Agent framework
Also known as: LLM framework
An agent framework is a library that packages the harness for you: the agent loop, tool calling, memory, state management.
- You define tools and logic; the framework runs the loop.
- Examples span LangChain/LangGraph, CrewAI, AutoGen, and vendor SDKs.
- Trade-off: convenience vs. control and abstraction cost.
When to use one: frameworks shine for standard patterns (tool loops, multi-step plans). If your harness is exotic, a thin custom loop may be simpler than fighting the abstraction.