Why ordinary recording falls short
Your microphone is designed to capture your voice, not clean digital output from Zoom, Teams, Meet, Slack, Webex, or FaceTime. Turning the speakers up and recording the room mixes both sides with echo, keyboard noise, and acoustic processing. A cloud meeting bot avoids that acoustic problem but becomes another participant, may require calendar access, and sends the call through an external service.
The two-track approach
LokalBot records two synchronized audio sources: the selected microphone and a Core Audio process tap for the selected meeting application's output. The transcript labels describe where the audio was captured, not a verified speaker identity. A shared microphone, headphones leaking into the mic, or meeting audio played into the room can put the same voice on either source.
When several voices occur in the system-audio source, optional on-device diarization can estimate additional speaker turns. Diarization is probabilistic, so identities and boundaries may still need editing after a difficult or overlapping conversation.
What the other participants see
No account joins the room, no virtual participant appears in the roster, and no bot announces itself. That makes the workflow less disruptive, but it does not make recording invisible in a legal or ethical sense. Tell people when required, obtain consent, and follow employer and platform policies. LokalBot is a personal recorder; it cannot decide whether a particular meeting may be recorded.
Permissions and controls
- Microphone access captures your side.
- System-audio access enables the process tap for the other side.
- Ask via notification is the fresh-install default: LokalBot asks before it records a detected meeting.
- Automatic or never-auto-record modes can be chosen in Settings.
- Calendar access is optional and can help detect and title scheduled meetings.
After the call
The selected local speech model transcribes the synchronized audio. The built-in language model can turn the transcript into a TL;DR, decisions, and action items. Because the capture sources are separate, you can replay the microphone and selected-app system audio independently and search from a result back to the relevant timestamp. Those sources still describe audio provenance rather than verified identity.
A pre-meeting test worth doing
Open the actual meeting app, play remote audio, speak into your microphone, and make a short recording. Verify that both tracks show activity and play back correctly. macOS permissions and device routing can change when you switch headsets, docks, or output devices, so repeat this check before an unusually important call.