Agentic retrieval (concept)¶
Agentic retrieval is iterative, tool-driven retrieval. A large language model (LLM) agent plans steps, issues search tool calls, fuses candidates, and ranks documents until it has enough context. One-pass retrieval sends a single static query through dense or hybrid search and returns chunk-level hits.
NeMo Retriever Library includes a first-class agentic query path. retriever query --agentic, POST /v1/query with agentic=true, and the agentic_query Model Context Protocol (MCP) tool run a Reason and Act (ReAct) loop over the same LanceDB table that one-pass retrieval uses. You do not have to implement the agent loop in application code.
The agentic path ranks documents rather than chunks, and returns the same hit fields as one-pass retrieval for each selected document. Local CLI and harness runs default to an in-process vLLM agent LLM. Retriever Service requires a remote OpenAI-compatible chat-completions endpoint. A self-hosted vLLM-backed NIM must enable automatic tool choice and a tool-call parser. Helm answer_llm does not turn those options on by default.
For commands, service configuration, request and response contracts, and failure behavior, refer to Workflow: Agentic retrieval.