Home/Use Cases/Real-time meeting agents
// Real-time agents

Build real-time meeting agents that hear, think, and respond

Real-time meeting agents need live audio in and actions out while people are still talking. MeetStream streams meeting media over WebSocket at around 200ms and lets your agent act in the room.

Roughly 200ms audioOne stream per speakerChat, images and GIFsZoom, Meet, Teams
SIGN UPREAD THE DOCS
Overview

What a live agent actually needs

An agent that responds after the call is a notetaker. Three things make one that participates.

SIGN UPREAD THE DOCS
One request returns an MP4 recording, per-participant audio and video, and a transcript
Send a meeting link, and when the call ends you get all three.
Low-latency audio

So it hears the conversation as it happens.

Live transcripts

Sent to your webhook, so its model has text to work on.

A way to act inside

So its output reaches the participants.

All three from one bot

Your agent architecture stays a single loop.

Real-time audio carries the speaker on every frame.

What you get

Live audio, one stream per speaker

Audio and video stream over a secure WebSocket at around 200ms, one stream per participant.

  • It knows who is talking
    Without running source separation of its own.
  • It can wait to be addressed
    And react only when the host speaks to it.
  • Per-person context
    Held through the whole conversation.
Meeting linkcreate_botone requestBot in the callrecords and streamsWebhookbot.doneYour agentacts during the call
The same five steps on every platform. Only the last box is yours.

What comes back from every call

Recording API
Composed MP4, plus one audio and video stream per participant.
Transcription API
Speaker-labelled transcripts from six engines, live or after the call.
Webhooks
Every event your handler will see, with delivery rules.

Feed the streams to your own streaming speech-to-text, or take the real-time transcripts with speaker labels sent to your webhook.

LIVE PRODUCTConfiguring an AI meeting agent, choosing pipeline or realtime mode and a voice, in the MeetStream dashboard
Configure a real-time agent's mode and voice, then send it into the call.
In your product

Act inside the meeting

MeetStream bots can speak, read the meeting chat, send chat messages, and display images and GIFs.

  • Answer in chat
    With a link to the doc.
  • Post the chart
    When someone asks about last quarter.
  • Read what people type
    Participants often send requests to a bot in writing rather than out loud.
MeetingBotin the call/audio socketPCM16 and speakerYour serverSTT, agent, coachingControl socketsendaudio, sendchat
Audio out to your server, control back into the room.

Put an agent in the room with one call

Deploying the agent is one request with the meeting URL, and the bot joins within seconds.

No paid meeting plan is required, and host consent for recording is asked only on Zoom. Webhook lifecycle events let your backend manage the agent session from join to teardown.

Design

Designing the response loop

A responsive agent is a latency budget problem.

  • About a second to answer
    Roughly 200 ms of transport leaves that for speech-to-text and a model pass.
  • Acknowledge, then answer
    Reply in chat at once, then follow with the full answer.
  • Trigger on the voices you care about
    Per-participant streams cut inference cost, on speakers or wake phrases.

Audio in, audio out

48 kHzsigned 16-bit PCM, mono
One frameper WebSocket message, speaker length-prefixed at the front
Control channelaudio to play, chat text to post, and an interrupt
MIAa managed version of the same loop

The real-time audio page has the byte layout and a decoder. A MIA voice agent handles listening, reasoning, tool calls and speech inside MeetStream.

MEETSTREAM DASHBOARDThe MeetStream API Playground with a bot request and its live JSON request body
Configure a bot in the API Playground. Watch the request build itself, field by field.
Security

Enterprise-level security, keep your data private.

The certificate, the audit status and the Trust Center are all on the security page.

SIGN UPREAD THE DOCS
Robust and reliable

MeetStream is ISO 27001 certified. SOC 2 Type 2 is under audit rather than complete.

GDPR compliant

MeetStream is GDPR compliant, and bots run in the United States by default.

A visible guest, always

The bot appears in the participant list under the name you give it.

SOC 2Type 2 (Under Audit)
ISO 27001Certified
GDPRCompliant

Agent limits and defaults

4 hoursdefault session, five when you raise it
No capconcurrent bots, and no limit to raise
30 daysstorage, or your own S3 bucket
$0.35per bot-hour, and $0.45 with transcription
United StatesThe default regionEU, Australia, JapanAnd the Middle EastOther regionsAvailable on request
Bots run in the United States by default.

More than 30 AI products run on this infrastructure in production, live agent workloads included.

FAQ

Got a question? We got the answer.

Common questions about live agents. Latency, how an agent responds, speech-to-text and platform coverage.

SIGN UPREAD THE DOCS
What latency should I actually expect?

Media streams over secure WebSocket at around 200ms. Your end-to-end response time then rests on your own transcription and model stack. That is why per-participant streams and trigger-based processing matter. They keep you inside a conversational window.

How does the agent respond in the meeting?

Through the meeting itself. It can speak with sendaudio over the control channel, send chat messages, and display images or animated GIFs that all participants see. It can also read the chat, so people can address it in writing.

Do I have to run my own speech-to-text?

No. Real-time transcripts with speaker labels are sent via webhook during the call. In-house transcription is an add-on. If you want raw control, take the WebSocket audio streams and bring your own models.

Which platforms can the agent join?

Zoom, Google Meet, and Microsoft Teams, all through the same API and the same integration code. No host permission or paid meeting plan is required on the meeting being joined.

An agent that takes part in the call.

Live audio at around 200ms, one stream per speaker, and chat, images and speech back into the room.

SIGN UPTALK TO US
Roughly 200ms audio
One stream per speaker
Live transcripts
Chat and images
Voice agents

Keep exploring

AI meeting notetakerUse caseConversation intelligenceUse caseWebSocket streamingGlossaryReal-time transcriptionGlossaryMeeting botGlossaryMeeting Bot APIProductZoomProductBuild a zoom meeting botBlogAudio transcription bot for google meetBlogVoice agentsProductReal-time audioProductBridge serverGlossary

Build it on MeetStream

One API call, $5 of free credit, no card. One API joins, records, streams, and transcribes across Zoom, Google Meet, and Microsoft Teams.

SIGN UPTALK TO A HUMAN