How It Works — LLM

Language models are stateless and forgetful, but applications need chat history, tools, and up-to-date knowledge. These flows show how each of those layers is bolted on top of a plain prompt.

LLM [LLM] {1}

From a bare prompt to a full agent loop.

Prompt and Completion [Prompt] {1}

Chat with History [Chat History] {1}

System Prompt and Roles [System Prompt] {1}

Tool Use / Function Calling [Tool Use] {1}

RAG — Retrieval-Augmented Generation [RAG] {1}

Agent Loop (ReAct) [Agent] {1}

MCP — Model Context Protocol [MCP] {1}