Welcome to Retrieval-Augmented Agents, the sixth module. Your agent can loop, use tools, remember, and plan — but it still answers from a fixed pool of training knowledge that can’t see your documents, your latest data, or anything private. When it doesn’t know, it tends to guess confidently. This module fixes that by giving the agent retrieval: the ability to look things up in a knowledge base you control, and to ground its answers in what it finds.

You’ll start with why agents need retrieval — grounding, fresh and private knowledge, and fewer confident wrong answers. You’ll build a knowledge base: chunk documents into passages, embed them, and search by similarity. You’ll then expose retrieval as a tool so the agent decides when to look something up, mid-loop — agentic RAG. And you’ll add the discipline that makes retrieval trustworthy: grounding and citations — answer only from retrieved sources, cite them, and refuse honestly when the knowledge base comes up empty. The module ends by giving Atlas a destination knowledge base it searches and cites before it plans.

Every retrieval pattern here — the knowledge base, retrieval-as-a-tool, and grounded answers with a similarity floor — is real, runnable Python, verified end to end against the agent loop you’ve built. The knowledge base stays dependency-free (the keyword embedding from the memory module), with the production swap to real embeddings called out as you go. Start with Lesson 1 on why agents need retrieval.