This section tests your understanding of LiveKit's integration ecosystem — RAG patterns, LangChain/LangGraph, provider selection, and the plugin architecture that makes the voice pipeline composable.
Test your knowledge
Question 1 of 5
A client wants their agent to answer questions from a knowledge base. What is the recommended RAG architecture with LiveKit?
A Implement RAG as a tool: when the agent needs information, it calls a search tool that queries a vector database (e.g., Pinecone, Weaviate, pgvector), retrieves relevant chunks, and returns them as context. The LLM then generates a response grounded in the retrieved documents. This keeps the system prompt focused on agent behavior while the knowledge base scales independently. B Load the knowledge base into the agent's system prompt using a sliding window approach. On each conversation turn, dynamically select the most relevant sections of the knowledge base to include in the prompt based on the user's latest question. This avoids the latency of a tool call and ensures the LLM always has context available without needing to request it. C Use LiveKit's built-in knowledge base plugin, which pre-indexes documents at agent startup and injects relevant context into the LLM prompt automatically based on STT transcripts. The plugin uses keyword matching (not vector search) for speed and runs locally in the agent process without external database dependencies. D Configure a RAG middleware in the pipeline that intercepts LLM requests. Before each LLM call, the middleware runs the user's latest utterance through a vector search, appends the top 5 results to the conversation history as system messages, then forwards to the LLM. This transparent approach requires no changes to agent code or tool definitions.