What this is

An agent joins your Google Meet as a participant and makes things while you talk. These are the docs for when that goes right — and for when it doesn't.

There are two kinds of reader here, and the docs are split that way.

If you're a person who's about to be in a call with one, you need about two minutes:

before your first call, and — if it ever seems deaf — call language.

If you're an agent driving a bot in someone's meeting, you want what to ask for

and the tools, by name.

The shape of it

A macOS app puts a participant in a Google Meet with an emoji face and a synthetic voice. It

hears by reading Meet's own captions. It can write to a shared whiteboard, present that board,

and drive a web page it's presenting. One person runs the app; everyone else just turns up.

What it can't do. It cannot join invisibly — a human clicks Admit, and it arrives with a

name and a face. It cannot control anyone else's computer. It does not write a transcript file,

and there is no automatic summariser.

google-meet-provider.js:1879 · server.js:2043 · agent-workdir.js:157

What the honest version costs

Requires macOS on Apple Silicon, Google Meet, and the operator's own Claude subscription — the

app holds no model key of its own. Windows installs but does not start the agent. Slack huddles

are partial: joining and speaking are not wired.

claude-install.js:111 · main.js:6555 · slack-provider.js:10

Everything on these pages is checked against the app's source before it ships. Where we could

not verify something, it is either marked or absent. Blank beats wrong.