OpenAI Agents SDK
OpenAI
Production successor to Swarm for agentic OpenAI workflows with broader model support.
Curated by Tiberiu ArvaVerified
Strengths
- Hosted tracing support
- Production successor to Swarm
- Handoffs and tool orchestration
Practitioner note
Fast path for OpenAI-centric agent builds, though portability and provider concentration should be weighed before making it a platform default.
Warning
Governance posture
Reviewed 2026-05-26Library is self-hosted, but built-in tracing sends run data to OpenAI's platform unless disabled.
Sourcepip-installed SDK (self-hosted); API calls go to OpenAI's hosted endpoints. (self-hosted)
SourceHosted tracing logs runs to OpenAI's platform; self-hosted logging is operator-implemented.
SourceSelf-hosted library; SOC 2 applies to the operator's infrastructure, not the package.
Self-hosted library; ISO 27001 applies to the operator's ISMS, not the package.
Self-hosted library; ISO 42001 applies to the deploying organisation.
Developer component, not an AI system placed on the market; obligations rest with the deployer. (role: not-applicable)
MIT — permissive, OSI-approved. (low)
SourceChange history
Source-backed and auto-detected events for this tool, newest first.
- Release
OpenAI Agents SDK 0.18.2 added GPT-5.6 request controls and hosted multi-agent beta support, plus sandbox and usage fixes.
Release v0.18.2 · openai/openai-agents-python - Release
OpenAI Agents SDK 0.18.1 added GPT-5.6 model defaults and migrated examples, with cache-write usage fixes.
Release v0.18.1 · openai/openai-agents-python - Release
OpenAI Agents SDK 0.18.0 changed the default RealtimeAgent model to gpt-realtime-2.1.
Release v0.18.0 · openai/openai-agents-python - Release
OpenAI Agents SDK 0.17.8 added invalid final-output recovery handling and fixed realtime session cancellation propagation.
Release v0.17.8 · openai/openai-agents-python - ReleaseHigh impact
OpenAI Agents SDK 0.17.7 exposed configurable WebSocket size limits, added buffered Chat Completions tool-call streaming, made AdvancedSQLiteSession add_items atomic, and tightened sandbox, guardrail, and Realtime multi-agent edge cases.
Release v0.17.7 · openai/openai-agents-python - ReleaseHigh impact
OpenAI Agents SDK 0.17.6 added pre-approval tool-input guardrails and SDK-only custom tool-output metadata, tightening human approval control points around agent tool execution.
Release v0.17.6 · openai/openai-agents-python - ReleaseHigh impact
OpenAI Agents SDK 0.16.0 changed the default model to gpt-5.4-mini for runs without an explicit model, altering baseline behavior, reasoning defaults, and potential cost/perf expectations.
Release v0.16.0 · openai/openai-agents-python · GitHub - ReleaseHigh impact
OpenAI Agents SDK 0.15.0 changed refusal handling to raise ModelRefusalError explicitly instead of surfacing empty outputs, a behavior change that can affect run-loop integrations.
Release v0.15.0 · openai/openai-agents-python · GitHub - ReleaseOpenAI Agents SDK release hardens tracing and SQL session handling
The OpenAI Agents SDK April 9 release hardened SQLite and SQLAlchemy session handling and expanded tracing documentation, signaling continued stabilization of the production agent stack.
Releases · openai/openai-agents-python
Explore the category
Compare this tool against the rest of its category and the cloud platform foundation layer.
Return to the category grid to compare governance posture across every tracked tool.
Review Microsoft Foundry, Amazon Bedrock, and Gemini Enterprise Agent Platform as the foundation layer.
Track releases, deprecations, license changes, and other market movements.