How do you implement multi-source RAG combining different knowledge bases?

Multi-source RAG queries multiple knowledge bases and synthesizes results into a unified answer. Implementation involves defining data sources including internal documentation, product databases, CRM data, and external knowledge. A query router uses the query intent and metadata to determine which sources to query, or queries all relevant sources in parallel. Each source may use different retrieval strategies optimized for its data characteristics. Results from multiple sources are merged using reciprocal rank fusion or a learned fusion model. The combined context is provided to the LLM with source attribution for each chunk. Handle conflicts between sources by including source authority metadata and instructing the LLM on precedence rules. For example, product specifications override marketing materials, and recent updates override older documentation. LlamaIndex routers and LangChain agents provide frameworks for implementing multi-source RAG with configurable routing logic.