> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.meetstream.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.meetstream.ai/_mcp/server.

# Three ways to integrate

> Choose how to integrate MeetStream: record and stream meetings with the Bot API, run your own agent in the meeting through a bridge server or the Agent Kit, or let MeetStream host the voice agent with the Agents API (MIA).

Every integration starts with a MeetStream bot in Zoom, Google Meet or Microsoft Teams. What you add to that bot decides the flow.

| Flow                                   | Use it when                                                                                             | Who runs the models                                            | Price per meeting hour                                             |
| -------------------------------------- | ------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------- | ------------------------------------------------------------------ |
| **1. Recording only** (Bot API)        | You want recordings, transcripts, participants, chat or live data. Nobody needs the bot to talk.        | Nobody, or your own server after the data arrives              | Bot price                                                          |
| **2a. Bridge server** (your own stack) | You want an agent that talks in the meeting, and you want to pick every model, including local ones.    | You: your own speech-to-text, LLM and voice, hosted or local   | Bot price only                                                     |
| **2b. Your own agent** (Agent Kit)     | You already run an agent (Claude Code, Codex, Hermes, OpenClaw and others) and want it to be the brain. | You: your agent on your own subscription. MIA handles speech.  | Bot price + \$0.25 agent infrastructure                            |
| **3. Managed agent** (Agents API, MIA) | You want a voice agent without running any servers.                                                     | MeetStream hosts the agent, with inbuilt keys or your own keys | Bot price + \$0.25 agent infrastructure + the models that you pick |

The bot price is $0.35 per hour ($0.25 per hour at volume). See [Voice agent pricing and keys](/guides/mia/pricing-and-keys) for the agent prices and keys.

## Which flow?

| Your question                                                                         | Flow   |
| ------------------------------------------------------------------------------------- | ------ |
| Do you only need data from the meeting (notes, transcript, recording, live captions)? | **1**  |
| Must the bot talk, and do you want to run your own models, including local models?    | **2a** |
| Must the bot talk, and do you want an agent that you already use to give the answers? | **2b** |
| Must the bot talk, and do you want no infrastructure work?                            | **3**  |

If you want an agent but are not sure which flow: run your own stack (flow 2a, any models, including local) or let MeetStream host it (flow 3).

---

## 1. Recording only: the Bot API

Use the Bot API for notetakers, meeting summaries, CRM updates, sales coaching and any product that reads the meeting. One call puts a bot into the meeting:

```bash
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": "https://meet.google.com/abc-defg-hij", "bot_name": "Notetaker", "video_required": false}'
```

Then decide when you need the data: after the call, or during the call.

| You need                                                                         | Use                                                                                                                                                                                                                                                                                                                                                                           |
| -------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **After the call** (post-call): recordings, a transcript, participants and chat  | [Webhooks](/guides/webhooks/webhooks-and-events) tell you when each item is ready. Then get the [recordings](/guides/transcription-recordings/retrieve-recordings), the [transcript](/guides/transcription-recordings/post-call-transcription) and the [participants and speaker timeline](/guides/features/participants-and-speaker-timeline).                               |
| **During the call** (live): a transcript, audio or meeting control as it happens | [Live transcription](/guides/transcription-recordings/live-transcription) to your webhook, [real-time audio](/guides/websockets/real-time-audio-streaming) over a WebSocket, the [meeting control socket](/guides/websockets/meeting-control-patterns), [chat and visuals](/guides/features/chat-and-visuals), and [raise and lower hand](/guides/features/raise-lower-hand). |

**Price.** Bot price. Add $0.10 per hour for a post-call transcript from the [MeetStream engine](/guides/transcription-recordings/providers/meetstream). With your own transcription provider key, that $0.10 is waived.

---

## 2. Agents in the meeting on your own stack

Start here when you want an agent in the meeting and you want to control the models. There are two setups.

