v0.2.1

Key changes

1. Realtime agents are now available. Note that this is in Beta and may have bugs, please report them!
2. Support for structuredContent in MCP tool results.

What's Changed

• Realtime: use SDK types for all messages by @rm-openai in #1134
• Realtime: introduce a demo without a UI by @rm-openai in #1135
• Realtime: handoffs by @rm-openai in #1139
• Realtime: move demo audio to separate thread by @rm-openai in #1141
• MCP: add support for tool_result.structuredContent by @rm-openai in #1150
• Correct error message for invalid output types with strict JSON schema by @DanielHashmi in #1101
• Add Streaming of Function Call Arguments to Chat Completions by @devtalker in #999
• Empty mcp tool result error by @thoo in #1142
• Unbreak tests by @rm-openai in #1157
• Realtime: web demo by @rm-openai in #1149
• Realtime docs by @rm-openai in #1153
• v0.2.1 by @rm-openai in #1151

Full Changelog: v0.2.0...v0.2.1

v0.2.0

Please note that this version includes a small breaking change (breaks Agent into AgentBase + Agent), hence the minor version bump. See https://openai.github.io/openai-agents-python/release/ for the breaking change changelog.

Key changes include:

1. Support for Sessions, a way to manage conversation sessions.
2. Beta support for Realtime agents (likely buggy, and we are evolving/fixing it as we speak)

What's Changed

• docs: mention 'name' is required in Agent constructor . fix: mention 'name' as required in Agent init doc by @maneeshanif in #1033
• Fix agent lifecycle example output to reflect correct hook execution order by @DanielHashmi in #1015
• Support mcp prompts by @thoo in #1010
• Add on_start support to VoiceWorkflowBase and VoicePipeline by @vrtnis in #922
• Add Makefile check rule to run all the linters and checkers by @ant31 in #1038
• Support for file_input content (#fix 893) by @ant31 in #1009
• description_override is not properly used for function_schema.description by @HafizFasih in #1000
• avoid concat tool call delta id by @DeoLeung in #990
• Fix #976 MCP filtering, make agent and run ctx optional by @njbrake in #977
• Add Sessions for Automatic Conversation History Management by @knowsuchagency in #752
• Updated llm_as_a_judge.py by @zaidrj in #1055
• [1/n] Break Agent into AgentBase+Agent by @rm-openai in #1068
• [2/n] Introduce RealtimeAgent by @rm-openai in #1069
• [3/n] Config and items for realtime by @rm-openai in #1070
• [4/n] Transport interface + events by @rm-openai in #1071
• [5/n] OpenAI realtime transport impl by @rm-openai in #1072
• [6/n] RealtimeSession + events by @rm-openai in #1073
• [7/n] Demo for realtime by @rm-openai in #1074
• Fix typo in Evaluator Instructions (capitalization of "After 5 attemp… by @mshsheikh in #1067
• [8/n] Make realtime more like the rest of agents sdk by @rm-openai in #1076
• Tests for realtime runner by @rm-openai in #1079
• Tests for realtime session by @rm-openai in #1080
• Add tests for openai realtime model by @rm-openai in #1081
• Fix #968 by upgrading openai package to the latest by @seratch in #1034
• Realtime guardrail support by @rm-openai in #1082
• Realtime tracing by @rm-openai in #1084
• enhancement: Add tool_name to ToolContext to support shared tool handlers by @vrtnis in #1043
• Fix incorrect comment order when popping session items by @MuhammadHassaanArain in #1086
• Back out "enhancement: Add tool_name to ToolContext to support shared tool handlers (#1043)" by @rm-openai in #1105
• Realtime: run guardrails without blockign event loop by @rm-openai in #1104
• Run make format by @rm-openai in #1106
• Realtime: forward exceptions from transport layer by @rm-openai in #1107
• Realtime: update model to have a single send_event method by @rm-openai in #1111
• Realtime: send session.update event at connection/handoffs by @rm-openai in #1112
• add numpy dev dependency by @jhills20 in #1117
• feat(function_schema): Add support for pydantic Field annotations in tool arguments (for tools decorated with @function_schema) by @georg-wolflein in #1124
• Fix typos and update async for loop variable by @sarmadalikhanofficial in #1122
• Realtime: export everything by @rm-openai in #1118
• Update examples docs with realtime section by @jhills20 in #1113
• Retry: Add tool_name to ToolContext for generic tool handlers by @vrtnis in #1110
• update-input-guardrail-name by @Quratulain-bilal in #1053
• Realtime: fix item parsing by @rm-openai in #1119

