OpenAI Agents SDK

license

MIT(SDK)

source

public

core features withheld

no — one root MIT LICENSE across the 1,684-path tree at f007b3f with no ee/, enterprise/, commercial/ or proprietary/ path

commercial

OpenAI-API

The SDK is MIT licensed and its single package builds from the public repository, with no enterprise directory or paid edition. It runs against other providers' models as well as OpenAI's, so the paid API it most often calls is a separate service rather than a part kept out of the code.

• https://pypi.org/pypi/openai-agents/json recorded 2026-10-07

  PyPI metadata for openai-agents 0.23.1: license expression MIT, Repository https://github.com/openai/openai-agents-python, one pure-Python wheel and one sdist.

• https://raw.githubusercontent.com/openai/openai-agents-python/main/LICENSE recorded 2026-10-07

  The root LICENSE is the MIT text: "MIT License" and "Copyright (c) 2025 OpenAI".

• https://raw.githubusercontent.com/openai/openai-agents-python/main/pyproject.toml recorded 2026-10-07

  The package manifest builds the openai-agents distribution from the repository with license = "MIT"; its required dependencies are open-source libraries (openai, pydantic, griffelib, mcp and others).

• https://raw.githubusercontent.com/openai/openai-agents-python/main/README.md recorded 2026-10-07

  The README says the SDK is "provider-agnostic, supporting the OpenAI Responses and Chat Completions APIs, as well as 100+ other LLMs" and closes "We're committed to continuing to build the Agents SDK as an open source framework".

Adoption is read from PyPI downloads of openai-agents over the trailing month, the package the SDK installs from. Some of that volume comes from CI runs and from other packages that depend on it, so it counts installs rather than developers, and the JavaScript version on npm is not included.

• https://pypistats.org/api/packages/openai-agents/recent recorded 2026-10-07

  12,449,784 downloads in the trailing 30 days for openai-agents (pypistats last_month)

The OpenAI Agents SDK covers multi-agent handoffs, guardrails, sessions, tracing and sandboxed long-running agents with a deliberately small set of primitives. It sits a step below LangGraph, whose graph-structured runs give finer control over flow and checkpoint their state to resume where they stopped.

• https://openai.github.io/openai-agents-python/ recorded 2026-10-07

  The documentation home says "The Agents SDK has a very small set of primitives" and lists sandbox agents with "resumable sandbox sessions", guardrails run "in parallel with agent execution", function tools and MCP server tool calling.

• https://raw.githubusercontent.com/openai/openai-agents-python/main/README.md recorded 2026-10-07

  The README lists agents, sandbox agents, realtime and voice agents, agents as tools and handoffs, tools (functions, MCP, hosted tools), guardrails, human in the loop, sessions and tracing as core concepts; no benchmark figure.