Ready · free & open source

Speak.
It just types.

Drift is local-first voice dictation for macOS. Hold a key, talk, and clean, punctuated text lands in whatever app you're in, transcribed entirely on your Mac. No account, no API key, nothing uploaded.

100% on-device No sign-up MIT licensed
1,248 words
Hold to talk
342Dictations
47This week
6Day streak
Why Drift

Dictation that respects your Mac

A native Swift menu-bar app. No Electron, no cloud round-trip, no subscription.

Push-to-talk anywhere

Hold Right Option (⌥) in any app, speak, and release. Text is inserted right at your cursor, system-wide.

Private by default

On-device transcription and on-device cleanup. No account, no API key, and your audio never leaves the Mac.

Fast English dictation

FluidAudio's local Parakeet model gives you quick, accurate English transcription, ready to type instantly.

Multilingual, optional

Switch to the WhisperKit backend for Hindi, Tamil, Malayalam, Kannada, Telugu and more, still fully on-device.

Smart cleanup

Filler words removed and punctuation fixed instantly. Want more polish? Plug in a cloud or local LLM, your call.

Native & lightweight

A tiny Swift menu-bar app with a live overlay. No Electron bloat, no battery drain. MIT licensed.

Three steps

From thought to text in one breath

No modes to learn. Hold, talk, release. That's the whole interaction.

1
Right Option

Hold

Click into any text field and press & hold the Right Option key.

2

Speak

Say what you mean. A floating overlay shows Drift is listening, and the live text.

3
text

Release

Let go. Drift transcribes, cleans up filler & punctuation, and types it at your cursor.

Local-first, for real

Your voice stays on your Mac

By default, audio is transcribed on-device and cleaned on-device. Nothing is uploaded. Drift isn't private-by-marketing. It's private by architecture.

  • Audio never leaves the device. Transcription runs locally with the Parakeet model.
  • No account, ever. No login, no telemetry, no API key required to use it.
  • Cloud cleanup is opt-in. If you choose it, only the text (never your audio) is sent to an endpoint you configure.
Microphone capture16 kHz, in-memory
On device
Speech → textFluidAudio Parakeet / WhisperKit
On device
Cleanup & formattingDeterministic, instant
On device
LLM cleanupOpenAI-compatible / Ollama
Optional
Cleanup, your way

Raw transcript, or fully polished

The cleanup layer is model-agnostic. Stay 100% local, or point it at any OpenAI-compatible endpoint.

Default

On-device

Runs on your Mac

Instant, private formatting. Strips filler words and fixes punctuation with zero setup.

Hosted

Cloud LLM

OpenAI-compatible API

Highest quality. Works with OpenAI, Groq, Sarvam, or LM Studio. Just a base URL and key.

Advanced

Local LLM

Ollama on your Mac

Private LLM-grade cleanup that still never touches the network. Run Ollama and pick a model.

Verbatim

Raw

No processing

Insert the transcription exactly as heard. Perfect when you want every word as spoken.

Stop typing. Start talking.

Free, open source, and private. Drag it to Applications, grant two permissions, and dictate anywhere on your Mac.

No Homebrew, no Terminal, no separate downloads. Everything's handled inside the app.