9 Pull requests merged by 6 people

• Fix #892 by adding proper tool description in context.md

  #937 merged Jun 25, 2025

• Fix #604 Chat Completion model raises runtime error when response.choices is empty

  #935 merged Jun 25, 2025

• Remove duplicate entry from __init__.py

  #897 merged Jun 25, 2025

• Add exempt-issue-labels to the issue triage GH Action job

  #936 merged Jun 25, 2025

• Fix #890 by adjusting the guardrail document page

  #903 merged Jun 24, 2025

• Point CLAUDE.md to AGENTS.md

  #930 merged Jun 24, 2025

• Bugfix | Fixed a bug when calling reasoning models with store=False

  #920 merged Jun 24, 2025

• Removed lines to avoid duplicate output in REPL utility

  #928 merged Jun 24, 2025

• Add is_enabled to handoffs

  #925 merged Jun 24, 2025

6 Pull requests opened by 5 people

• Add on_start support to VoiceWorkflowBase and VoicePipeline

  #922 opened Jun 23, 2025

• Add safety check handling for ComputerTool

  #923 opened Jun 23, 2025

• feat(tools): add conversation history support to ToolContext #904

  #926 opened Jun 24, 2025

• Ensure that input_guardrails can block tools from running

  #931 opened Jun 24, 2025

• Annotating the openai.Omit type so that ModelSettings can be serialized by pydantic

  #938 opened Jun 25, 2025

• Support Claude extended thinking

  #941 opened Jun 25, 2025

27 Issues closed by 4 people

• [Bug]: SDK crashes when `choices` is `None` (provider-error payload)

  #604 closed Jun 25, 2025

• Configure `service_tier` param

  #934 closed Jun 25, 2025

• the same isinstance(output, ResponseFunctionToolCall) check twice in "_run_impl.py "

  #658 closed Jun 25, 2025

• How to ensure a specific tool sequencing?

  #918 closed Jun 25, 2025

• Cannot run Gemini

  #829 closed Jun 25, 2025

• Make FuncTool and @function_tool decorated function callable

  #708 closed Jun 25, 2025

• openai.BadRequestError: Error code: 400 - {'error': {'message': 'Invalid request: prompt is empty or malformed.', 'type': 'invalid_request_error', 'param': None, 'code': None}}

  #716 closed Jun 25, 2025

• Make intermediate results available when `MaxTurnExceededException` is thrown

  #719 closed Jun 25, 2025

• ValidationError` from `InputTokensDetails` when using `LitellmModel` with `None` cached\_tokens

  #760 closed Jun 25, 2025

• Bug Report: Reasoning Mode Incompatible with Tools on Bedrock Claude via LiteLLM model class

  #810 closed Jun 25, 2025

• Bug: style guideline and formatting inconsistencies

  #611 closed Jun 25, 2025

• no attribute error occurs while calling MCP

  #630 closed Jun 25, 2025

• ValueError: <Token var=<ContextVar name='current_trace' default=None at 0x7f9c4462cf40> at 0x7f9c42267c80> was created in a different Context

  #634 closed Jun 25, 2025

• Feature Request: Allow Separate Models for Tool Execution and Final Response in OpenAI Agent SDK

  #684 closed Jun 25, 2025

• ImportError: cannot import name 'MCPServerStdio' from 'agents.mcp'

  #691 closed Jun 25, 2025

• Streaming Fails Due to `{"include_usage": True}`

  #442 closed Jun 25, 2025

• input_guardrail is skipped

  #576 closed Jun 25, 2025

• Context restrictions

  #283 closed Jun 25, 2025

• Errors from custom model providers dont handled

  #380 closed Jun 25, 2025

• Add the possibility to add extra header fields in the RunConfig or Agents

  #549 closed Jun 25, 2025

• Add HTTP Streamable support for MCP's

  #600 closed Jun 25, 2025

• Typo in Output Guardrails documentation: "First, the guardrail receives the same input passed to the agent."

