3 Releases published by 1 person

• v0.2.0

  published Jul 15, 2025

• v0.2.1

  published Jul 16, 2025

• v0.2.2

  published Jul 17, 2025

43 Pull requests merged by 18 people

• Update all translated document pages

  #1184 merged Jul 18, 2025

• Fix image_generator example error on Windows OS

  #1180 merged Jul 18, 2025

• fix: Apply strict JSON schema validation in FunctionTool constructor

  #1041 merged Jul 18, 2025

• Mark some dataclasses as pydantic dataclasses

  #1131 merged Jul 18, 2025

• fix: fallback to function name for unnamed output_guardrail decorators

  #1133 merged Jul 18, 2025

• Add missing guardrail exception import to quickstart

  #1161 merged Jul 18, 2025

• fix: ensure ResponseUsage token fields are int, not None (fixes #1179)

  #1181 merged Jul 18, 2025

• Fix spelling, grammar, and minor formatting issues in running agents documentation

  #1128 merged Jul 18, 2025

• Update all translated document pages

  #1173 merged Jul 17, 2025

• [realtime] Add direct access to model layer from session

  #1172 merged Jul 17, 2025

• v0.2.2

  #1170 merged Jul 17, 2025

• Realtime: only update model settings from session

  #1169 merged Jul 17, 2025

• Adjust #598 to only create a PR rather than pushing changes to main branch

  #1162 merged Jul 17, 2025

• Add a new GH Actions job to automatically update translated document pagse

  #598 merged Jul 16, 2025

• v0.2.1

  #1151 merged Jul 16, 2025

• Realtime docs

  #1153 merged Jul 16, 2025

• Realtime: web demo

  #1149 merged Jul 16, 2025

• Unbreak tests

  #1157 merged Jul 16, 2025

• Empty mcp tool result error

  #1142 merged Jul 16, 2025

• Add Streaming of Function Call Arguments to Chat Completions

  #999 merged Jul 16, 2025

• Correct error message for invalid output types with strict JSON schema

  #1101 merged Jul 16, 2025

• MCP: add support for tool_result.structuredContent

  #1150 merged Jul 16, 2025

• Realtime: move demo audio to separate thread

  #1141 merged Jul 16, 2025

• Realtime: handoffs

  #1139 merged Jul 16, 2025

• Realtime: introduce a demo without a UI

  #1135 merged Jul 15, 2025

• Realtime: use SDK types for all messages

  #1134 merged Jul 15, 2025

• v0.2.0

  #1120 merged Jul 15, 2025

• Realtime: fix item parsing

  #1119 merged Jul 15, 2025

• update-input-guardrail-name

  #1053 merged Jul 15, 2025

• Retry: Add tool_name to ToolContext for generic tool handlers

  #1110 merged Jul 15, 2025

• Update examples docs with realtime section

  #1113 merged Jul 15, 2025

• Realtime: export everything

  #1118 merged Jul 15, 2025

• Fix typos and update async for loop variable

  #1122 merged Jul 15, 2025

• feat(function_schema): Add support for pydantic Field annotations in tool arguments (for tools decorated with @function_schema)

  #1124 merged Jul 15, 2025

• add numpy dev dependency

  #1117 merged Jul 15, 2025

• Realtime: send session.update event at connection/handoffs

  #1112 merged Jul 14, 2025

• Realtime: update model to have a single send_event method

  #1111 merged Jul 14, 2025

• Realtime: forward exceptions from transport layer

  #1107 merged Jul 14, 2025

• Run make format

  #1106 merged Jul 14, 2025

• Realtime: run guardrails without blockign event loop

  #1104 merged Jul 14, 2025

• Back out "enhancement: Add tool_name to ToolContext to support shared tool handlers (#1043)"

  #1105 merged Jul 14, 2025

• Fix incorrect comment order when popping session items

  #1086 merged Jul 14, 2025

• enhancement: Add tool_name to ToolContext to support shared tool handlers

  #1043 merged Jul 14, 2025

14 Pull requests opened by 10 people

• feat(function_schema): add enforce_type_annotations flag for stricter schema validation with type enforcement, clearer errors, and updated docstrings

  #1092 opened Jul 13, 2025

• fix: convert litellm response with reasoning content to openai message

  #1098 opened Jul 14, 2025

• Fix handling of text_message_outputs output structure in parallel translation example

  #1115 opened Jul 14, 2025

• docs: update GuardrailFunctionOutput docstrings for clarity

  #1152 opened Jul 16, 2025

• fix(logging): include system prompt in debug logs in case of response…

  #1154 opened Jul 16, 2025

• Example of Connecting Twilio to Realtime Agent SDK

  #1155 opened Jul 16, 2025

• Initial pass at PostgreSQL session management

  #1163 opened Jul 17, 2025

• feat: Add reasoning item handling to skip non-message content

  #1166 opened Jul 17, 2025

• Fix MCP tool name conflicts by adding server name prefixes

  #1178 opened Jul 18, 2025

• Realtime: add user agent header

  #1185 opened Jul 18, 2025

• docs(examples): use UserLocation dataclass in WebSearchTool example

  #1188 opened Jul 19, 2025

• Create pylint.yml

  #1189 opened Jul 19, 2025

• fix(mcp-server): assign original_name with type ignore and wrap long docstring

  #1190 opened Jul 19, 2025

• Feature: Make trace_include_sensitive_data configurable via environment variable

