ComponentsBuild with this
rag-agenticmainstream
Agentic RAG
LLM decides when and what to retrieve. Self-directed search.
Overview
The LLM acts as an agent that decides whether retrieval is needed, formulates its own search queries, and iteratively retrieves until it has enough context. Handles complex, open-ended questions that require adaptive search strategies.
Architecture
Interactive walkthrough01
Agent Loop
ReAct Reasonerthought generation
Tool Selectorchooses next action
Observationfeeds back into loop
Loop Controllermax-iteration guard
02
Tool Registry
vector_searchsemantic recall
keyword_filtermetadata narrowing
summarizecontext compression
03
Generation
Evidence Assemblytool outputs merged
Final AnswerLLM synthesis
01
Agent Loop
ReAct Reasonerthought generation
Tool Selectorchooses next action
Observationfeeds back into loop
Loop Controllermax-iteration guard
02
Tool Registry
vector_searchsemantic recall
keyword_filtermetadata narrowing
summarizecontext compression
03
Generation
Evidence Assemblytool outputs merged
Final AnswerLLM synthesis
Summarized pipeline view. For the full interactive, scroll-driven walkthrough with clickable stages → Pipeline detail
When to use
Use when
- query complexity == high
- multi-step research workflows
- queries are unpredictable in nature
- user tolerance for higher latency
Avoid when
- latency < 1s required
- cost sensitivity is high
- simple factual lookups dominate
- deterministic retrieval is required for audit
Compatible vector databases
PineconeQdrantWeaviateChroma
Compatible frameworks
langchainllamaindexraw python
#agentic#tool-use#iterative-retrieval#react#self-directed
Ready to build with Agentic RAG?
Walk through the wizard to generate a complete, parameterized pipeline.