  #890 closed Jun 24, 2025

• [Feature Request] Add REPL-style `run_demo_loop` utility for rapid agent testing and debugging

  #784 closed Jun 24, 2025

• Bug | Reasoning on OpenAI's models doesn't work with `store=False`

  #919 closed Jun 24, 2025

• Missing docstring in fetch_user_age breaks tool understanding

  #892 closed Jun 23, 2025

• Multiplying by 10 in @function_tool return value does not work

  #896 closed Jun 23, 2025

• documentation

  #916 closed Jun 23, 2025

5 Issues opened by 5 people

• Go version of the Agents SDK?

  #942 opened Jun 26, 2025

• Expose tool call arguments in RunHooks on_tool_start and on_tool_end

  #939 opened Jun 25, 2025

• Document upload across gemini, claude and gpt (completions API)

  #933 opened Jun 24, 2025

• Provided By OpenAI Kittens SSO

  #929 opened Jun 24, 2025

• OpenAI SDK does not expose MCP session ID, making stateful streamable HTTP sessions impossible

  #924 opened Jun 23, 2025

158 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.

• feat: add MCP tool filtering support

  #861 commented on Jun 25, 2025 • 6 new comments

• add voice option "ballad" in src/agents/voice/model.py

  #278 commented on Jun 25, 2025 • 1 new comment

• Fix typos in event names and comments across multiple files

  #155 commented on Jun 25, 2025 • 1 new comment

• parallel_tool_calls default is True in the client, despite docs saying False

  #762 commented on Jun 25, 2025 • 0 new comments

• Yield reasoning delta in the response or add hooks to handle

  #825 commented on Jun 25, 2025 • 0 new comments

• Streaming Output issue got unnecesory text in output

  #824 commented on Jun 25, 2025 • 0 new comments

• Error initializing MCP server: Connection closed

  #822 commented on Jun 25, 2025 • 0 new comments

• openai_stt.py _turn_audio_buffer maybe empty

  #821 commented on Jun 25, 2025 • 0 new comments

• Adding a ToolRAG that the user can enable when he gives to the agent many tools.

  #913 commented on Jun 25, 2025 • 0 new comments

• Adding a knowledge parameter within the Agent class

  #914 commented on Jun 25, 2025 • 0 new comments

• CodeInterpreter tool is appending data to the end of the filename in containers

  #911 commented on Jun 25, 2025 • 0 new comments

• When using a llmproxy to access model, the part of code remove the model prefix, but llm proxy need it.

  #910 commented on Jun 25, 2025 • 0 new comments

• Support Anthropic Prompt Caching

  #905 commented on Jun 25, 2025 • 0 new comments

• When running in non-streaming mode, retrieve past history during tool execution.

  #904 commented on Jun 25, 2025 • 0 new comments

• `output_type=List[str]` silently disables “thinking” in Qwen-3 (and similar models), cutting run-time by ~50 % compared with `output_type=str`

  #902 commented on Jun 25, 2025 • 0 new comments

• OpenAI 'file' message type is supported by Chat Completions

  #893 commented on Jun 25, 2025 • 0 new comments

• KeyError: ~TContext while importing with Python 3.11.0

  #891 commented on Jun 25, 2025 • 0 new comments

• FileSearchTool runs despite InputGuardrailTripwireTriggered

  #889 commented on Jun 25, 2025 • 0 new comments

• Support for short‑term and long‑term memory in Agents SDK

  #887 commented on Jun 25, 2025 • 0 new comments

• Add uv as well

  #884 commented on Jun 25, 2025 • 0 new comments

• Get pydantic serializer warnings in calling Runner.run

  #880 commented on Jun 25, 2025 • 0 new comments

• MCP tools turn upstream HTTP errors into AgentsException, crashing the run

  #879 commented on Jun 25, 2025 • 0 new comments

• LiteLLM + Gemini 2.5 Pro: cached_tokens=None crashes Agents SDK with Pydantic int-validation error

