OpenAI has released version 0.19.0 of the OpenAI Agents Python SDK, introducing a significant new feature area: Programmatic Tool Calling for supported OpenAI Responses models.

  • Added `agents.tool.ProgrammaticToolCallingTool`, allowing models to generate JavaScript to coordinate eligible tools with support for per-tool allowed callers and structured outputs.
  • Introduced the public `agents.decorators` module and a shorter `@tool` alias, while enabling async callable objects for function tools.
  • SDK configuration now consistently accepts either typed settings objects or dictionaries across agents, runs, models, sessions, sandboxes, and voice pipelines.
  • Hardened error and diagnostic logging to prevent exposing raw sensitive payloads while preserving debugging context.
  • Improved compatibility with AnyLLM and LiteLLM, preserved session history across model retries, and added retries for WebSocket overloads.
  • Added `VercelCloudBucketMountStrategy` for mounted sessions, which excludes bucket contents from workspace persistence.

This release enables more complex agent orchestration through programmatic tool coordination while improving configuration flexibility and security in logging.