  #1192 opened Jul 19, 2025

34 Issues closed by 11 people

• Error Using Gemini API Key with Agent-as-Tool Pattern in openai-agent-sdk

  #986 closed Jul 20, 2025

• Agent unable to perform multiple Web Searches by using WebSearchTool

  #993 closed Jul 20, 2025

• Converter.items_to_messages cannot handle assistant message with both content and tool_calls

  #995 closed Jul 20, 2025

• How should I handle when the response from MCP or some tool is JSON, which contains the base64 encoding of the file.

  #1001 closed Jul 20, 2025

• Tool Call Execution Problem with Structured Output

  #1012 closed Jul 20, 2025

• Passing Local Context to MCP Server Tools

  #1021 closed Jul 20, 2025

• Reasoning only coming before final output

  #1036 closed Jul 20, 2025

• Import error - MCPServer

  #1059 closed Jul 19, 2025

• Specifying `model_provider` for each `Agent`

  #1137 closed Jul 18, 2025

• xAI Grok 4 Usage

  #1056 closed Jul 18, 2025

• Anthropic Tool calls failing with litellm after update to 0.2.0

  #1147 closed Jul 18, 2025

• FunctionTool ignores strict_json_schema parameter

  #992 closed Jul 18, 2025

• Streaming: ResponseUsage validation error when token fields are None

  #1179 closed Jul 18, 2025

• Retreive context variables on MCP server

  #1175 closed Jul 18, 2025

• Realtime Voice Not Quite Working

  #1136 closed Jul 16, 2025

• tool_name parameter on as_tool() does not default to agent name as expected

  #1159 closed Jul 16, 2025

• Agent(name=…) accepts non-string types due to Pydantic coercion

  #1148 closed Jul 16, 2025

• Dependencies for v0.2 Realtime Not Up to Date

  #1158 closed Jul 16, 2025

• Empty tool output incorrectly triggers “Errored MCP tool result”

  #1035 closed Jul 16, 2025

• Support Streaming of Function Call Arguments

  #834 closed Jul 16, 2025

• When will realtime capabilities be added like they are in the Typescript SDK?

  #894 closed Jul 16, 2025

• Intent Classifier Support

  #628 closed Jul 16, 2025

• Agent Initialization Allows name as Empty String or None Without Error — Missing Runtime Validation

  #1143 closed Jul 16, 2025

• Support pydantic `Field` annotations in tool arguments

  #1123 closed Jul 15, 2025

• Agent allows None for 'name' even though it's typed as str

  #1125 closed Jul 15, 2025

• How to provide nested arguments description for function_tool?

  #1096 closed Jul 15, 2025

• Sub-Agent Usage in Multi-Agent System with `Agent.as_tool`

  #1108 closed Jul 15, 2025

• Using LiteLLM router (with fallback and retry/cooldown) in an Agent

  #813 closed Jul 15, 2025

• Error initializing MCP server: Connection closed

  #822 closed Jul 15, 2025

• Add `tool_name` to `ToolContext` for generic tool handlers

  #1030 closed Jul 14, 2025

• Openai-agents memory package not present

  #1103 closed Jul 14, 2025

• MCP Streamable HTTP Server Client times out

  #1029 closed Jul 14, 2025

• Agent Class type

  #1095 closed Jul 14, 2025

• LLM Bypasses Tool Calls When output_type is Defined, Even With Explicit Tooling Instructions

  #1090 closed Jul 13, 2025

18 Issues opened by 16 people

• Draft issue: Additional information needed

  #1187 opened Jul 19, 2025

• Suggestions for using user feedback as tools?

  #1186 opened Jul 18, 2025

• Add support for global or environment-based default for `trace_include_sensitive_data` in `RunConfig`

  #1183 opened Jul 18, 2025

• cannot use custom cert when connecting to server with MCPServerSse

  #1182 opened Jul 18, 2025

• [Question] Relationships between instructions, prompt, handoff, and input_type/output_type in openai-agent-sdk

  #1177 opened Jul 18, 2025

• tool_name and tool_description for Agent#as_tool() method

  #1174 opened Jul 17, 2025

• Realtime API with Multiple Tool Calls Results in No Voice Output

  #1168 opened Jul 17, 2025

• Please fix the issue where duplicate tool names in multiple MCP tools cause the agent list to hang

  #1167 opened Jul 17, 2025

• items_to_messages method not handle "ResponseReasoningItemParam"

  #1165 opened Jul 17, 2025

• Unhandled item type or structure with resoning item

  #1164 opened Jul 17, 2025

• Tool Call Results are not Appearing in Realtime API and Tool Calls are Not Showing Up in Tracing