  #758 commented on Jun 25, 2025 • 0 new comments

• support for bedrock prompt caching

  #750 commented on Jun 25, 2025 • 0 new comments

• Add Session Memory

  #745 commented on Jun 25, 2025 • 0 new comments

• Unable to use tool calling when using Llama 4 scout via LiteLLM Proxy

  #723 commented on Jun 25, 2025 • 0 new comments

• Allow agent to return logprobs

  #715 commented on Jun 25, 2025 • 0 new comments

• LLM Cannot Access tool_context.context (e.g. user_uuid) for Tool Input Interpolation in MCP Workflows

  #711 commented on Jun 25, 2025 • 0 new comments

• Prevent Token Wastage When Input Guardrails Delay Error Generation

  #867 commented on Jun 25, 2025 • 0 new comments

• Agent.as_tool hides nested tool‑call events — blocks parallel sub‑agents with streaming

  #864 commented on Jun 25, 2025 • 0 new comments

• Tools selection from MCP server

  #863 commented on Jun 25, 2025 • 0 new comments

• Mixture of Tool-Calling and Handoff

  #858 commented on Jun 25, 2025 • 0 new comments

• Filter MCPServerStdio server tools

  #851 commented on Jun 25, 2025 • 0 new comments

• Support automatic "back" handoffs to orchestrating Agents

  #847 commented on Jun 25, 2025 • 0 new comments

• Inconsistent timeout and sse_read_timeout Types in MCPServerSseParams and MCPServerStreamableHttpParams

  #845 commented on Jun 25, 2025 • 0 new comments

• acknowledged_safety_checks during response

  #843 commented on Jun 25, 2025 • 0 new comments

• When using the Azure LLM key and employing "Runner.run_streamed", the token usage returned is 0.

  #838 commented on Jun 25, 2025 • 0 new comments

• Support Predicted Outputs

  #837 commented on Jun 25, 2025 • 0 new comments

• Connection Error on Windows When Running Official LiteLLM Example

  #836 commented on Jun 25, 2025 • 0 new comments

• Support Streaming of Function Call Arguments

  #834 commented on Jun 25, 2025 • 0 new comments

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

  #831 commented on Jun 25, 2025 • 0 new comments

• Update all examples details and links in examples.md

  #250 commented on Jun 25, 2025 • 0 new comments

• Add tool call parameters for `on_tool_start` hook

  #253 commented on Jun 25, 2025 • 0 new comments

• replace .py file with .ipynb for Jupyter example

  #262 commented on Jun 25, 2025 • 0 new comments

• Update examples.md

  #271 commented on Jun 25, 2025 • 0 new comments

• Support a callable model

  #312 commented on Jun 25, 2025 • 0 new comments

• add reasoning content to ChatCompletions

  #494 commented on Jun 25, 2025 • 0 new comments

• Make input/new items available in the run context

  #572 commented on Jun 25, 2025 • 0 new comments

• fix: add ensure_ascii=False to json.dumps for correct Unicode output

  #639 commented on Jun 26, 2025 • 0 new comments

• Added support for gpt4o-realtime models for Speect to Speech interactions

  #659 commented on Jun 25, 2025 • 0 new comments

• Add Sessions for Automatic Conversation History Management

  #752 commented on Jun 26, 2025 • 0 new comments

• Fix and Document `parallel_tool_calls` Attribute in ModelSettings

  #763 commented on Jun 25, 2025 • 0 new comments

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

  #820 commented on Jun 25, 2025 • 0 new comments

• feat: separate stream items for tool_call_item and tool_call_output_item

  #833 commented on Jun 25, 2025 • 0 new comments

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

  #839 commented on Jun 25, 2025 • 0 new comments

• Added support for "return" handoffs (#1)

  #869 commented on Jun 25, 2025 • 0 new comments

• Add reasoning content - fix on #494

  #871 commented on Jun 25, 2025 • 0 new comments

• add example of gradio in streaming agent with mcp cases

  #888 commented on Jun 25, 2025 • 0 new comments

• feat: Support Anthropic Claude prompt caching key "cache_control"

  #908 commented on Jun 25, 2025 • 0 new comments

• Add uv as an alternative Python environment setup option for issue #884

  #909 commented on Jun 25, 2025 • 0 new comments

• Can tools ingest images attached to image_url?

