Dictation for lawyers who cannot send audio to a server.

The question that stops most dictation apps at the door

Privilege and confidentiality do not care how good a vendor's security is. The moment a recording of a client conversation is transmitted to a third party, you have made a decision on someone else's behalf, and it is a decision a website's privacy page cannot make for you. That is why dictation usually dies somewhere in a firm's approval process rather than in the trial.

Zumbo answers it by removing the transmission. The speech engine ships inside the download and runs on your Mac's own silicon. There is no server for your voice, no account, no API key, and no network request needed to turn speech into text. Unplug the router and it works exactly the same.

What it does send, in full

• Your licence key, once, when you activate it.
• A version number, once a day, to check for an update.
• A feedback message, only if you write one and press Send.

• Speech engine inside the app
• 570 MB, downloaded once
• Works with the Wi-Fi off
• No account, no API key
• Unplug the internet and it works exactly the same.

That is the complete list. Block all three and dictation is unaffected. There is no data retention policy to read because there is no database holding anything to retain.

The vocabulary problem in legal work

General speech models spell what they think they heard, and what they think they heard is ordinary English. A case citation becomes an address. A Latin term becomes something adjacent. Counsel's surname becomes a common noun. Each one costs you twice: once to notice and once to fix, and in a document that goes out, once more if you do not.

There is a legal vocabulary pack covering contract and court terms, switched on during setup. For everything else, including the names on this particular matter, correct a word once with Teach and it is heard your way from then on. The corrections live on your Mac and never sync anywhere.

Attendance notes while the meeting happens

Meeting mode transcribes the room through the microphone as the conversation goes, a timestamp for each paragraph, each speaker labelled. The labelling model is an optional add on of about 220 MB downloaded once; everything else is already in the app. On a call, use the speakers rather than headphones and the other side is in the note too.

Nothing joins the call. There is no participant in the attendee list and no bot to explain to the other side. Whether you may record a particular conversation remains entirely your judgment and theirs; the tool simply does not add a third party to the question.

What to take to your risk people

Three sentences, which are the ones on the privacy page of this site and are the only ones that matter. Zumbo has no server for your voice: there is no account, no API key and no network request needed to transcribe. Audio is processed on your Mac by the speech engine that ships inside the app, and nothing is uploaded, cached or logged anywhere else. Zumbo has no data retention policy to read, because there is no database to retain anything in.

That is a materially different conversation from asking a firm to approve a processor, and it is usually a much shorter one.

What it costs

Fifteen dollars once for one Mac, twenty for three, seventy nine for a team of ten. No per seat subscription, no minutes metered, free updates. Three day trial with everything unlocked and no card, which is enough to test it on a real matter.

Related: dictation with the Wi-Fi off, the 21 vocabulary packs, and how Zumbo keeps your voice on your Mac.