things do, and all three are named here rather than left in a policy for you to find.
- your account
- An identity record and a subscription record, so the app knows which pet is yours and whether the licence is active. It holds no note content, no audio and no transcripts.
- the photo, once
- Sent once to draw your pet, then deleted. No biometric identifier or template is created, compared or stored, and no facial recognition is performed at any point.
- your email, if you asked to be told
- An address on the waitlist, and the date you left it. Used once, to tell you the app is out, and deleted after that. Reply to any message from us to be taken off it sooner.
one plan
$12 a month, or less on the annual. 7 days free first, and one click to cancel with no retention screen in the way.
mochi
$12a month
Or $96 a year, which is $8 a month.
The plan, when it ships: 7 days free, cancel any time.
Everything is included:
- Unlimited on-device transcription
- All three model tiers, swappable any time
- Meeting mode with speaker separation
- Writing to your Obsidian vault
- Screen awareness, on request only
- Every model provider, using your own key
- One pet, plus an allowance of re-rolls
- Every update while you are subscribed
A card is required to start the trial. You are not charged if you cancel inside 7 days. After that, 14 days to change your mind, anywhere in the world, no questions.
- Do I pay for transcription?
- No. Speech recognition runs on your Mac, so it costs nothing per minute and there is no allowance to run down.
- Do I pay for the tidied notes?
- You pay your model provider directly, with your own key. Mochi never sees that traffic and takes no margin on it.
- What does the trial cost?
- Nothing. A card is required to start it, and you are not charged if you cancel inside 7 days.
- How many Macs?
- Your licence activates on several Macs. If you reach the limit, free one up in the billing portal.
- How do I cancel?
- One click, straight to the billing portal. No retention screen, no counter-offer, no survey between you and the button.
- What if I want a refund?
- 14 days, anywhere in the world, no questions asked. Payments go through a merchant of record, so refunds and invoices come from them.
you already say the good part out loud.
this is just the bit where it gets written down.
Signed and notarised DMG · Apple Silicon