Futurelab Studio
Open Source · macOS 26.0+

MacTalk — Local Voice Dictation for macOS

An open-source menu bar app that transcribes your voice locally on your Mac. Choose between Whisper for accuracy or Parakeet for real-time streaming. Metal-accelerated, privacy-first, and built for people who talk to their computers more than they type.

macOS Whisper Parakeet
Download for Mac View on GitHub

Why MacTalk exists

MacTalk started as an ASR experiment — a playground for comparing on-device speech recognition engines on macOS. Instead of committing to one model early, the goal was to support multiple engines side by side and learn which trade-offs actually matter in daily use.

That experiment turned into a genuine tool. MacTalk is now a lightweight, always-available voice dictation app that lives in your menu bar and pastes transcriptions directly into whatever you are working on.

The fastest way to get a thought into a coding agent is to say it out loud.

Two engines, two philosophies

Whisper
High accuracy, multilingual, batch transcription
Parakeet
Real-time streaming, instant feedback, Core ML

Whisper (via whisper.cpp) offers multiple model sizes from tiny to large-v3-turbo. It processes audio in chunks and returns high-confidence results, making it strong for accuracy-sensitive work and multilingual dictation.

Parakeet (via FluidAudio) streams results as you speak. Words appear in near real-time, which makes the interaction feel less like dictation and more like a live conversation with your computer. This is the engine that helped shape the thinking behind Ora.

Key features

  • 100% local processing — no network calls, no cloud, no telemetry. Your voice stays on your Mac.
  • Metal acceleration — optimized for Apple Silicon with GPU-accelerated inference.
  • Menu bar app — lightweight and always one hotkey away. No dock icon, no window clutter.
  • Auto-paste — transcriptions go to your clipboard and optionally paste directly into the frontmost app.
  • Mic + App Audio — capture your microphone or combine it with system audio from calls and meetings via ScreenCaptureKit.
  • Multiple Whisper models — choose from tiny (fastest) to large-v3-turbo (most accurate) depending on your hardware.
  • Global hotkeys — start and stop transcription with customizable keyboard shortcuts.
  • Signed and notarized — the release DMG is Developer ID signed, notarized, and stapled for clean Gatekeeper behavior.

Built for talking to coding agents

The use case that brought MacTalk back into daily rotation is surprisingly simple: talking to coding agents. When the bottleneck is not the model but how fast you can externalize your intent, speaking a prompt is faster and more natural than typing one.

MacTalk's hotkey workflow — press, speak, release, paste — means the transcription lands directly in the agent's input. No extra windows, no copy-paste chain. Just voice to text to agent.

This works especially well with tools like Pi, Codex, and similar coding agents where the prompt interface is a text field and the interaction rhythm is short, iterative instructions.

How MacTalk led to Ora

MacTalk was the experiment. Ora was the product that grew out of it. By comparing Whisper and Parakeet side by side in a real desktop workflow, the shape of a broader local-first voice assistant became clearer. Parakeet's real-time feel was the specific insight that made Ora feel viable — and the full story is here.

Requirements

  • macOS 26.0 or later
  • Apple Silicon (M1 or newer) recommended
  • 8 GB RAM minimum

Get started

  1. Download MacTalk from the latest GitHub release
  2. Open the DMG and drag MacTalk to Applications
  3. Launch it — the icon appears in your menu bar
  4. Press ⌥Space to start recording, speak, then press again to stop
  5. Your transcript is on the clipboard (and optionally auto-pasted)

Models download automatically when selected. No manual setup required.