### 2a. Bridge server

The bot opens two WebSockets to your server:

| WebSocket  | `create_bot` field                    | Direction         | What it carries                                                                     |
| ---------- | ------------------------------------- | ----------------- | ----------------------------------------------------------------------------------- |
| Live audio | `live_audio_required.websocket_url`   | MeetStream to you | Meeting audio, with the speaker of each frame                                       |
| Control    | `socket_connection_url.websocket_url` | Both ways         | Your commands: send audio, chat messages and images into the meeting, and interrupt |

Your server runs the speech-to-text, the LLM and the voice. Use any provider, hosted or local. For example, run Parakeet for local speech-to-text, a local LLM and a local text-to-speech model.

```bash
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": "https://meet.google.com/abc-defg-hij",
    "bot_name": "My Assistant",
    "live_audio_required": { "websocket_url": "wss://your-bridge.example.com/bridge/audio" },
    "socket_connection_url": { "websocket_url": "wss://your-bridge.example.com/bridge" }
  }'
```

**Price.** Bot price only. MeetStream charges no agent infrastructure and no model tokens. Your model providers, if any, bill you directly.

* [Bridge Server Architecture](/guides/websockets/bridge-server-architecture): the two endpoints, the session lifecycle and a full working skeleton.
* [Real-time Audio Streaming](/guides/websockets/real-time-audio-streaming): the audio frame format and how to decode it.
* [Meeting Control and Command Patterns](/guides/websockets/meeting-control-patterns): the commands that you send back.
* Starter repo: [meetstream-ai/meetstream-agent](https://github.com/meetstream-ai/meetstream-agent), an open-source FastAPI bridge. It uses OpenAI Realtime by default. Replace that part with your own models, including local models.

### 2b. Your own agent as the brain

Use the Agent Kit when you already run an agent, for example Claude Code, Codex, Hermes or OpenClaw, on your own computer or your own subscription. MIA joins the call, listens and speaks. Your agent writes the answers. Set `model.provider` to `"local_agent"` and connect your agent with the open-source [MeetStream Agent Kit](https://github.com/meetstream-ai/agent-kit).

**Price.** Bot price + \$0.25 per hour for the agent infrastructure. MeetStream charges no model tokens for your agent's reasoning.

> **Warning**
>
> **Use this setup with care.** What people say and type in the meeting reaches an agent that has access to your computer and your terminal. Someone in the meeting can try to give your agent instructions (prompt injection). Keep the access level as low as possible: `readonly` is the default. Your computer, your agent and the bridge must stay on for the whole call.

See [MIA Screen Sharing and Custom Agents](/guides/features/mia-screen-sharing-and-custom-agents#4-your-agent-bring-claude-code-codex-and-more) for the setup, the activation code and the access levels.

---

## 3. Managed agent infrastructure: the Agents API (MIA)

Use the Agents API (API name: MIA) when you want a voice agent and no infrastructure work. MeetStream hosts the agent: it listens, speaks, chats and calls tools in the meeting. You save an agent configuration with `POST /api/v1/mia`, then send a bot with its `agent_config_id`.

For each pipeline section (`transcriber`, `model`, `voice`), pick the key:

* **MeetStream's inbuilt keys.** Set `"key_source": "meetstream"` (pipeline mode only). The inbuilt keys cover OpenAI and Deepgram today. You need no provider account. If the API answers `MeetStream-managed keys are not available for <section>. Use your own key.`, inbuilt keys are not on for your account yet: use your own key for that section. Do not retry the same request in a loop.
* **Your own keys.** Add your provider keys (OpenAI, Anthropic, Deepgram, ElevenLabs and others) under **Integrations** and pick any supported model.

**Price.** Bot price + \$0.25 per hour for the agent infrastructure + the models that you pick. With inbuilt keys, the dashboard agent editor shows the per-hour price of your models before the bot runs. With your own keys, your provider bills you for model usage.

> **Note**
>
> A MIA bot takes only `agent_config_id`. Do not add `live_audio_required` or `socket_connection_url` to it: the agent then stays silent. Flows 2a and 3 use separate bots.

* [Voice agent pricing and keys](/guides/mia/pricing-and-keys): prices, `key_source` and a copy-paste example for each setup.
* [What is MIA?](/guides/mia/what-is-mia): modes, providers and features.
* [Create an Agent](/guides/mia/create-an-agent): set up an agent in the dashboard.
* [MIA API Guide](/guides/mia/mia-api-guide): create and update agents with the API.