New Contributors

• @maneeshanif made their first contribution in #1033
• @thoo made their first contribution in #1010
• @vrtnis made their first contribution in #922
• @ant31 made their first contribution in #1038
• @HafizFasih made their first contribution in #1000
• @DeoLeung made their first contribution in #990
• @knowsuchagency made their first contribution in #752
• @zaidrj made their first contribution in #1055
• @mshsheikh made their first contribution in #1067
• @MuhammadHassaanArain made their first contribution in #1086
• @georg-wolflein made their first contribution in #1124
• @sarmadalikhanofficial made their first contribution in #1122
• @Quratulain-bilal made their first contribution in #1053

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Please note that this version includes a small breaking change to MCP servers, hence the minor version bump. See https://openai.github.io/openai-agents-python/release/ for the breaking change changelog.

What's Changed

• Add is_enabled to handoffs by @rm-openai in #925
• Removed lines to avoid duplicate output in REPL utility by @DanieleMorotti in #928
• Bugfix | Fixed a bug when calling reasoning models with store=False by @niv-hertz in #920
• Point CLAUDE.md to AGENTS.md by @rm-openai in #930
• Fix #890 by adjusting the guardrail document page by @seratch in #903
• Add exempt-issue-labels to the issue triage GH Action job by @seratch in #936
• Remove duplicate entry from __init__.py by @Lightblues in #897
• Fix #604 Chat Completion model raises runtime error when response.choices is empty by @seratch in #935
• Fix #892 by adding proper tool description in context.md by @Abbas-Asad in #937
• Duplicate Classifier in pyproject.toml by @DanielHashmi in #951
• fix: add ensure_ascii=False to json.dumps for correct Unicode output by @KatHaruto in #639
• Add uv as an alternative Python environment setup option for issue #884 by @Sucran in #909
• Fix and Document parallel_tool_calls Attribute in ModelSettings by @Rehan-Ul-Haq in #763
• replace .py file with .ipynb for Jupyter example by @ccmien in #262
• feat: add MCP tool filtering support by @devtalker in #861
• Tweak in pyproject.toml by @seratch in #952
• Fix incorrect argument description in on_trace_end docstring by @shirazkk in #958
• Fix #944 by updating the models document page to cover extra_args by @seratch in #950
• Annotating the openai.Omit type so that ModelSettings can be serialized by pydantic by @tconley1428 in #938
• Import Path Inconsistency by @DanielHashmi in #960
• Add reasoning content - fix on #494 by @axion66 in #871
• Add safety check handling for ComputerTool by @rm-openai in #923
• v0.1.0 by @rm-openai in #963

New Contributors

• @Lightblues made their first contribution in #897
• @Abbas-Asad made their first contribution in #937
• @DanielHashmi made their first contribution in #951
• @KatHaruto made their first contribution in #639
• @Sucran made their first contribution in #909
• @devtalker made their first contribution in #861
• @shirazkk made their first contribution in #958
• @tconley1428 made their first contribution in #938
• @axion66 made their first contribution in #871

Full Changelog: v0.0.19...v0.1.0

v0.0.19

What's Changed

• Make Runner an abstract base class by @pakrym-oai in #720
• Prepare 0.0.19 release by @pakrym-oai in #895

Full Changelog: v0.0.18...v0.0.19

v0.0.18

Key changes

1. Added support for dynamic prompt templates via the OpenAI Prompts feature
2. REPL support
3. Bug fixes

What's Changed

• Add REPL run_demo_loop helper by @rm-openai in #811
• Crosslink to js/ts by @rm-openai in #815
• Add release documentation by @rm-openai in #814
• Fix handoff transfer message JSON by @jhills20 in #818
• docs: custom output extraction by @jleguina in #817
• Added support for passing tool_call_id via the RunContextWrapper by @niv-hertz in #766
• Allow arbitrary kwargs in model by @rm-openai in #842
• Fix function_schema name override bug by @rm-openai in #872
• adopted float instead of timedelta for timeout parameters by @DanieleMorotti in #874
• Prompts support by @rm-openai in #876
• v0.0.18 by @rm-openai in #878

New Contributors

• @jleguina made their first contribution in #817
• @niv-hertz made their first contribution in #766

