Magic Record transcribes the audio coming out of any app on your Mac — meetings other people host, softphone calls, podcasts, course videos. Then paste the transcript into ChatGPT or Claude for a summary, action items, anything.
Whisper runs locally. No cloud. No login. Nothing leaves your Mac.
Magic Record lives next to the clock. There's no main window, no Dock icon, no app to learn. Four clicks from sound to summary.
Pick microphone, or any app you can hear — Zoom, Meet, Teams, a softphone, a course video tab. Live or batch mode.
Join the meeting. Take the call. Watch the lecture. Magic Record listens in the background and builds the transcript as it plays.
Text lands on your clipboard instantly. Optionally also written to ~/Documents/Magic Record/ as a dated .txt.
Drop it into ChatGPT, Claude, Gemini, or Magic Note. Ask for a summary, action items, a translation — whatever you need.
Magic Record ships with the model already inside it. The base 4-bit Whisper weights are bundled in the .app — first transcription is instant, with no setup or download. Upgrade to a larger model anytime; everything keeps running locally.
Zoom calls you joined as a guest, Google Meet, Teams, Webex — anywhere the host won't (or can't) hit "Record." Magic Record captures the audio from your end. No bots in the call, no permission requests.
Dialpad, RingCentral, Aircall, Vonage, Microsoft Teams Phone, native FaceTime audio — any softphone that pipes audio through macOS. Sales calls, customer calls, support escalations.
Udemy, Coursera, MasterClass, a Loom embed in your internal LMS, a YouTube lecture, a podcast. If the audio is playing on your Mac, Magic Record captures it.
Every transcript copies to your clipboard automatically. Drop it into ChatGPT, Claude, Gemini, or Magic Note. Ask for a summary, action items, a translation, a tweet thread.
Audio buffers and transcripts never leave your Mac. Whisper runs on-device. No telemetry, no login, no API key, no cloud anywhere in the loop.
Live mode streams text as the meeting plays, VAD-chunked, model kept warm. Batch mode records straight through and transcribes once at the end. Pick per recording.
Trial starts automatically the first time you launch the app. Buy anytime — your existing recordings stay where they are.
Most transcription tools upload your audio to a cloud server. Some keep it. Some train on it. Magic Record never sends audio anywhere — the model runs on your machine, the transcript is generated on your machine, and the only thing that crosses the network is a one-time license check when you activate a paid copy.
There's no account to create. No analytics in the app. The bundle you install is the bundle that runs.
Magic Record is the standalone tray transcriber. If you want voice → text plus a notebook that organizes your prompts, screenshots, web imports, and AI threads, the full Magic Note app has all of it — and it's free.
A notebook for your prompts. Local-first, multi-provider AI, free for macOS.