AI agent glossary
Agent SDK
A software library that packages the agent loop, tool handling, and common features so you can build agents without writing that plumbing yourself.
When you build an agent directly on a model API, you write the loop, the tool dispatch, error handling, and context management yourself. That is a great way to learn, but it gets repetitive. An agent SDK gives you those pieces ready-made.
Typical features include a built-in agent loop, simple ways to declare tools, support for MCP servers, subagents or handoffs, streaming, permission controls, and tracing. Well-known examples include the Claude Agent SDK and the OpenAI Agents SDK.
The tradeoff is control. An SDK makes the common case easy but adds its own concepts and defaults. Many builders write a minimal loop by hand first so they understand what the SDK is doing for them.
Learn more in the course
Related terms
Go from definitions to doing
Learn to use and build AI agents in 65 practical lessons.