# What is MIA?

> MIA, the MeetStream Infrastructure Agent, is the agent layer built into MeetStream that turns a meeting bot into a voice agent: it processes the live.

Source: https://meetstream.ai/glossary/mia

*By Sidhdharth Sivasubramanian, Co-founder and CEO, MeetStream*

MIA, the MeetStream Infrastructure Agent, is the agent layer that turns a [meeting bot](https://meetstream.ai/glossary/meeting-bot) into a voice agent.

MIA takes the live meeting audio through an AI pipeline you configure.

It answers by voice, in chat, or with silent tool actions.

It works inside Zoom, Google Meet and Microsoft Teams.

- **Bot**: per-speaker audio
- **MIA**: STT, LLM, TTS or realtime
- **Tools**: MCP, functions
- **Response**: voice, chat, action

*MIA sits on the bot: it hears, reasons, calls tools and answers inside the meeting.*

## How MIA works

An agent is a saved configuration you attach to any bot by its ID.

The config holds a mode, providers, a system prompt, a response modality and tools.

The bot sends per-speaker audio to the agent in under 100 milliseconds.

The agent reasons, then speaks back through the same bot.

You create it with the [create agent config endpoint](https://docs.meetstream.ai/api-reference/api-endpoints/mia/create-agent-config).

## Realtime mode and pipeline mode

Two modes, and the choice is speed against control.

- **Realtime**: One speech-to-speech model Around 300 milliseconds end to end OpenAI Realtime, gpt-realtime-mini by default Google Gemini native audio, and xAI
- **Pipeline**: Three parts you pick yourself Speech to text, a language model, a voice Around 600 to 900 milliseconds Wake words and speaker-aware prompts

*One model end to end, or three parts you pick yourself.*

Pipeline mode also adds adaptive interruption handling.

## Tools and responses

Agents call MCP servers and custom HTTPS functions during the conversation.

- Response mode
- What the agent does
- Where it shows up
- **Voice**: Speaks the answer aloud In the meeting audio
- **Chat**: Posts the answer In the meeting chat
- **Action**: Runs the tool quietly In your own systems

*The response mode decides what happens with a tool result.*

Interruptions, turn detection, echo suppression and Anam avatars are settings, not code.

The [voice agents page](https://meetstream.ai/voice-agents) has examples.

## Frequently asked questions

### What does MIA stand for?

MeetStream Infrastructure Agent.

### Do I need my own model keys?

Yes, MIA is bring-your-own-key. You store provider keys in the dashboard, and MIA reads them when a session starts.

### Which platforms does MIA work on?

Zoom, Google Meet and Microsoft Teams, through the same bot that records.

### Can MIA run without speaking?

Yes. Chat mode responds in the meeting chat and action mode only calls tools.

## Keep exploring

- [Voice agents](https://meetstream.ai/voice-agents): Product
- [Bridge server](https://meetstream.ai/glossary/bridge-server): Glossary
- [Real-time meeting agents](https://meetstream.ai/use-cases/real-time-meeting-agents): Use case
- [WebSocket streaming](https://meetstream.ai/glossary/websocket-streaming): Glossary
- [Meeting bot](https://meetstream.ai/glossary/meeting-bot): Glossary
- [Transcription](https://meetstream.ai/integrations/transcription): Integration

## Put the vocabulary to work

One API to join, record, stream and transcribe meetings across Zoom, Google Meet and Microsoft Teams.

---

Full site index for agents: https://meetstream.ai/llms.txt
API contract (OpenAPI 3.1): https://meetstream.ai/openapi.json
MCP discovery: https://meetstream.ai/.well-known/mcp