// reference

Agentic AI & RAG on one page

Every concept, framework, library, and design pattern from the 2026 agentic-AI stack — RAG over structured and unstructured data, embeddings, vector indexes, the four core agent components, memory management, orchestration frameworks, token economics, evaluation, guardrails, and the LoRA / DAPT / RAG fine-tuning decision. Each topic gets one tools note (what to reach for) and one build note (how to actually do it).

RAG and agents in one sentence each. RAG = embed your data once, embed every user query at runtime, retrieve the top-k closest chunks, hand them to an LLM as context. Agent = an LLM + a state + tools + an orchestrator that loops "think → act → observe" until the task is done. Agentic RAG = the agent decides when, what, and how to retrieve — often in multiple rounds.