Push-to-talk transcription for macOS Apple Silicon only No subscription, ever

Fast speech-to-text anywhere on your Mac.

Humlet Transcribe is a fast, push-to-talk dictation app for macOS with no subscription. Use a local model for free offline transcription, or bring your own API key for streaming cloud speeds.

Download for macOS
Requires Apple Silicon and macOS 14+
What’s new in 0.7 build 19
Live
Notes · Untitled Cmd + N
Cursor was here
Try it:
Humlet widget
Local / Local / openai_whisper-base.en

A small overlay that appears when you hold your trigger. No tab to focus. No popup to dismiss.

How it works

You decide
when, how, and where.

Your trigger, your model, your insertion target. Humlet only listens while you are holding the key, and the text only lands where your cursor already is.

01

Hold.

Press your chosen trigger: Fn, Option, a side mouse button, or any custom hotkey.

Option
02

Speak.

A small overlay shows your voice activity. Talk naturally for as long as you hold the trigger.

Local / Local / openai_whisper-base.en
03

Release.

Text lands wherever your cursor already was: notes, browsers, chat, editors, AI tools.

Inserted into front app
Model providers

You pick
who hears you.

Run local WhisperKit models when audio should stay on this Mac, or connect directly to AssemblyAI, OpenAI, ElevenLabs, or Deepgram with your own key. Humlet does not proxy cloud audio or charge a subscription.

On this Mac Offline and local.

No audio leaves your machine after the model is downloaded.

Bring your own key Cloud when speed matters.

Direct provider connection. Your account, your limits, your bill.

Latest model from each provider
A

AssemblyAI

Streaming, multilingual

O

OpenAI

Realtime Whisper

E

ElevenLabs

Scribe, keyterms

D

Deepgram

Nova, streaming

By design

Push-to-talk only.
No ambient listening.

Humlet starts capturing audio only when you intentionally hold the trigger. There is no always-on listening mode, no background recording, and no hidden buffer.

01 - Intent

Recording is a held key.

Release the key and the mic stops. There is nothing else to remember.

02 - Storage

Nothing saved by default.

Recordings and transcripts are off unless you opt in. Saved logs stay on this Mac.

03 - Routing

Direct to the model.

Cloud audio streams straight to your chosen provider. Humlet does not proxy.

Support and feedback

Help shape Humlet.

Found a bug, missing a provider, or want a workflow to feel smoother?

Bugs and tracked requests

File an issue.

Whenever you encounter a problem, don't hesitate to file a bug. They will be fixed promptly.

Open GitHub issues

Ideas and requests

Suggest a feature.

If you miss a feature, or have an idea how to improve the app, don't hesitate to submit it here.

Open GitHub discussions

Optional support

Support on Ko-fi.

If Humlet saves you time and you want to support ongoing development, you can support it on Ko-fi.

Buy me a coffee

FAQ

Questions.

Short answers to things people ask before downloading.

Yes. There is no subscription to use the app. Local transcription runs on your Mac with no per-use cost. Cloud transcription is bring-your-own-key, so you pay your chosen provider directly. If you want to help, you can submit feedback, file issues, or support Humlet on Ko-fi in the support section.

No. Humlet Transcribe is Apple Silicon only. macOS 14 Sonoma or later is required.

Yes, when you pick a local model. Once downloaded, transcription happens entirely on-device.

Wherever your cursor was when you held the trigger. Humlet types or pastes directly into the active text field.

No. Humlet is strictly push-to-talk. The mic activates only while you hold the trigger and stops when you release.

Because this website does not use analytics cookies or ad tracking. We only count basic visits and a few button clicks in a cookieless way, so there is nothing to accept before you can read the site.

Currently: AssemblyAI, OpenAI, ElevenLabs, and Deepgram.