A library inside your application
The Agents SDK provides a runner for an agent loop and handoffs. You run and deploy that code. This is a different operational model from asking the Agents API to run a managed harness through a durable session.
Control comes with ownership
Your application integrates storage, approval decisions, runtime services, and deployment. SDK session abstractions manage conversation history, but their identifiers and lifecycle should not be confused with managed API session resources.
Compare integration boundaries
Existing applications with substantial orchestration or custom runtime requirements may prefer a library. Applications seeking a managed harness can evaluate the Agents API. Compare the actual workflow, tools, and failure handling before choosing.
Read the official reference
Check the source for current API fields, account requirements, and service limits.
OpenAI: agents