AI Foundationsintermediate13 min · 10 steps
The Augmented LLM: Retrieval, Tools and Memory
How an LLM gets fresh knowledge, the ability to act and a memory, and the function-calling loop that makes tools work.
LLM calls1
context60tok
requesterrortool calldataresponse
metrics are illustrativeProblemStep 1 of 10
A bare LLM is frozen, passive and forgetful
A user asks: “Is my order 4471 shipped? If not, cancel it. And remember I prefer email updates.” A bare model can't do any of it.
Its knowledge stopped at training time, so it can't look up the order. It only produces text, so it can't act (cancel). Each conversation starts empty, so it can't remember the preference.
Watch outAsked anyway, the model will often invent a plausible status. Missing capabilities turn into hallucinations.
← Previous lessonRetrieval: Keyword, Semantic, Hybrid and HyDENext lesson →Agent Workflows I: Prompt Chaining and Routing
Lesson 28 of 34 in AI Foundations