Thanks — this sounds like an interesting project. I’m responding to your request about integrating a LiveKit AI agent into Microsoft Teams for real-time voice + AI participation in Teams calls. A few clarifying questions so I can give targeted guidance and recommend the best approach: 1. Do you control the Teams tenant (are you an admin) or are you building for external customers? Tenant/admin access changes what integration options are available. 2. Do you want the AI agent to join existing Teams calls as a participant (call-in) or sit in a parallel LiveKit session that’s bridged to Teams (gateway/relay)? 3. Is the AI agent expected to: ◦ speak and listen in real time (low latency voice), ◦ transcribe only, or ◦ perform other actions like sending messages, analyzing sentiment, or controlling call flow? 1. Do you already have an AI model or service you’re using for the agent (e.g., custom model, OpenAI, etc.), or do you want recommendations on that piece as well? 2. Do you have constraints on infrastructure (self-hosted LiveKit vs LiveKit cloud) or compliance requirements (recording, data residency, encryption)? 3. What developer environment/language are you using (Node, Python, .NET)? Any existing LiveKit or Teams code you’ve started with? High-level options I can walk you through once you answer the above: • Join-as-a-participant: Provision a bot/meeting participant that joins Teams calls (typically via Microsoft Graph Call Records / Bot Framework / Direct Routing) and have the bot bridge audio to a LiveKit room where your AI agent processes audio and returns synthesized audio. • SIP/Media Gateway: Use a SIP or media gateway to bridge Teams PSTN/media to a LiveKit room so the AI agent can listen/speak via LiveKit without a native Teams bot. • Parallel session + user workflow: Keep Teams call native for human participants and create a LiveKit session that mirrors audio (via client-side forwarding or server-side media relay) so the agent can participate without being a Teams participant. If you tell me which of the clarifying points above match your scenario, I’ll provide concrete architecture recommendations, required APIs/permissions, and an implementation plan with next steps and example code snippets.