Stop typing. Start talking. 150 words a minute.

Reply to emails three to five times faster.

Read the message on screen and answer it out loud. Zumbo writes the reply, punctuated, into Mail, Gmail, Slack or wherever the cursor is. People speak at about 150 words a minute and type between 30 and 50, measured in the studies on our speed page.

Chat with AI apps at talking speed.

Speak your prompt. Zumbo types it into the chat box with your file names, product names and technical terms spelled right, and you press Enter. A prompt that takes two minutes to type takes fifteen seconds to say.

Writes what you meant, not every sound you made.

Zumbo drops the ums and uhs, collapses repeated words and trims false starts before the text lands, so a sentence reads the way you meant it in your own words. Nothing is rewritten and nothing is paraphrased: every word that stays is one you said. Each rule is a switch in Settings, and the group off gives you verbatim.

About a fifth of a second. The fastest open source dictation app there is.

Stop talking and the sentence is in your app about 200 milliseconds later, transcribed on the Mac's own chip with no server in the loop. Measured in the app over 30 recordings: 190 ms median. Whisper based apps on the same Mac took 20 to 30 seconds on a two minute recording.

The code that does it is open source, GPL-3, on GitHub. Read it, build it, or take the signed build with the model inside.

Every app. One shortcut.

Press your shortcut to start, press it again to stop. Or hold it and talk. Terminal, Slack, Mail, Xcode, Notion, a browser form: if you can put a cursor there, Zumbo can type there. Escape cancels. The shortcut is yours to pick in Settings.

Wrong once. Never again.

Select the word Zumbo got wrong, type what you meant, press Teach. From then on it hears it your way. Corrections stay on your Mac and stack into a vocabulary that is yours alone. The more you use it, the more it sounds like you.

Notes without a notes app.

Press New note and talk. The text stays in Zumbo, searchable, never pasted anywhere. Say a time and it becomes a reminder: “remind me to call Dana tomorrow at ten” sets one for tomorrow at ten.

Meetings, written down while they happen.

Turn on Meeting mode and Zumbo transcribes the room through the microphone, live, a timestamp per paragraph, each speaker labeled. On a call, use the speakers instead of headphones and the other side is in the notes too. Nothing leaves the Mac.

Otter, Fireflies and Krisp promise not to misuse your meeting data. Zumbo never receives it. There is nothing to retain, train on or subpoena.

It speaks your field.

21 vocabulary packs, from developer tools to legal, medical, design and marketing, switched on by what you dictate most. Product names, technical terms and commands come out spelled right the first time.

100 percent private. Built for teams that cannot send audio to a server.

Zumbo has no server for your voice. Audio is processed on the Mac by the local AI models that ship inside the app. Nothing is uploaded, cached or logged anywhere else, and there is no account.

For a security review, this is the complete list of what the app ever sends: the license key when you activate, a version number once a day to check for updates, and a feedback message only if someone writes one. Block all three at the firewall and dictation still works, on an air gapped Mac too.

There is no data retention policy to read and no processor agreement to sign for your voice data, because no third party ever receives it.

The source code is public on GitHub, so a reviewer can read what the app does instead of taking our word for it.

Team license: 10 Macs for $79, one download, installed like any other Mac app.