Send an AI agent into a Google Meet call
To send an AI agent into a Google Meet, paste the meeting link into the Video Calls page of Fleece AI Team and choose which agent attends. The agent joins the call as a participant with its 3D portrait on camera and its cloned voice active, waits for the host to admit it like any external guest, then listens and answers out loud for as long as the meeting runs. Nothing is installed in Google Workspace and no admin approval is required before the first call.
At a glance
- Meeting link format
- https://meet.google.com/xxx-yyyy-zzz
- Admission
- Host admits the agent from the knock prompt, like any guest
- Workspace app required
- None — nothing is installed in your Workspace
- What the room sees
- A participant tile with the agent's 3D portrait
- What the room hears
- The agent's cloned voice, answering in real time
- Billing
- Per minute from join to leave; 200 min/agent/month on Team
What actually happens when the agent joins
You open Video Calls in the dashboard, pick an agent from your organisation, and paste the Google Meet URL. The agent connects, appears in the meeting as a participant, and shows its generated portrait rather than a blank avatar or a static logo.
Because it arrives as an external participant, Google Meet does what it does with any unknown guest: it asks somebody already in the call to let it in. The agent waits at that prompt. Once admitted, its microphone opens and it starts following the conversation.
From that point the agent behaves like the colleague you configured. It answers when addressed, draws on the knowledge sources attached to its profile, and keeps the personality traits you selected. When you end the session from the dashboard, it leaves the call and the meter stops.
Google Workspace policies that get in the way
Two Workspace settings account for nearly every failed first attempt. The first is the host-management policy that decides whether external participants can join at all: if your administrator has restricted meetings to people inside the organisation, the agent will be refused at the door regardless of who clicks admit.
The second is quick access. With quick access off, every participant who is not on the calendar invitation has to be admitted individually — so somebody has to be watching for the prompt when the agent knocks. Turning quick access on for a recurring internal meeting removes that step, at the cost of letting anyone with the link walk in.
Neither setting is something Fleece AI Team can override, and it should not be able to. If your agent cannot get into a call it could join yesterday, the Workspace policy is the first thing to check.
Meet's own AI is not the same thing
Google Meet already includes AI features through Gemini: transcription, summaries, note-taking against the calendar event. Those act on the meeting after the fact, and they act on behalf of the platform rather than on behalf of a colleague you defined.
An agent from Fleece AI Team is the opposite arrangement. It has no privileged access to Meet; it participates on the same terms as a human guest, and it carries your instructions, your knowledge base and your voice rather than the platform's. The two coexist without conflict — Gemini can keep summarising a call your agent is speaking in.
What the agent brings into the room
The portrait is generated once, on the agent's profile, and is what appears on camera. Teams that skip this step get a tile showing initials, which reads as a bot and gets treated as one.
The voice comes from a 30-second sample you cloned, or from one of the preset voices if you would rather not clone anything. It is the same voice the agent uses on a phone call or a WhatsApp voice note, which is what makes it recognisable across channels.
The instructions are the agent's system prompt plus its knowledge sources. A support agent that has been given your refund policy will apply that policy in a meeting exactly as it does in a Slack thread — there is no separate meeting configuration to keep in sync.
How to send an AI agent into a Google Meet call
- 01
Give the agent a face and a voice
On the agent's profile, generate a 3D portrait and clone a voice from a 30-second sample, or pick a preset voice. Without both, the agent joins as an unnamed tile and the room treats it as a bot.
- 02
Copy the Google Meet link
Take the meet.google.com URL from the calendar event or the Meet window. The agent joins from the link — there is nothing to install in Google Workspace.
- 03
Start the session from Video Calls
Open Video Calls in the dashboard, select the agent, paste the link and start. Your credit balance is shown before the session opens.
- 04
Admit the agent
Google Meet asks somebody in the call to let the agent in, as it would for any external guest. Once admitted, the agent's camera and microphone come on.
- 05
End the session when the meeting ends
Leave from the dashboard. Time is metered from join to leave, so an agent left in an empty room keeps costing minutes.
Questions, answered
Do I need a Google Workspace add-on or admin approval?
+
No. The agent joins from the meeting link as an external participant, so there is no Workspace Marketplace app to install and no administrator to approve before your first call. Existing Workspace restrictions on external participants still apply.
Why is my agent stuck waiting to be admitted?
+
Google Meet is asking somebody already in the call to let it in. If nobody clicks admit, the agent waits. Turning on quick access for that meeting removes the prompt for everyone with the link, which is a trade-off worth making only for internal recurring calls.
Can the agent join a Meet that has already started?
+
Yes. Nothing about the flow depends on the agent being there at the start — you can send it in mid-call when a question comes up that it is the right colleague to answer.
Does the agent see screen shares?
+
The agent participates as a video and audio participant. Treat what is said aloud as what it can act on, and put anything it needs to reason about into its knowledge sources rather than assuming it read a slide.
Will people know they are talking to an AI teammate?
+
Yes, and they should. The agent appears under its own name in the participant list with its own portrait. Nothing about the setup is designed to pass an agent off as a human colleague.
How much does an hour-long Google Meet cost?
+
Meeting time is metered per minute from join to leave. The Team plan includes 200 minutes per agent per month, so an hour-long weekly call fits inside the allowance with room to spare; Scale plans remove the cap.
Read next
- Send an agent into Zoom
Zoom's waiting room behaves differently from Meet's knock prompt, and host controls can mute a joining participant.
- Send an agent into Microsoft Teams
Lobby behaviour on Teams is set by tenant policy rather than per meeting, which changes who has to act.
- Fleece AI Team vs Otter.ai
Otter's notetaker also shows up in the Meet participant list. The difference is whether the tile answers.
- All video meeting platforms
How metering, portraits and voices work across all three platforms.
Send your first agent into a call
Create an agent, generate its portrait, clone a voice, then paste a meeting link. It joins as a participant and answers out loud — and stays reachable on six other channels afterwards.