  #875 commented on Jun 25, 2025 • 0 new comments

• Invalid parameter: messages with role 'tool' must be a response to a preceeding message with 'tool_calls'.

  #873 commented on Jun 25, 2025 • 0 new comments

• OPENAI Authentication error

  #870 commented on Jun 25, 2025 • 0 new comments

• Add `method_tool` Functionality

  #94 commented on Jun 25, 2025 • 0 new comments

• Very high response times at random during hand-offs

  #717 commented on Jun 25, 2025 • 0 new comments

• [Bug]: UnicodeDecodeError when importing litellm_model on Windows

  #610 commented on Jun 25, 2025 • 0 new comments

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

  #787 commented on Jun 25, 2025 • 0 new comments

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

  #894 commented on Jun 25, 2025 • 0 new comments

• Mem0ai integration as example for create memory agents

  #832 commented on Jun 26, 2025 • 0 new comments

• results field in Code Interpreter always returns None

  #917 commented on Jun 26, 2025 • 0 new comments

• fix(#94): handle special parameters (self/cls) correctly with context parameter

  #137 commented on Jun 25, 2025 • 0 new comments

• perf: Optimize has_more_than_n_keys function and BatchTraceProcessor

  #143 commented on Jun 25, 2025 • 0 new comments

• Add FastAPI example

  #168 commented on Jun 25, 2025 • 0 new comments

• Add PDF extraction example agent with verification

  #176 commented on Jun 25, 2025 • 0 new comments

• Create a Tauri desktop app with Lynx UI for OpenAI Agent SDK

  #179 commented on Jun 25, 2025 • 0 new comments

• Update README.md

  #180 commented on Jun 25, 2025 • 0 new comments

• Add Firecrawl integration for web scraping and information extraction

  #182 commented on Jun 25, 2025 • 0 new comments

• UI-for-examples

  #228 commented on Jun 25, 2025 • 0 new comments

• update bullets for documentation in tool.py

  #247 commented on Jun 25, 2025 • 0 new comments

• Custom model provider agent does not use tool calling when an output type is specified

  #332 commented on Jun 25, 2025 • 0 new comments

• Feature Request: Support Manual Interruption of OpenAI Agent During Execution

  #329 commented on Jun 25, 2025 • 0 new comments

• `tts_settings` should be available per Agent

  #327 commented on Jun 25, 2025 • 0 new comments

• Question about Tool Call Streaming

  #326 commented on Jun 25, 2025 • 0 new comments

• Retry mechanism for ModelBehaviorError

  #325 commented on Jun 25, 2025 • 0 new comments

• Callable model

  #310 commented on Jun 25, 2025 • 0 new comments

• Streamed Voice Agent Demo - Multiple Performance Issues

  #301 commented on Jun 25, 2025 • 0 new comments

• Integrate more 3rd party tools into AgentSDK tool

  #299 commented on Jun 25, 2025 • 0 new comments

• Best practices for long running tools?

