Why Overshow doesn't send a bot to your meetings

Most AI meeting tools work by joining the call. A notetaker appears in the participant list, sits through the conversation, and uploads what it hears to the vendor's servers. Overshow does not do that, and the reasons are worth setting out plainly.

A bot in the participant list changes the meeting

People speak differently when a recorder with an unknown audience sits in the room. The bot belongs to one attendee, but everyone else has to adjust to it: the aside about a difficult stakeholder gets swallowed, the half-formed idea stays unsaid, and the first two minutes become a negotiation about what "Notetaker" is and who admitted it.

That is a poor trade. The candour that makes a meeting worth remembering is exactly what a visible recording participant suppresses.

The platforms now treat bots as suspects

This is no longer just an etiquette question. By June 2026, Microsoft had documented Teams controls that detect external meeting bots, hold detected bots in the lobby, and require explicit approval under the default policy. Microsoft says these tools can introduce security, privacy, compliance, and data-leakage risks, while also warning that some bots may not be detected (Microsoft Learn).

Google moved in the same direction. Its safeguarded guest admit flow for Meet rolled out from late March into April 2026, queues higher-risk join requests separately, and denies them by default unless the host intervenes (Google Workspace Updates, 2026); industry reporting notes that third-party notetaker bots are among the requests flagged this way (UC Today, 25 March 2026).

If your record of a meeting depends on a bot being admitted, that record now depends on platform policy and on the host clicking the right button, every single time.

The consent question, honestly

In August 2025, a proposed class action was filed against Otter.ai in the Northern District of California, alleging that its notetaker bot recorded meeting participants who never consented and that the recordings were used to train its models (NPR, 15 August 2025). The consolidated case is still pending: a motion to dismiss was heard in May 2026 and there has been no ruling, no certified class, and no finding on the merits either way (case tracker). Whatever the outcome, the filing describes the uneasy pattern many people have felt: consent gathered by a bot is often consent by ambush, a banner glimpsed mid-join.

Here is our honest position. Recording on your own device does not make consent someone else's problem. Depending on where you and the other participants are, the law may still require you to tell people you are recording, and you should. What changes is who owns that obligation: with Overshow it stays with you, a person in the meeting who can be asked and can answer, rather than being delegated to a bot's join screen. Our recording consent guide covers this in detail, and the pause control exists precisely so you can honour an objection instantly.

What Overshow does instead

Overshow runs on your Mac and captures meeting audio from the device itself, only while a meeting is active or you start a recording manually. Transcription happens on the device using a local speech model, the transcript is stored encrypted on your Mac, and raw audio is removed after terminal processing. Interrupted or unresolved audio stays encrypted on the Mac for recovery; in the rare case where speech produces no usable transcript, its encrypted carrier can remain quarantined within the shared ten-hour durable-audio cap until a model or result-filter revision retries it, you delete it, or oldest-first eviction removes it. There is no bot in the participant list, no third party in the room, and no meeting audio uploaded for transcription. Approved cloud clients can receive transcript snippets you ask them to retrieve, but nobody has to admit anything into the call and your record does not depend on a vendor's bot surviving the next platform policy change.