Hierarchical agentic RAG

A multi-tiered structure where higher-level agents oversee lower-level ones. It is the orchestrator-workers pattern applied across levels, with added strategic decision-making.

The workflow: a top-tier agent receives the query and assesses its complexity, then makes a strategic decision about which subordinate agents or data sources to prioritize (some sources may be judged more reliable for a given domain). It delegates to lower-level agents specialized in particular retrieval methods, which work independently, and then it aggregates and synthesizes their results.

Strengths: strategic prioritization (the top agent can rank sources by reliability or relevance), scalability across tiers for complex queries, and enhanced decision-making through layered oversight.

Challenges: coordination complexity grows with the number of levels, and allocating tasks across tiers without creating bottlenecks is hard.

The survey's example is a financial analysis system where the top-tier agent prioritizes reliable financial databases, a mid-level agent retrieves real-time market data, and lower-level agents handle web searches and expert-opinion recommendations, with everything compiled at the top.