Talk to your computer.
Nothing you say leaves it.

Tangora is free, open-source voice dictation that runs entirely on your machine. Hold a key, speak, and your words land wherever your cursor is — no cloud, no account, no trial minutes.

macOS · Apple Silicon  —  Windows & Linux coming
The story

Named after the fastest typist who ever lived

In 1923, Albert Tangora sat down at a manual typewriter in front of a crowd and typed 147 words per minute — sustained, for a full hour. For the next century, that record stood as the ceiling on how fast a human could turn thought into text.

Here's the thing, though: you already beat him, every day. You just do it out loud. Most people type 40–60 words a minute — and speak at 150 or more. The fastest input device you own is the one you were born with.

So I wanted to talk to my computer. Simple wish — with one catch. Every tool I looked at came with the same fine print: your voice gets processed on our servers. And your voice is a biometric. Workdays are full of things that shouldn't leave the room — names, plans, half-formed ideas rambled at a text field. Streaming the most personal signal you produce to someone else's cloud, forever, in exchange for convenience is a bad trade. I wanted the convenience without the trade: dictation that works on a plane, in a locked-down site, on day one of an NDA. Everything on the machine. No uploads. Ever.

For years that was technically naive. Then it quietly stopped being naive: open-weight speech models — like IBM Granite Speech, Apache 2.0 — got good enough to transcribe faster than real time on an ordinary laptop. The entire cloud apparatus became optional.

And the name? In the 1980s, IBM Research built one of the first real speech-recognition systems and called it Tangora — after Albert. The name was a promise: someday you'll beat him by talking. The machines of that era weren't ready. The promise was just forty years early.

This app is that promise, kept. On your machine, and no one else's.

Four things, without compromise

Private

Audio is transcribed on-device and deleted. There is no server side to trust, because there is no server side.

Actually free

No trial minutes, no Pro tier, no account. The models are openly licensed and downloaded once.

Fast

Granite Speech and Parakeet run faster than real time on Apple Silicon. Speak at 150 WPM; it keeps up.

Open source

MIT licensed, forkable, auditable. Read every line of code that touches your voice.

Up and running in two minutes

  1. Download the latest .dmg from Releases and drag Tangora to Applications.
  2. First launch: right-click → Open (builds aren't notarized yet), then grant Microphone and Accessibility permissions.
  3. Pick a model in onboarding — Granite Speech 4.1 2B for accuracy, Parakeet for speed.
  4. Click into any text field, hold the hotkey, and talk.
  5. Release. Your words are there. Go beat Albert.

Where your words go

Audio is captured, transcribed locally, and deleted per your retention settings. Transcription history lives on your disk, only as long as you configure. The app touches the network for exactly one thing: downloading the models you ask for. That's the whole disclosure — one paragraph long, because there's nothing else to disclose.

Support the work

⭐ Star the repo

Stars are how open-source projects get found. github.com/asapalejandro/tangora

🐛 File issues

Found a bug, or want Windows/Linux sooner? Tell us what broke or what's missing.

💙 Support upstream

Tangora builds on Handy by CJ Pais. If you love the core experience, support the upstream project.