# What is a bridge server? > In MeetStream's real-time audio model, a bridge server is a small service you run between a MeetStream bot and your own AI session. The bot sends it. Source: https://meetstream.ai/glossary/bridge-server *By Sidhdharth Sivasubramanian, Co-founder and CEO, MeetStream* A bridge server is a small service you run between a meeting bot and your own AI session. The bot sends it live meeting audio on one WebSocket and control commands on another. The bridge decodes, resamples, filters and forwards that audio. Your own voice stack can then take part in a real meeting without joining it. - **Bot**: in the meeting - **/bridge/audio**: binary PCM - **Bridge**: decode, resample, filter - **/bridge**: JSON control - **AI session**: your stack *Audio one way, control both ways, one live session per bot.* ## Why two WebSockets The two carry very different traffic, so MeetStream keeps them apart. - **Audio socket**: Hundreds of frames a minute Binary frames, one way Tagged with the speaker ID and name 48 kHz meeting audio - **Control socket**: The occasional short message Commands both ways Speak, send chat, send an image Interrupt, and relay a user message *One steady flow, one short message at a time.* An interrupt lands the instant you send it, instead of queueing behind audio. That is the difference between stopping on cue and talking over people. The [bridge server guide](https://docs.meetstream.ai/guides/websockets/bridge-server-architecture) names each one. ## What the bridge does Four jobs, and the guide ships a working Python skeleton with all four. - **Hold one session**: Both sockets share it - **Resample**: 48 kHz to your model rate - **Skip its own voice**: Matched by speaker name - **Pump the reply back**: Audio, text and interrupts *One session, two sample rates, one voice skipped, one route back.* You give each bot the two socket URLs at create time. The rest is [WebSocket streaming](https://meetstream.ai/glossary/websocket-streaming). ## When you do not need one For a managed version of the same loop, [MIA](https://meetstream.ai/glossary/mia) runs it end to end inside MeetStream. Teams who already own a voice stack keep the bridge and take only the in-meeting part. ## Frequently asked questions ### Do I need a bridge server for a voice agent? A bridge is optional, because MIA runs the loop for you. Teams who already own a voice stack use a bridge to take just the in-meeting part from MeetStream. ### What audio format arrives? 48 kHz mono PCM16, little-endian. Every frame carries the speaker ID and name, each with a length prefix. ### What sample rate should I send back? 48 kHz PCM16, base64 encoded, on the control socket. The guide shows a 24 kHz model rate being resampled up to it. ### How does the bridge avoid hearing itself? It keeps only the frames from human speakers, matching the bot by its display name. You can extend the list of names it skips. ## Keep exploring - [Voice agents](https://meetstream.ai/voice-agents): Product - [MIA](https://meetstream.ai/glossary/mia): Glossary - [WebSocket streaming](https://meetstream.ai/glossary/websocket-streaming): Glossary - [Real-time audio API](https://meetstream.ai/real-time-audio-transcript-api): Product - [Per-participant audio](https://meetstream.ai/glossary/per-participant-audio): Glossary - [Meeting bot](https://meetstream.ai/glossary/meeting-bot): Glossary ## 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