Full Changelog: v0.0.17...v0.0.18

v0.0.17

What's Changed

• fix Gemini token validation issue with LiteLLM by @handrew in #735
• Fix visualization recursion with cycle detection by @rm-openai in #737
• Update MCP and tool docs by @rm-openai in #736
• Fix Gemini API content filter handling by @rm-openai in #746
• Add Portkey AI as a tracing provider by @siddharthsambharia-portkey in #785
• Added RunErrorDetails object for MaxTurnsExceeded exception by @DanieleMorotti in #743
• Fixed Python syntax by @sarmadgulzar in #665
• Small fix for litellm model by @robtinn in #789
• Fix typo in assertion message for handoff function by @Rehan-Ul-Haq in #780
• Fix typo: Replace 'two' with 'three' in /docs/mcp.md by @luochang212 in #757
• Update input_guardrails.py by @venkatnaveen7 in #774
• docs: fix typo in docstring for is_strict_json_schema method by @Rehan-Ul-Haq in #775
• Add comment to handoff_occured misspelling by @rm-openai in #792
• Fix #777 by handling MCPCall events in RunImpl by @seratch in #799
• Ensure item.model_dump only contains JSON serializable types by @westhood in #801
• Don't cache agent tools during a run by @rm-openai in #803
• Only start tracing worker thread on first span/trace by @rm-openai in #804
• Add is_enabled to FunctionTool by @rm-openai in #808
• v0.0.17 by @rm-openai in #809

New Contributors

• @siddharthsambharia-portkey made their first contribution in #785
• @sarmadgulzar made their first contribution in #665
• @robtinn made their first contribution in #789
• @Rehan-Ul-Haq made their first contribution in #780
• @luochang212 made their first contribution in #757
• @venkatnaveen7 made their first contribution in #774
• @westhood made their first contribution in #801

Full Changelog: v0.0.16...v0.0.17

v0.0.16

Biggest change is support for new hosted tools - remote MCP, code interpreter, image generator and local shell.

What's Changed

• Create AGENTS.md by @dkundel-openai in #707
• Added mcp 'instructions' attribute to the server by @DanieleMorotti in #706
• Add Galileo to external tracing processors list by @franz101 in #662
• DRAFT: Dev/add usage details to Usage class by @WJPBProjects in #726
• Upgrade openAI sdk version by @rm-openai in #730
• Hosted MCP support by @rm-openai in #731
• Add support for local shell, image generator, code interpreter tools by @rm-openai in #732
• v0.0.16 by @rm-openai in #733

New Contributors

• @franz101 made their first contribution in #662
• @WJPBProjects made their first contribution in #726

Full Changelog: v0.0.15...v0.0.16

v0.0.15

Main big change is support for Streamable HTTP in MCP servers.

What's Changed

• Fixed a bug for "detail" attribute in input image by @DanieleMorotti in #685
• feat: pass extra_body through to LiteLLM acompletion by @AshokSaravanan222 in #638
• Update search_agent.py by @leohpark in #677
• feat: Streamable HTTP support by @Akshit97 in #643
• v0.0.15 by @rm-openai in #701

New Contributors

• @AshokSaravanan222 made their first contribution in #638
• @leohpark made their first contribution in #677
• @Akshit97 made their first contribution in #643

Full Changelog: v0.0.14...v0.0.15

v0.0.14

What's Changed

• Add usage to context in streaming by @rm-openai in #595
• Make the TTS voices type exportable by @mangiucugna in #577
• docs: add FutureAGI to tracing documentation by @NVJKKartik in #592
• Update litellm version by @pakrym-oai in #626
• 0.0.14 release by @pakrym-oai in #635

New Contributors

• @mangiucugna made their first contribution in #577
• @NVJKKartik made their first contribution in #592

Full Changelog: v0.0.13...v0.0.14

v0.0.13

What's Changed

• Adding extra_headers parameters to ModelSettings by @jonnyk20 in #550
• Examples: Fix financial_research_agent instructions by @seratch in #573
• Allow cancel out of the streaming result by @handrew in #579
• Create to_json_dict for ModelSettings by @rm-openai in #582
• Prevent MCP ClientSession hang by @njbrake in #580
• Fix stream error using LiteLLM by @DanieleMorotti in #589
• More tests for cancelling streamed run by @rm-openai in #590
• v0.0.13 by @rm-openai in #593

New Contributors

• @jonnyk20 made their first contribution in #550
• @handrew made their first contribution in #579
• @njbrake made their first contribution in #580

Full Changelog: v0.0.12...v0.0.13