Download and start
Download the DMG matching your Mac: arm64 = Apple silicon, x86_64 = Intel. Open it, drag Hall-e to Applications, eject the disk image, and open Hall-e. ZIP alternatives and SHA-256 checksums are included. Requires macOS 14+; meeting-app audio requires 14.2+.
These downloads are ad-hoc signed and are not Apple-notarized. After verifying the download source, try to open Hall-e once. If macOS blocks it, close the warning, open System Settings → Privacy & Security, scroll to Security, click Open Anyway beside Hall-e, authenticate with Touch ID or your Mac login password, then click Open. This approves Hall-e specifically; do not disable Gatekeeper globally. Managed Macs may require IT approval.
Your first meeting
Follow four guided setup steps, allow the microphone, and choose your meeting language. For cloud transcription, connect your own Deepgram or Speechmatics key with Save & test, then allow audio processing. Speechmatics also requires a processing region and confirmation that Model Training is off. The connection check sends no audio and creates no paid transcription job; provider credit and model access still depend on your account. Supported local transcription instead needs Speech Recognition permission and an on-device language model, with no account or API key. Try the optional eight-second audio check.
Choose Start recording, name your meeting, and select Microphone + your meeting app for online calls. Join the call first and tell participants before recording. If app audio is unavailable, Hall-e shows a microphone-only warning and a retry action. Notes save automatically on your Mac; export your notes and transcript together as Markdown.
Existing recordings and provider choices are preserved. CI tests and fresh-profile startup checks run separately for Apple silicon and Intel. Real microphone permissions and live calls still need an interactive check on each Mac; automatic AI summaries are optional integrations, not part of local transcription.
Reliability fixes
Deleted recordings no longer reappear when a delayed processing callback runs. Deletion waits for active processing, and scheduled retries are cancelled after deletion. Failed recording sessions release their UI state correctly. Speechmatics submissions now require durable local checkpoints before uploading audio and before polling a newly created job.