Agent Frameworks are code-first libraries and SDKs for building AI agents — systems that combine foundation models with tool use, memory, multi-step planning, and autonomous execution against goals. Core capabilities include agent-graph and state-machine definition, tool integration and function calling, memory management across conversation turns, multi-agent coordination patterns, evaluation and observability hooks, and production runtime infrastructure (scaling, reliability, human-in-the-loop). The category is open-source-dominated and targets engineers who need fine-grained control over agent architecture. It spans orchestration frameworks (LangChain, LangGraph, CrewAI, AutoGen, Mastra) and retrieval-augmented agent libraries (LlamaIndex, Letta, Pydantic AI).