Keep the words close

Offline meeting transcription for Mac.

Scribe turns meeting audio into a transcript on your Mac. Once a speech model is installed, normal recording and transcription do not need a network connection.

That makes a transcript useful for an in-person interview, a conversation on unreliable Wi-Fi, or work you want to keep local. A video call still needs the connection required by its meeting service. Offline transcription describes how Scribe processes your recording, not how Zoom or another call app connects people.

Prepare before you disconnect

1. Install the Mac app. Download Scribe, open the DMG, and drag it into Applications. You need macOS 15 or newer on an Apple silicon or Intel Mac.
2. Download a transcription model. Follow guided setup or open Scribe Settings. The app offers English, multilingual, and compact choices and shows whether a model is installed, selected, or downloading. English is the recommended starting point for English conversations.
3. Let the download finish. Installing a speech model requires internet access and available disk space. Confirm that your intended model is selected before you leave. An installer alone does not include a prepared transcription setup.
4. Test the workflow. Use Settings → Review setup and test audio…, then make a short practice recording and review its transcript. Test with your usual microphone and call setup so you can resolve permission or audio problems while connected.

Record and review on your Mac

Tell participants you plan to record and obtain the consent your conversation requires. Choose Record yourself. Scribe captures your microphone and Mac system output as separate tracks; a detected-call prompt never starts the recording without your choice.

Choose Stop when you finish. Scribe saves the audio and processes transcription locally using the selected speech model. If you chose audio-only mode during setup, prepare transcription before expecting a transcript. Allow processing to finish, then listen to important passages and check names, numbers, and technical terms.

Two audio tracks do not identify every person in a group. Your microphone is distinct from the call track, while individual remote speakers need your review and confirmed tags. The Mac meeting-recording guide covers permissions and speaker names in more detail.

A transcript and an AI summary are separate

You can record, transcribe, and read the transcript without connecting an AI provider. Summaries, decisions, action items, and open questions require a capable model connection. Scribe does not turn fragments of the transcript into a pretend AI analysis when no model is configured.

For local summaries, choose On-device Scribe → Use an installed local summary model and select a llama.cpp executable with a compatible GGUF model. This is a separate setup from the speech model. Prepare the executable and model beforehand, then use Test connection. That test uses synthetic text rather than one of your meetings.

A remote AI provider is optional. It requires your own API configuration and a network connection when used. If you want an offline workflow, use installed local models or continue with transcripts and your own notes. Model quality varies, so review generated claims before sharing.

Keep a deliberate record

Scribe’s default recordings folder is ~/Recordings. Meetings live in ordinary folders containing audio, a readable transcript, and note files. There is no account required and normal recording and transcription do not upload your meeting to a transcription service.

Where you place or share those files is a separate choice. A cloud-backed folder may sync through its provider, and sending an exported note shares it with that destination. Keep a separate backup of recordings you need to retain. For a smaller, readable document to carry into your project, export meeting notes as Markdown.