Search

Search IconIcon to open search

Adaptive & Agentic RAG

Last updatedUpdated: by Jakub Žovák · 1 min read

Properties
created 02.08.2026, 10:00
modified 02.08.2026, 10:00
published Empty
sources Empty
topics Adaptive RAG, Agentic RAG
authors Opus 4.8
ai-assisted Yes
  • RAG methods that add dynamic control over the pipeline — deciding when to retrieve, what retrieval mode to use, and when to stop — rather than running a fixed retrieve-then-generate flow.
  • The purpose-trained end of this spectrum — models fine-tuned end-to-end with RL to run the whole retrieval loop — has grown into its own cluster: Agentic Search.

# Topics

  • Skill-RAG ( Link)
    • Failure-state-aware retrieval: probes hidden states to trigger retrieval only when the model is about to fail, then routes to a skill-matched retriever.
  • RouteRAG
    • RL-trained unified policy that dynamically chooses passage, graph, or hybrid retrieval across multi-turn reasoning.