No account, so no profile
Nothing to sign up for. There is no database of Briska users to breach, subpoena or sell, because we never built one.
macOS · Apple silicon
Stop typing prompts. Just say it. Anywhere on your Mac.
No window to switch to, no prompt to write. Hold a key, say what you mean, let go — Briska writes it at your cursor in any app, removing filler words, fixing self-corrections, and punctuating as it goes. On-device, start to finish. None of it uploaded.
An address, not an account. The download starts on the same click.
- Zero bytesNothing you say is uploaded
- Any appAnywhere you can type
- OfflineWorks with the Wi-Fi off
- No accountNothing to sign in to
0:08
Not a raw transcript
You ramble. Briska writes what you meant.
Traditional dictation types every false start, stutter and trailing thought. Briska listens to the whole context instead, stripping the noise, resolving mid-sentence corrections, and handing you finished prose.
What you spoke
um hi sam so i was thinking let's move the handover to wednesday no sorry thursday at four uh so the whole team can make it and yeah i'll send the notes across beforehand
What gets typed
Hi Sam, let's move the handover to Thursday at 4 so the whole team can make it. I'll send the notes across beforehand.
- Fillers and false starts cleanly removed
- ‘Wednesday, no sorry, Thursday’ resolved to one correct date
- Punctuation, capitalization, and numbers automatically applied
- It falls back rather than guesses. When the cleanup is ambiguous Briska keeps your exact words instead of inventing a tidier version of them.
- It types where your cursor already is. Mail, Slack, Notion, Google Docs, your terminal. Briska inserts at the OS level, so there is nothing to install per app. Without Accessibility granted it hands you the text on the clipboard instead, and says so.
Beyond dictation
Fix it. Don't retype it.
Dictation makes words from nothing. These three are for the words that already exist: a rough draft, a message someone sent you, an hour-long call.
One key, one rewrite, in place
Select text anywhere, press the key, and it is replaced where it sits. Three Skills ship with the app; write your own instruction once and it binds to the next free key.
- Your wording, not ours. A Skill is just an instruction you saved, so “make this concise” is as valid as “polish”.
- It refuses rather than truncates. Select more than the model can hold and Briska says so, instead of silently cutting your document.
+Create a skillBinds to the next free key
Polish · what you selected → what it becomes
“I just wanted to check in and see if maybe you had any thoughts about the proposal yet.”
Could you share your feedback on the proposal when convenient?
When it isn’t on a key, say it
Hold ⌥0, state the instruction, release. With text selected it rewrites in place; with nothing selected it still sets reminders, schedules meetings, drafts email and opens apps.
- Phrased any way you like. “Turn this into bullets”, “make it shorter”, “remind me to send this on Friday”.
- Nothing happens without approval. Anything with a real-world side effect shows an editable confirmation card first.
Try sayingStyles: Polish · Email · List
Or in your own words
It can sit in on the call, too
Briska records your microphone and your Mac's system audio as separate tracks, then writes the summary, the decisions and the action items.
- Ask it afterwards. Questions about the discussion come back with citations linked to exact timestamps.
- Why beta: it works, and it is the part still being shaped. Dictation is what we would stake the app on today.
Me
Let's ship the on-device inference engine first and test offline.
Speaker 1
Agreed. I'll finish the speech engine migration by Thursday.
Speaker 2
Can we make sure the changelog is drafted before release?
Me
Yes, I will handle the documentation.
Summary
The team aligned on shipping the on-device inference pipeline first and verifying offline performance.
Action items
- Speech engine migration by Thursday · Speaker 1
- Draft documentation before release · Me
Your words
It learns the names you actually say.
Every dictation tool can spell “meeting”. The one that matters is whether it can spell your colleague, your product, and the acronym your team invented last quarter. That is the one you would otherwise stop to fix by hand, every time.
Teach it your words
Add a colleague's name, a product name or a piece of jargon once, with how it sounds if that helps, and Briska spells it right from then on: in dictation, in commands, in meeting transcripts.
