Thanks — regarding whether an agent can use multiple models and fail over if the primary fails: Yes. LiveKit Inference supports multi-model/fallback behavior for LLM, STT, and TTS components. You can configure explicit fallback models (for example

fallback_models

in the OpenRouter LLM plugin) so if the primary model or provider is unavailable, the agent automatically tries the alternatives. Provider-level fallback is also supported (e.g., TTS voice clones will route to another provider that hosts the voice). The system’s modular pipeline (VAD → STT → LLM → TTS) and built-in reliability patterns (retries, timeouts, multiple workers) enable graceful degradation and automatic switching. If you’d like, tell me which component (LLM, STT, or TTS) and which providers/models you plan to use and I’ll suggest a specific fallback configuration. Sources: Introducing LiveKit Inference: A unified model interface for voice AI | Liv... | OpenRouter LLM plugin guide | LiveKit Documentation | Custom voices | LiveKit Documentation | How does end-of-utterance detection work in conversations? | Sequential Pipeline Architecture for Voice Agents | LiveKit, with interactive elements