  #295 commented on Jun 25, 2025 • 0 new comments

• agents.exceptions.ModelBehaviorError: Invalid JSON when parsing

  #280 commented on Jun 25, 2025 • 0 new comments

• Handoff does not work with Claude 3.7 Sonnet

  #270 commented on Jun 25, 2025 • 0 new comments

• Truncate span input when input is too large

  #260 commented on Jun 25, 2025 • 0 new comments

• Provide a agent hook `on_get_response` to recalculate the messages passed to the model

  #257 commented on Jun 25, 2025 • 0 new comments

• Ordering of events in Runner.run_streamed is incorrect

  #583 commented on Jun 25, 2025 • 0 new comments

• Missing Handling for `delta.reasoning_content` in `agents.models.chatcmpl_stream_handler.ChatCmplStreamHandler.handle_stream`

  #578 commented on Jun 25, 2025 • 0 new comments

• bugs in run.py

  #570 commented on Jun 25, 2025 • 0 new comments

• Reasoning model items provide to General model

  #569 commented on Jun 25, 2025 • 0 new comments

• Why does the Computer protocol not have the goto method?

  #547 commented on Jun 25, 2025 • 0 new comments

• History Cleaning

  #545 commented on Jun 25, 2025 • 0 new comments

• from agents.extensions.models.litellm_model import LitellmModel

  #666 commented on Jun 25, 2025 • 0 new comments

• Best practices for integrating Gradio's Agent and Tool

  #862 commented on Jun 24, 2025 • 0 new comments

• Empty assistant message after a hand-off

  #856 commented on Jun 24, 2025 • 0 new comments

• Best setup for tracing with Azure

  #846 commented on Jun 24, 2025 • 0 new comments

• Add llms.txt in the documentation

  #556 commented on Jun 25, 2025 • 0 new comments

• Usage tokens no longer automatically show

  #548 commented on Jun 25, 2025 • 0 new comments

• Passing the Encoded image to @tool

  #883 commented on Jun 25, 2025 • 0 new comments

• BadRequestError while trying out input guardrails

  #129 commented on Jun 25, 2025 • 0 new comments

• Accessing Reasoning Summary for ComputerUseTool-based Agent

  #202 commented on Jun 25, 2025 • 0 new comments

• Computer tool can not be used with Web search call.

  #254 commented on Jun 25, 2025 • 0 new comments

• Agents SDK v0.0.4: Local Traces Missing Critical LLM Outputs and Tool Interaction Data

  #222 commented on Jun 25, 2025 • 0 new comments

• invalid_request_error when using "chat_completions" with triage agent (gemini -> any other model)

  #237 commented on Jun 25, 2025 • 0 new comments

• Add `reasoning_content` to ChatCompletions

  #415 commented on Jun 25, 2025 • 0 new comments

• Rate limit in RunConfig

  #381 commented on Jun 25, 2025 • 0 new comments

• Only use certain tools from MCP server

  #376 commented on Jun 25, 2025 • 0 new comments

• The voice models aren't waiting for a pause(a very natural phenomenon when using a voice agent) and generating an output mid sentence when talking to the voice agent implemented using "VoicePipline" .

  #369 commented on Jun 25, 2025 • 0 new comments

• Random transcript gets printed/generated when talking to the voice agent implemented using "VoicePipline" . Eg - "Transcription: Kurs." Mind you there is no background noise.

  #368 commented on Jun 25, 2025 • 0 new comments

• Implementing a comprehensive workflow

  #366 commented on Jun 25, 2025 • 0 new comments

• Mermaid-based visualization

  #352 commented on Jun 25, 2025 • 0 new comments

• Access to Chat History in Hooks

  #346 commented on Jun 25, 2025 • 0 new comments

• Add Support for Image Return in Agent Tools

  #341 commented on Jun 25, 2025 • 0 new comments

• Custom model provider ignored when using agents as tools

  #663 commented on Jun 25, 2025 • 0 new comments

• First-class streaming tool output

  #661 commented on Jun 25, 2025 • 0 new comments

• OAuth support for MCPServerSse

  #657 commented on Jun 25, 2025 • 0 new comments

• Function calling fails on “application/json” MIME type with the latest Gemini models

  #656 commented on Jun 25, 2025 • 0 new comments

• Human-In-The-Loop Architecture should be implemented on top priority!

