Local Transcriber
Free · Open source · for macOS

Local Transcriber

Meetings, calls and dictation – transcribed on your Mac. No cloud. A small, helpful tool in the menu bar that listens, takes notes and knows who said what.

Install – paste into Terminal: 
curl -fsSL https://transcriber.teconomix.net/install.sh | bash

Version 0.7 · October 14, 2026 · macOS 14 or later

Instructions Source code on GitHub
  • Mac with Apple silicon
  • Around 100 languages
  • No account, no subscription

What Local Transcriber does

Record, transcribe, identify speakers, dictate – all on your Mac, nothing in someone else’s data center.

Fully local & private

Recording, speech recognition and speaker recognition run entirely on your Mac. No cloud, no account, no subscription, no license server. Audio never leaves your computer.

Record meetings

One click in the menu bar. Microphone and system audio – that is, the other people on the call – are recorded as separate tracks.

Transcription uses Whisper large-v3-turbo on the Apple silicon GPU: one hour of audio, including speaker recognition, in about three to four minutes.

Who said what

Speaker recognition (pyannote) with a voice database that learns: name a voice once, and it is recognized next time. Uncertain matches are marked with “(?)”. Set the number of people or separate speakers again later, share voices with your team.

Video calls

Detects Teams, Zoom, Google Meet & co. and asks: “Record?” The recording uses the microphone the call uses – even if you switch it midway – and stops automatically when the call ends. Your voice and the others’ are kept apart via separate tracks – in testing, even with speaker echo.

Calendar

The current event becomes the file name, and its attendees are suggested when you name the voices.

Dictation everywhere

Hold the right ⌘, speak, release: the text appears at the cursor – in Mail, Slack, your browser or Terminal. 10 seconds of dictation in about 0.2 s.

With the right ⌥, Apple Intelligence (on-device) cleans up filler words and punctuation.

Meaningful file names

From your calendar or via on-device AI:

2026-10-10 14-03 Q4 budget planning
Recording 14-03

For power users

Folder watcher (optional): files in ~/Transcriber are transcribed automatically. Output as Markdown, JSON and SRT, a glossary for names and technical terms, a command line and an MCP server for AI agents such as Claude.

How it works

Three steps from installation to your first transcript.

  1. Install

    One command in Terminal – the installer takes care of the rest.

  2. Grant permissions

    Microphone and system audio. Optional: Accessibility for dictation and access to your calendar.

  3. Record or dictate

    Click in the menu bar or hold the right ⌘. Shortly afterwards, the transcript is saved next to the recording.

A 70-second tour: record a call, recognize speakers, dictate. No sound.

Clear, simple settings

A few understandable switches – just what you would expect from a Mac app.

Recording settings: record system audio, voice processing for the microphone, transcribe and give a descriptive name after recording, detect calls and offer to record.
Recording
Dictation settings: keys for verbatim dictation (right Command key) and AI cleanup (right Option key), text of the active window as context, feedback via sound and icon.
Dictation
Calendar settings: use the event for the recording (title, attendees) and choose which calendars to include.
Calendar

Installation

One command, about ten minutes of patience – after that, everything runs offline.

Requirements

  • Mac with Apple silicon (M1 or later)
  • macOS 14 Sonoma or later – for AI features such as cleanup and file names: macOS 26 with Apple Intelligence
  • about 8 GB of free disk space (the speech recognition model alone ~3 GB)
  • 16 GB of memory recommended
  • Internet only for installation

Step by step

  1. Open Terminal: press ⌘ + Space, type “Terminal”, press ↩. (Or in the Finder: Applications › Utilities › Terminal.)
  2. Paste the command: click “Copy”, then press ⌘ + V and ↩ in Terminal. 
    curl -fsSL https://transcriber.teconomix.net/install.sh | bash
  3. Wait about ten minutes. The installer downloads the Python environment, the models and the app and shows every step. At the end it asks whether to watch the folder ~/Transcriber: y for yes, just ↩ for no. 
    Transcriber 0.7 – Installation

▸ Checking requirements
  ✓ macOS 15.6, Apple Silicon, 212 GB free
▸ Downloading Transcriber 0.7
  ✓ Checksums verified
▸ Setting up Python environment (uv, Python 3.12, packages – takes a few minutes)
  ✓ Python environment ready
▸ ffmpeg
  ✓ ffmpeg installed (static build)
▸ Models
  ✓ Speaker recognition (pyannote community-1, CC-BY-4.0)
▸ Downloading speech recognition model (Whisper large-v3-turbo, ~3 GB – may take a while)
  ✓ Whisper large-v3-turbo
▸ Installing the app
  ✓ ~/Applications/Transcriber.app
  Watch the folder ~/Transcriber (automatically transcribe audio files dropped there)? [y/N]

Done. Transcriber 0.7 is installed.
  4. Grant permissions: Transcriber starts in the menu bar. macOS asks once per feature the first time you use it – click “Allow”.
    • Microphone – your voice
    • System audio – the others in the call
    • Accessibility – for dictation only
    • Calendars – optional, event as file name

Prefer Homebrew? The same in one command, then update with brew upgrade:

brew install --cask teconomix/tap/local-transcriber

About Gatekeeper: The app is not (yet) notarized by Apple. The installer sets it up so that it starts anyway.

Uninstall: bash ~/.local/share/transcriber/uninstall.sh (with Homebrew: brew uninstall local-transcriber). Your recordings and transcripts are kept.

Stuck? Ask in the GitHub Discussions – other users can help there too.

Frequently asked questions

Does it cost anything?

No. Transcriber is open source under the MIT license and free of charge – no account, no subscription, no ads.

Does it need an internet connection?

Only for installation. After that, everything works offline; recordings and transcripts stay on your Mac. If you agree, the app checks the current version number once a day to tell you about updates – nothing else.

Which languages are supported?

Around 100 – all the languages the Whisper speech recognition model knows. German, English, Spanish and French are preselected; you can add more in the settings. Widely spoken languages work best, rare ones considerably less well.

Am I allowed to record conversations?

That is your responsibility: get the consent of everyone involved before you record – depending on the country, this is required by law.

Where can I get help?

In the GitHub Discussions: ask questions, suggest ideas, show what you use Transcriber for. The easiest way to report a bug is right in the app, with the bug icon in the panel.