Thanks — you’re asking about the disadvantages of having multiple agents in a room. Below are the key downsides and what to watch for. • Unintended speech processing — agents don’t process other agents’ audio unless you enable it (

RoomInputOptions.allow_agent_input=True

), so inter-agent communication may not work by default. • Audio interference and crosstalk — multiple agents speaking or reacting at once can produce overlapping replies or chaotic conversations unless you coordinate turn-taking or implement handoff logic. • Handoff latency — transferring control between agents can introduce dead air or delays if transitions aren’t handled (filler audio and streaming optimizations help but must be implemented). • State and context complexity — keeping conversation context synchronized across agents requires explicit context passing (e.g.,

chat_ctx

) or you risk repetition and lost context. • Tool/permission conflicts — overlapping tool sets or publishing permissions (publish/subscribe) can cause race conditions unless permissions are managed per agent. • Resource overhead — each agent may use STT/TTS/VAD and consumes CPU/RAM; multiple agents increase infrastructure cost and load. • Debugging and observability challenges — tracing issues is harder when multiple agents interact; robust logging and metrics are needed. • Scalability constraints — job distribution assigns jobs to one agent server; uncoordinated multi-agent deployments can lead to inefficient resource use. If you tell me your use case (number of agents, expected traffic, whether agents need to hear each other, and if you plan handoffs), I can recommend an architecture and specific mitigations. Sources: The Handoff Pattern for Voice Agents That Replaces IVR Menus | LiveKit | The Human-in-the-Loop (HITL) Pattern for Voice Agents | LiveKit | Server options | LiveKit Documentation | Self-hosted deployments | LiveKit Documentation | Avoiding crosstalk when multiple developers share a LiveKit project | LiveK..., with interactive elements