LlamaIndex is a suitable framework for implementing Retrieval-Augmented Generation (RAG) in voice AI agents. The LiveKit Agents framework provides examples of using LlamaIndex for RAG, specifically in the

examples/voice_agents/llamaindex-rag

directory. There are three approaches available: • chat_engine.py: A straightforward solution using LlamaIndex's

as_chat_engine

, but it lacks function calling support. • query_engine.py: Offers a balance of flexibility and control, enabling function calling and custom behaviors without excessive complexity. This is the recommended approach. • retrieval.py: Manually injects retrieved context into the system prompt using LlamaIndex's retriever. If you need further details or guidance on implementing these approaches, feel free to ask! Sources: Handling RAG Delays in Voice Agents | LiveKit Docs | Tree View: | dse-workbench