  #636 commented on Jun 25, 2025 • 0 new comments

• Intent Classifier Support

  #628 commented on Jun 25, 2025 • 0 new comments

• on_agent_start hook should be more performant

  #623 commented on Jun 25, 2025 • 0 new comments

• Can we use agent.run instead of Runner.run(starting_agent=agent)

  #622 commented on Jun 25, 2025 • 0 new comments

• Resource tracker warning (leaked semaphores) with MCPServerStdio

  #618 commented on Jun 25, 2025 • 0 new comments

• Random Invalid URL completion error on agents SDK

  #816 commented on Jun 25, 2025 • 0 new comments

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

  #813 commented on Jun 25, 2025 • 0 new comments

• Azure OpenAI rejects system prompt from prompt_with_handoff_instructions

  #806 commented on Jun 25, 2025 • 0 new comments

• Feature Request: Enhanced Run Lifecycle Management - Interrupt and Update Active Runs

  #798 commented on Jun 25, 2025 • 0 new comments

• SDK support for retrieving historical traces

  #793 commented on Jun 25, 2025 • 0 new comments

• Rate Limit Support

  #782 commented on Jun 25, 2025 • 0 new comments

• Bug: Pydantic Warnings and `.final_output` Serialization Issue with Code Interpreter File Citations

  #772 commented on Jun 25, 2025 • 0 new comments

• Only 1 handoff getting called no matter what

  #771 commented on Jun 25, 2025 • 0 new comments

• function_tool calling - agents orchertratins

  #769 commented on Jun 25, 2025 • 0 new comments

• Agents Reproducibility - seed ? top_p=0?

  #768 commented on Jun 25, 2025 • 0 new comments

• Support for MCP prompts and resources

  #544 commented on Jun 25, 2025 • 0 new comments

• Websocket streaming audio in realtime from client

  #536 commented on Jun 25, 2025 • 0 new comments

• Cannot get the last tool_call_output event in stream_events when MaxTurnsExceeded

  #526 commented on Jun 25, 2025 • 0 new comments

• Timeout after 300 seconds with any error message. Could it be rate limiting?

  #511 commented on Jun 25, 2025 • 0 new comments

• Obvious Pauses Between Text Segments in Current `OpenAITTSModel` Implementation Affect Speech Fluency

  #493 commented on Jun 25, 2025 • 0 new comments

• Add reasoning support for custom models.

  #492 commented on Jun 25, 2025 • 0 new comments

• Automatic Audio Feedback on Silence in VoicePipeline

  #489 commented on Jun 25, 2025 • 0 new comments

• Add Intro message function for VoicePipeline

  #488 commented on Jun 25, 2025 • 0 new comments

• Duplicate tool names across MCP servers cause errors

  #464 commented on Jun 25, 2025 • 0 new comments

• Support hosted tools across different model providers

  #461 commented on Jun 25, 2025 • 0 new comments

• SSL error for streaming voice terminal app

  #423 commented on Jun 25, 2025 • 0 new comments

• Custom-LLM <think> tag handling

  #703 commented on Jun 25, 2025 • 0 new comments

• MCP server restart cause Agent to fail

  #693 commented on Jun 25, 2025 • 0 new comments

• Feature Request: Support streaming tool call outputs

  #692 commented on Jun 25, 2025 • 0 new comments

• Please add time travel

  #688 commented on Jun 25, 2025 • 0 new comments

• Agent attempts to use non-existing tool

  #686 commented on Jun 25, 2025 • 0 new comments

• Add response cost in the Usage

  #683 commented on Jun 25, 2025 • 0 new comments

• Unable to use reasoning models with tool calls using LitellmModel

  #678 commented on Jun 25, 2025 • 0 new comments

• Tool calling with LiteLLM and thinking models fail

  #765 commented on Jun 25, 2025 • 0 new comments

• Error code: 400 "No tool output found for function call"

  #673 commented on Jun 25, 2025 • 0 new comments