- A guarantee, not a suggestion. Your words aren't just hints to the model. A separate correction pass rewrites near-misses to your exact spelling, and leaves ordinary words that merely resemble a term alone.
- Your jargon stays yours. The list is one small file on your Mac, like everything else here. It is never uploaded, because there is nowhere to upload it to.
Kubernetestermsounds like “kuber-netus”
Jonpersonsounds like “john”
Xylotechcompanysounds like “zylo tech”
Corrected as you dictate
Heard “kuber-netus”, wrote Kubernetes.
Privacy you can check in four seconds
Runs on your Mac. Stays on your Mac.
Briska has no account, no server and no telemetry. Not as a policy we could revise, but as a fact about how it is built: there is no mechanism that could send your voice anywhere.
The four-second proof
Turn off your Wi-Fi and keep dictating. Everything works. An app that had to send your voice somewhere couldn't do that, and no policy document can fake it.
Your words train nothing
What you say is never used to improve a model, ours or anyone else's. The models arrive finished and stay that way.
Sensitive fields stay unwritten
Dictate into a password field and nothing is saved: not the text, not the audio.
One folder, yours to empty
Everything lives in a single folder on your Mac that you can inspect or purge in one click.
Compare
The same job, two different bargains.
Briska isn't the only dictation tool, or the only on-device one. So the right-hand column is a category, not a product: these are the trade-offs you usually meet, not a claim about any particular app.
| What you get | Briska | Other dictation tools |
|---|---|---|
| Finished writing, not a transcript | Every time, on your Mac | Most get there by sending your voice to a server |
| Works with the Wi-Fi off | Both models sit on your disk | The cloud ones stop working entirely |
| An account before you can start | None. There is nothing to sign in to | Usually required before the first word |
| Your names and jargon, spelled right | A correction pass guarantees it | Hints to the model at best, and often nothing |
| Rewrites text already on your screen | ⌥1–9, in any app | Rare. Most only transcribe what you say next |
| What it costs a year from now | One payment, never a subscription | Usually a monthly bill that doesn't stop |
Questions
The things worth knowing.
Including the answers that aren't flattering. We would rather you found out here than after installing.
Where does my voice go?
Nowhere. There is no account, no server and no telemetry, so the app has nothing to send audio to. Its only network requests are model downloads: most on first run, and one each for voice commands and speaker labels if you switch those on. It works with the network off after that. The one thing that does leave your Mac is a reminder, calendar event or email draft you explicitly confirm, which is handed to Apple's apps and syncs wherever you've set them to.
Is Briska open source?
No, so we won't pretend you can audit it line by line. What we can do is not ask for trust where trust isn't necessary: every privacy claim we make is either checkable in four seconds (turn off your Wi-Fi and keep dictating) or a thing we simply never built, like an account system or telemetry.
What languages does it understand?
English only, for now. The speech model is English-only and we would rather say that plainly than have you find out after installing.
What Mac do I need?
An Apple silicon Mac running macOS 14.5 or later. The models are built for the Neural Engine and the Metal GPU, so Intel Macs are not supported. The download is Apple-silicon-only, which means an Intel Mac finds that out straight away rather than after 3.2 GB.
How much disk does it use?
About 3.2 GB, downloaded once: that's the app plus the two models it needs to dictate. Allow 4 GB free for headroom and the optional models. What you dictate takes up almost nothing next to that, and you can have Briska discard the audio as soon as it has the text.
What does it cost?
Nothing. Briska is free. If that ever changes, anyone who installed before then keeps it free, for life — so installing today costs you nothing later either. And it would never be a subscription: Briska has no server, so there is nothing to rent you.
When can I get it?
Now. And yes: the privacy app is asking for your email, and we know exactly how that reads. So, plainly: it's an address, not an account. Nothing to sign in to, nothing to verify, no “check your inbox”. The download starts on the same click, and the address just sits in a table until it has a reason not to. That reason is rare: a new version, or a change to Briska big enough that you would want to hear it from us. Nothing else, ever.
Turn off your Wi-Fi. Then try it.
Briska is free, and it stays free for you even if that ever changes. An address, not an account: the download starts on the same click.