AI meeting agent API
An AI meeting agent API that puts an agent into Zoom, Google Meet or Microsoft Teams with one call. It listens, speaks, posts in chat and acts on tools, on the model you choose.
Everything an agent does in a meeting
Attach an agent to a bot and it joins the call as a participant. It hears the room, decides when it is being spoken to, and answers out loud, in the chat, or silently through your tools.

Spoken replies with barge-in and echo control.
Answers and links in the meeting chat.
Calls your tools with no output at all.
An image or an avatar as its video tile.
Put an agent in the call
Create the agent once, then pass its id when you send a bot. The full reference is in the docs.
- Any of three platformsZoom, Google Meet and Microsoft Teams from the same request.
- Realtime or pipelineRealtime for the lowest latency, or pick each stage yourself.
- Your modelOpenAI, Anthropic, Google or xAI, on your own keys.
curl -X POST "https://api.meetstream.ai/api/v1/bots/create_bot" \
-H "Authorization: Token <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{
"meeting_link": "<ZOOM_MEET_OR_TEAMS_LINK>",
"bot_name": "Meeting Agent",
"agent_config_id": "<YOUR_AGENT_ID>"
}'The controls that make it usable
A voice agent earns its place by speaking at the right moment. These controls are set per agent.
- Wake wordIn pipeline mode the agent waits for a wake phrase you set, such as its name, then keeps listening for 30 seconds. One to one calls can skip it.
- InterruptionsIn pipeline mode a short "uh-huh" does not stop the agent, but a real interruption does, so it stops only when it should.
- Turn detectionSilence detection or a turn detection model decides when a person has finished speaking.
- Three ways to replySpeak, post in the meeting chat, or act silently through tools with no output at all.
Connect the agent to your systems
Agents reach your systems through MCP servers or custom functions, so a meeting can update a ticket, a CRM or a doc while it is still running.
- MCP servers with an allow-list of tools
- Up to 20 custom functions per agent
- An optional filler line while a custom function runs
- Silent turns for pure side effects
The agent hears the room as it happens
Audio reaches the agent while the call is running, with live transcripts that carry speaker names.
- Mixed audio or one track per person
- Live transcripts with speaker names
- Echo control built in
- Over a WebSocket you host
What teams build with it
The same agent runtime powers very different products, from notetakers to agents that act on what they hear.
- Notetakers and summaries
- Sales coaching mid-call
- CRM updates with no typing
- Voice agents that answer
It joins as a named participant
The agent appears in the participant list under your product's name, and everyone can see it is there.
- A named tile in the call
- A joined notice in the chat
- Speaks, chats or stays silent
- Leaves on your rules
Priya
DanielEnterprise-level security, keep your data private.
The certificate, the audit status and the Trust Center are all on the security page.
Keep recordings with MeetStream, or write them straight to your own storage bucket.
Support on Slack, email and chat, with a response inside one working day.
Concurrent bots scale on their own. There is no cluster for you to run.
Got a question? We got the answer.
Common questions about platforms, models, voice control and pricing.
What is a meeting agent API?
An API that puts a software participant into a live video call and lets it work with what happens there: listen, transcribe, speak, post in chat and call tools.
How do I build an AI agent that responds during meetings?
Create an agent with one POST to /api/v1/mia, then pass its agent_config_id when you send a bot with create_bot. MeetStream connects the live audio for you, so there is no WebSocket server to run. The agent joins Zoom, Google Meet or Microsoft Teams as a named participant, hears the room with speaker names, and replies by voice, in the meeting chat or through a silent tool call. Realtime mode gives the lowest latency. Pipeline mode adds a wake phrase, so the agent answers only when someone addresses it.
How do I control what an AI meeting agent knows in each meeting?
Put named placeholders, written in double curly braces, in the agent's system prompt and first message, then pass the values in agent_config_params when you send the bot with create_bot. One saved agent then starts each meeting with the account notes, agenda or instructions you pass for that call. To change what it knows during the call, POST to /api/v1/bots/{bot_id}/update_mia with new variable values or a new system prompt. Live changes apply to that session only, so the saved agent stays the same.
Which models can a MeetStream agent use?
OpenAI, Anthropic, Google and xAI. Realtime mode uses one provider end to end. Pipeline mode lets you choose speech recognition, the model and speech separately.
How fast does the agent reply?
Realtime mode, with one provider end to end, has the lowest latency. Pipeline mode runs speech to text, the model and speech in turn, which trades some speed for choice.
Can the agent stay silent and only act?
Yes. In action mode the agent produces no voice and no chat. It only calls your tools, which suits CRM updates and ticket creation.
Which platforms does it support?
Zoom, Google Meet and Microsoft Teams from the same request. WhatsApp calls are in Beta.