Agentic RAG
Retrieval-augmented generation where the system decides when, what, and how to retrieve — routing between sources, checking whether the results suffice, and re-retrieving when they don't — instead of running one fixed lookup per question.
Production AI is a system: context, tools, permissions, traces, evals, and feedback loops around the model.
What it is
Classic RAG retrieves top-k chunks for every question, whether the question needs them or not. Agentic RAG gives the retrieval loop judgment: decide whether a turn needs retrieval at all, choose the source and the query, evaluate whether the results actually answer the question, and go back for a wider or different pass when they don't.
Why it matters
Fixed retrieval fails at both ends. It burns tokens retrieving for questions that needed none, and it answers thinly when the first lookup missed. Spending retrieval effort where the question demands it improves both cost and grounding quality.
How we build it
A deterministic gate decides whether a turn retrieves. Hybrid search — vector similarity fused with full-text ranking through reciprocal rank fusion — runs the lookup. A post-answer sufficiency check triggers one bounded re-retrieval when grounding is thin. Keeping the loop bounded keeps latency and cost predictable; an unbounded retrieval loop is its own failure mode.
Related resources
The pattern where an AI agent retrieves relevant context from your data before generating an answer — instead of relying only on what the model learned during training.
The search behind your AI combines two ways of looking — by exact words and by meaning — then gives the best candidates a careful second read, so answers that exist in your documents actually get found.
A deliberate, tested decision about how many search results get an expensive careful second read, and with which model — tuned per workflow instead of copied from a tutorial.
How an AI agent finds the right document, chunk, or row to ground its answer in — and why the part that matters is the pipeline around the database, not the database itself.