Core Concepts and Architecture

Agent Architecture

A typical LLM agent follows this flow:

  1. Observe - Receive input from environment
  2. Think - Process with LLM reasoning
  3. Act - Execute tools or respond
  4. Reflect - Learn from outcomes

The ReAct Pattern

ReAct (Reasoning + Acting) is a popular agent pattern:

Thought: I need to find the weather in Seoul
Action: search_weather("Seoul")
Observation: Current temperature is 5°C
Thought: I have the information needed
Answer: The weather in Seoul is 5°C

Memory Types

  • Short-term: Current conversation context
  • Long-term: Persistent knowledge storage
  • Episodic: Past interaction histories
← Introduction to LLM Agents