Agentic Retrieval overview

Agent-driven iterative retrieval pipeline where an LLM agent decides what to search, evaluates results, and re-queries until sufficient context is gathered. The agent decomposes complex questions into sub-queries, selects which retrieval backends to query (vector store, knowledge graph, API, code search), evaluates if retrieved context answers the question, and iterates if not. Supports self-reflective retrieval (is this enough?), corrective retrieval (this answer seems wrong, let me re-search), and multi-source orchestration.