Design the knowledge layer for this use case. Inventory the authoritative sources, file types, databases, APIs, update frequency, access rules, retention requirements, expected questions, and citation needs.

Compare direct file access, SQL, keyword search, metadata filtering, vector retrieval, RAG, APIs, and MCP resources. Explain where structured querying is more reliable than semantic retrieval and where both should be combined.

Recommend an ingestion, indexing, retrieval, freshness, deletion, and access-control design. Include source identity, chunking only when appropriate, metadata, versioning, duplicate handling, citations, observability, cost, and tests for missing or stale material.

Do not create infrastructure yet. Distinguish verified requirements, assumptions, and questions requiring a human decision.