  #1156 opened Jul 16, 2025

• multiple inputs with session support

  #1140 opened Jul 16, 2025

• 2 Messages after a successful tool call

  #1129 opened Jul 15, 2025

• Halt `Runner` on specific tool call

  #1126 opened Jul 15, 2025

• Requesting x-request-id from the API Response

  #1121 opened Jul 15, 2025

• Add is_enabled option to Agent as a tool pattern

  #1097 opened Jul 14, 2025

• Better Error Handling For Long Contexts

  #1093 opened Jul 14, 2025

• Union type not permitted in AaaT

  #1091 opened Jul 13, 2025

39 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Added runtime validation for Agent name field

  #998 commented on Jul 18, 2025 • 1 new comment

• Only 1 handoff getting called no matter what

  #771 commented on Jul 14, 2025 • 0 new comments

• Function tool choice causing pydantic validation error

  #980 commented on Jul 19, 2025 • 0 new comments

• agents.exceptions.UserError: Unhandled item type or structure when use reasoning model with tool call.

  #1004 commented on Jul 19, 2025 • 0 new comments

• Add PDF extraction example agent with verification

  #176 commented on Jul 13, 2025 • 0 new comments

• Add tool call parameters for `on_tool_start` hook

  #253 commented on Jul 16, 2025 • 0 new comments

• feat(tools): run sync tools in a thread to avoid event loop blocking

  #820 commented on Jul 14, 2025 • 0 new comments

• [Fix]: Get logger from tracing folder instead of agents folder

  #839 commented on Jul 17, 2025 • 0 new comments

• Feat: separate tool_call_item and tool_call_output_item in stream events

  #974 commented on Jul 18, 2025 • 0 new comments

• Add MCP Tools to visualization.

  #979 commented on Jul 19, 2025 • 0 new comments

• feat(agents): Add on_llm_start and on_llm_end Lifecycle Hooks

  #987 commented on Jul 16, 2025 • 0 new comments

• Support OPENAI_BASE_URL in tracing

  #997 commented on Jul 16, 2025 • 0 new comments

• Add run configuration inheritance for agent-as-tool workflows

  #1006 commented on Jul 16, 2025 • 0 new comments

• Feat/support mcp resources

  #1042 commented on Jul 16, 2025 • 0 new comments

• Introduce RealtimeAgent

  #1045 commented on Jul 16, 2025 • 0 new comments

• wip

  #1046 commented on Jul 16, 2025 • 0 new comments

• Add automation script to generate mkdocstrings files

  #1048 commented on Jul 19, 2025 • 0 new comments

• Add streaming function call arguments example

  #1052 commented on Jul 16, 2025 • 0 new comments

• feat: #864 support streaming nested tool events in Agent.as_tool

  #1057 commented on Jul 16, 2025 • 0 new comments

• Feat/prompts caching

  #1087 commented on Jul 15, 2025 • 0 new comments

• function_tool calling - agents orchertratins

  #769 commented on Jul 14, 2025 • 0 new comments

• Agent attempts to use non-existing tool

  #686 commented on Jul 14, 2025 • 0 new comments

• reasoning_item_created never called on stream

  #1025 commented on Jul 14, 2025 • 0 new comments

• Support automatic "back" handoffs to orchestrating Agents

  #847 commented on Jul 14, 2025 • 0 new comments

• Local memory and context window length

  #177 commented on Jul 15, 2025 • 0 new comments

• Agent run with previous_response_id fails - No tool output found for function call call_WdnUUKXKvwy3jk....

  #1061 commented on Jul 15, 2025 • 0 new comments

• Input guardrail tripwire allows tool execution to continue after exception is raised

  #991 commented on Jul 15, 2025 • 0 new comments

• Error "Item ‘rs_ABCD’ of type ‘reasoning’ was provided without its required..." when using CodeInterpreter

  #985 commented on Jul 15, 2025 • 0 new comments

• Agent as a tool input_type

  #984 commented on Jul 16, 2025 • 0 new comments

• Cannot run visualization

  #978 commented on Jul 16, 2025 • 0 new comments

• tool_call_item event fires only after tool results instead of before

  #1037 commented on Jul 16, 2025 • 0 new comments

• Pydantic requirements mismatch

  #967 commented on Jul 17, 2025 • 0 new comments

• separate stream items for "tool_call_item" and "tool_call_output_item"

  #831 commented on Jul 17, 2025 • 0 new comments

• Add support for reference images in Agents SDK for image generation

  #1017 commented on Jul 17, 2025 • 0 new comments

• from LangChain and Langgraph to openai-agents

  #128 commented on Jul 17, 2025 • 0 new comments

• Prevent Token Wastage When Input Guardrails Delay Error Generation

  #867 commented on Jul 17, 2025 • 0 new comments

• No OpenAI tracing with Azure Foundry models

  #1051 commented on Jul 18, 2025 • 0 new comments

• Handoff to agent that uses websearch tool call fails. But works fine if agent is called as a tool

  #972 commented on Jul 18, 2025 • 0 new comments

• Processing image/ multi modal responses in function tool results?

  #787 commented on Jul 18, 2025 • 0 new comments