MBox AI Meet

MBox reads the caption stream during a Google Meet session and turns it into a working transcript while the call is still going. You can watch the text build up in the side panel, which helps when you want to check that a name or a figure came through correctly. Nothing is recorded to disk, so the transcript is the only artifact.

When the call ends, the extension sends a condensed AI meeting summary to the address tied to your Google account. It covers the main topics, the decisions, and the loose ends, usually in a few short paragraphs. The email is the default delivery route, so the recap is waiting even if you close the tab straight away.

A slider in the panel sets how much detail you get. A brief mode returns a handful of bullets for a status call, while a deeper setting expands into context, reasoning, and next steps for a planning session. You pick per meeting rather than committing to one format for everything.

The model scans the conversation for commitments and flags them as tasks with the person attached. It catches phrasing like "I'll send the draft Friday" that a keyword search would miss. You still confirm the list before it goes anywhere, which matters when people talk in hypotheticals.

Each line in the transcript carries a speaker label, so a long discussion stays readable after the fact. In a five person call the labels make it obvious who raised a concern and who answered it. Accuracy depends on distinct voices and a decent microphone.

Audio and video never leave the call, and the Pro plan adds end to end encryption for the text it keeps. The same extension handles multi-language meetings, so a Spanish or Japanese session produces notes in the language spoken. Pro also unlocks a stronger Gemini Pro configuration with larger token limits, which helps on long calls where the basic model trims detail.