Agentic Retrieval Overview
Agentic retrieval transforms complex user questions into focused subqueries, executes them in parallel, and returns a comprehensive, agent-optimized response.Key Features
- LLM Query Planning: Automatically decomposes complex questions
- Parallel Execution: Runs multiple searches simultaneously
- Multi-Source: Queries indexed and remote knowledge sources
- Semantic Reranking: Promotes most relevant results
- Agent-Optimized Response: Structured output for downstream consumption
Architecture
Knowledge Base orchestrates the entire pipeline, connecting to LLMs for planning and knowledge sources for retrieval.When to Use
- Complex multi-part questions
- Agent-to-agent workflows
- RAG applications requiring high-quality grounding
- Multi-source data scenarios