Choose a Retrieval Stack for a Growing Agent Knowledge Base

ROLE Act as a search infrastructure advisor. OBJECTIVE Recommend a retrieval stack for [KNOWLEDGE COLLECTION]. CONTEXT Representative queries: [QUERY EXAMPLES]. Collection size and update frequency: [COLLECTION PROFILE]. Permission requirements: [ACCESS REQUIREMENTS]. Existing infrastructure: [CURRENT INFRASTRUCTURE]. INSTRUCTIONS Compare keyword search, existing database search, vector retrieval, and hybrid search. Distinguish storage, indexing, retrieval, reranking, and answer generation rather than treating them as one capability. Evaluate exact identifiers, synonyms, freshness, metadata filtering, deletion propagation, permission enforcement, and operational complexity. Identify cases where adding embeddings would not solve the actual problem. Propose a retrieval-only trial with known relevant documents and difficult negative queries. Require access filtering before restricted content reaches the model. For named products, verify capabilities against current official documentation when available. OUTPUT FORMAT Provide an architecture comparison, minimum viable recommendation, trial dataset outline, and migration triggers that justify a more complex stack. QUALITY CHECK Flag assumptions and avoid fabricated statistics. Separate measured retrieval quality from answer fluency, and explain which requirements remain unverified.