Your agents can attend the meeting, not just read the notes
Fleece AI Team sends an AI agent into a video call as a live participant. You paste the meeting link, pick which agent goes, and it joins the call on camera with its own 3D portrait and cloned voice, hears what is said, and answers out loud in real time. It works with Google Meet, Zoom and Microsoft Teams, and the agent behaves exactly as it does on WhatsApp, Slack or the phone — same instructions, same knowledge, same voice.
Pick your platform
Google Meet
Admission prompts, Workspace policies on external participants, and what the agent's tile looks like to the room.
Zoom
Waiting rooms, passcode links, and the host controls that leave an agent present but inaudible.
Microsoft Teams
Lobby behaviour is tenant policy, not a per-meeting setting — which makes this the platform most likely to need one change first.
A participant, not a recorder
Most AI tooling around meetings sits outside the conversation. A notetaker joins, listens, and hands you a summary afterwards. That is useful, and it is also strictly retrospective: nobody in the room got an answer while the question was still live.
An agent from Fleece AI Team occupies a participant tile. It is announced in the roster like anyone else, it hears the discussion as it happens, and it speaks when addressed. If your support agent knows the refund policy, the sales call does not have to end with "let me check and get back to you".
The practical consequence is that a meeting stops being a place where questions are collected. Your specialist agent is in the room, so the answer arrives inside the meeting rather than in a follow-up thread the next morning.
The same teammate on every channel
An agent that joins your Monday standup is not a separate configuration from the one answering WhatsApp. It carries one identity across all seven channels: the same instructions, the same knowledge sources, the same personality traits, the same cloned voice.
That matters more than it sounds. Teams that wire a meeting bot separately from their chat automation end up maintaining two descriptions of the same colleague, and the two drift. Here the portrait your team sees on the call is the portrait on the agent's profile page, and the voice on the call is the voice on the phone line.
What it costs to run
Meeting time is metered by the minute, counted from the moment the agent joins to the moment it leaves — there is no rounding to the nearest block. The dashboard shows your remaining balance before you start, and a call will not begin without enough credit to run it.
The Team plan includes 200 meeting minutes per agent per month. Beyond that, minutes are billed as used; Scale plans lift the cap entirely. Because billing follows attendance, an agent parked in an empty room costs the same as one doing the talking — end the session when the meeting ends.
Every platform makes you knock
None of the three lets an unknown participant walk straight in, and the agent does not try to get around that. What differs is who holds the key, which is the single most useful thing to know before your first call.
On Google Meet, whoever is already in the call gets a prompt and admits the agent — the fastest of the three to get working, because the decision sits with the people in the room. On Zoom, the agent lands in the waiting room if the host enabled one, and the thing that catches people out is not admission at all but mute on entry: the agent joins, appears in the roster, and is inaudible until somebody unmutes it.
Microsoft Teams is the one that may need a change before anything works. Lobby behaviour and whether anonymous participants can join at all are tenant-wide policies set in the Teams admin centre, so the person hosting your meeting may simply have no ability to let the agent in. If that is your situation, the fix belongs with your administrator and a per-user meeting policy is the narrow version of it.
Which meetings are worth sending an agent to
Not all of them. An agent earns its place in a call where somebody is going to ask a question that the people present cannot answer — and that is a smaller set of meetings than it first appears.
External calls are the obvious case. A sales demo where the pricing question arrives mid-conversation, a technical review where the integration question needs the API documentation, a customer call where the policy question decides whether the call ends well. In each of those, the answer existing somewhere in your organisation is not the same as the answer being in the room.
Internal meetings qualify less often, and it is worth being honest about that. A weekly standup between five people who all know the work does not improve because an agent attended. Send agents where knowledge is missing, not where headcount is.
Questions, answered
Which video platforms are supported?
+
Google Meet, Zoom and Microsoft Teams. You paste the meeting link into the dashboard and the agent joins from there — there is nothing to install in the meeting platform itself, and no admin app to approve before your first call.
Do the other participants see and hear the agent?
+
Yes. The agent occupies a participant tile with its generated 3D portrait on camera and speaks aloud with its cloned voice. It is visible in the participant list under the agent's name, so nobody is recorded by something they cannot see.
Can the agent answer questions during the call?
+
That is the point of it. The agent hears the conversation live and responds out loud, using the system prompt, knowledge sources and personality you gave it. It is not transcribing for later — it is taking its turn in the discussion.
Does the agent need to be admitted by the host?
+
Usually yes, and that is deliberate. Every one of the three platforms puts an unknown participant through an admission step — Google Meet asks the host to let it in, Zoom uses the waiting room, Teams uses the lobby. The agent waits there like any external guest until somebody admits it.
How is meeting time billed?
+
Per minute, from join to leave. The Team plan includes 200 minutes per agent per month and Scale plans are uncapped. Your balance is shown before you start a call, and a session will not open without enough credit to run it.
Do I have to build the agent again for meetings?
+
No. Any agent in your organisation can be sent into a call, including ones you imported from OpenAI Assistants, LangChain, Anthropic Agents or your own API. Meetings are one more channel on an agent that already exists, not a separate product to configure.
Read next
- Fleece AI Team vs Otter.ai
Otter's notetaker sits in the same participant list but does not speak. The comparison sets out what changes when the tile answers back.
- Fleece AI Team vs Tavus
Tavus also puts a conversational agent into a Google Meet. The difference is who builds the application around it.
- Send an agent into Google Meet
Admission prompts, Workspace external-participant policies, and what the agent's tile looks like to the room.
- Send an agent into Microsoft Teams
Lobby policy is set by your tenant admin, which makes Teams the platform most likely to need one configuration change first.
Put a colleague in your next call
Create an agent, give it a face and a voice, and paste a meeting link. It joins as a participant and answers out loud — on Google Meet, Zoom or Microsoft Teams.