Ambient records room audio and calls on your Mac, transcribes speech locally, and separates speakers. Its menu bar controls recording; its window lets you browse, search, name speakers and export transcripts. No bot joins the call.
Start with Getting started for installation and a short test recording. Release builds target Apple Silicon Macs with macOS 14.4 or later and include the speech models.
Choose a guide
| Your task | Guide |
|---|---|
| Install and record for the first time | Getting started |
| Select audio devices or watched apps | Settings |
| Understand consent, storage and deletion | What is kept |
| Fix an empty recording or a permission problem | Troubleshooting |
| Use the CLI | Commands |
| Give an assistant read access to sessions | MCP |
| Change the app | Developing Ambient |
| Build the documentation | Building these docs |
Know the limits
Ambient is early software. Speaker separation groups voices; it does not know people’s names until you assign them. Recognition can get names and specialist terms wrong. Check the transcript before treating it as a record of a decision.
Capture, transcription and diarization have been exercised on an Apple Silicon Mac. The measurements name the hardware and fixtures; they are not a performance promise for every Mac. Hosted CI checks the code and bundle assembly, not microphone permissions or live capture.
The repository has releases through v0.0.3. See Releases for available builds. The release process signs, notarizes and staples the app; a fresh-machine installation and capture check remains part of release acceptance.
Keep control of the data
Capture and inference stay on the Mac. Exporting a transcript or connecting an MCP client creates another way to share its contents. A cloud-synced sessions folder can upload files independently of Ambient. Record only with appropriate permission, and read what is kept before changing retention.
Ambient’s source is MIT licensed. Models have separate terms documented in the repository’s third-party notices.