At the heart of PydanticAI is its agent-based architecture. Each agent acts as a container for managing interactions with LLMs, defining system prompts, tools, and structured outputs.