Thanks — I hear that TTS works in dev but in prod with those packages the agent starts but never speaks (you see “tts never arrived”). I can help troubleshoot — a few likely causes and quick checks plus one question to narrow this down. Quick checks (please verify in your prod environment) • Are provider API keys present in prod environment variables (e.g., CARTESIA_API_KEY, DEEPGRAM_API_KEY, ELEVENLABS_API_KEY) when you run agent plugins? Plugin-based integrations require those keys. • If you’re using LiveKit Inference string descriptors (e.g., "cartesia/.."), ensure you’re not expecting separate API keys — credentials are managed by LiveKit Cloud in that mode. Confirm which mode you use (plugin vs string descriptor). • Check agent logs in prod for TTS-related errors (authentication, network timeouts, or “no audio generated” messages). Look for lines around when session.say() or TTS synthesis is invoked. • Confirm outbound network egress from prod allows calls to provider endpoints (Deepgram/ElevenLabs/Google). Network restrictions often block TTS calls in prod. • Verify the same package versions/config (you listed 1.4.4). If dev and prod run identical versions, focus on environment/secret/network differences. • If you deploy via LiveKit CLI, ensure secrets were applied with

lk agent update-secrets

(or equivalent) so prod has the keys. One targeted question to narrow scope • Do you run TTS via plugin-based providers (requiring API keys) or via LiveKit Inference string descriptors? Send: • a recent excerpt of the agent logs showing the TTS attempt (error lines around the time the agent should speak) • how you configure TTS in AgentSession (code snippet) and whether API keys are set in prod I’ll use that to give exact next steps. , with interactive elements