{"id":89825,"date":"2026-08-03T22:00:38","date_gmt":"2026-08-03T14:00:38","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/89825.html"},"modified":"2026-08-03T22:00:38","modified_gmt":"2026-08-03T14:00:38","slug":"openai-agents-python-sdk-%e6%ba%90%e7%a0%81%e8%a7%a3%e6%9e%90-%e7%ac%ac%e5%8d%81%e4%ba%8c%e7%af%87%ef%bc%9a%e6%a8%a1%e5%9e%8b%e9%80%82%e9%85%8d%e5%b1%82%ef%bc%9aresponses%e3%80%81chat-completions","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/89825.html","title":{"rendered":"openai-agents-python-sdk \u6e90\u7801\u89e3\u6790 | \u7b2c\u5341\u4e8c\u7bc7\uff1a\u6a21\u578b\u9002\u914d\u5c42\uff1aResponses\u3001Chat Completions \u4e0e\u7b2c\u4e09\u65b9 Provider"},"content":{"rendered":"

<\/h2>\n

\u672c\u7bc7\u5bfc\u8bfb<\/h3>\n

\u524d\u5341\u4e00\u7bc7\u5df2\u7ecf\u628a Agents SDK \u7684\u8fd0\u884c\u4e3b\u7ebf\u62c6\u5f00\u4e86：<\/p>\n

  • Agent \u5b9a\u4e49\u89d2\u8272\u3001\u5de5\u5177\u3001handoffs \u548c\u8f93\u51fa\u7c7b\u578b\u3002<\/li>\n
  • Runner \u63a8\u8fdb\u4e00\u6b21 run\u3002<\/li>\n
  • Function Tool \u628a Python \u51fd\u6570\u66b4\u9732\u7ed9\u6a21\u578b\u3002<\/li>\n
  • Handoff \u548c Agent-as-tool \u7ec4\u7ec7\u591a Agent \u5de5\u4f5c\u6d41\u3002<\/li>\n
  • Streaming \u628a\u8fd0\u884c\u8fc7\u7a0b\u53d8\u6210\u4e8b\u4ef6\u6d41\u3002<\/li>\n
  • Sessions \u4fdd\u5b58\u591a\u8f6e\u5386\u53f2\u3002<\/li>\n
  • Guardrails \u63a7\u5236\u8f93\u5165\u3001\u8f93\u51fa\u548c\u5de5\u5177\u8fb9\u754c\u3002<\/li>\n
  • Tracing \u8bb0\u5f55\u4e00\u6b21 run \u7684\u53ef\u89c2\u6d4b\u94fe\u8def\u3002<\/li>\n

    \u8fd9\u4e9b\u673a\u5236\u6700\u7ec8\u90fd\u8981\u843d\u5230\u4e00\u4e2a\u6838\u5fc3\u52a8\u4f5c\u4e0a：<\/p>\n

    \u8c03\u7528\u6a21\u578b，\u62ff\u5230\u6a21\u578b\u54cd\u5e94，\u518d\u628a\u6a21\u578b\u54cd\u5e94\u5f52\u4e00\u5316\u6210 SDK \u80fd\u5904\u7406\u7684 items\u3002<\/p>\n

    \u8fd9\u5c31\u662f\u6a21\u578b\u9002\u914d\u5c42\u7684\u804c\u8d23\u3002<\/p>\n

    \u672c\u7bc7\u91cd\u70b9\u56de\u7b54\u4e94\u4e2a\u95ee\u9898：<\/p>\n

  • SDK \u5982\u4f55\u62bd\u8c61 Model \u548c ModelProvider\u3002<\/li>\n
  • Runner \u5982\u4f55\u51b3\u5b9a\u4e00\u6b21 run \u7528\u54ea\u4e2a\u6a21\u578b\u3002<\/li>\n
  • OpenAI Responses \u548c Chat Completions \u4e24\u6761\u8def\u5f84\u6709\u4ec0\u4e48\u5dee\u5f02\u3002<\/li>\n
  • ModelSettings \u5982\u4f55\u5408\u5e76\u3001\u8f6c\u6362\u548c\u4e0b\u53d1\u3002<\/li>\n
  • LiteLLM\u3001Any-LLM\u3001OpenAI-compatible endpoint \u5982\u4f55\u63a5\u5165\u3002<\/li>\n

    \u5148\u7ed9\u7ed3\u8bba：Runner \u4e0d\u5173\u5fc3\u5177\u4f53 Provider<\/h3>\n

    Agents SDK \u7684\u6838\u5fc3 run loop \u4e0d\u76f4\u63a5\u4f9d\u8d56 OpenAI Responses\u3001Chat Completions\u3001LiteLLM \u6216 Any-LLM \u7684\u8bf7\u6c42\u7ed3\u6784\u3002<\/p>\n

    \u5b83\u4f9d\u8d56\u7684\u662f\u7edf\u4e00\u63a5\u53e3：<\/p>\n

    class<\/span> Model<\/span>(<\/span>abc.<\/span>ABC)<\/span>:<\/span>
    \n async<\/span> def<\/span> get_response<\/span>(<\/span>.<\/span>.<\/span>.<\/span>)<\/span> –<\/span>><\/span> ModelResponse:<\/span>
    \n .<\/span>.<\/span>.<\/span><\/p>\n

    def<\/span> stream_response<\/span>(<\/span>.<\/span>.<\/span>.<\/span>)<\/span> –<\/span>><\/span> AsyncIterator[<\/span>TResponseStreamEvent]<\/span>:<\/span>
    \n .<\/span>.<\/span>.<\/span><\/p>\n

    \u4e5f\u5c31\u662f\u8bf4，run loop \u53ea\u9700\u8981：<\/p>\n

  • \u628a\u672c\u8f6e\u8f93\u5165\u3001\u5de5\u5177\u3001handoffs\u3001\u8f93\u51fa schema\u3001model settings \u4ea4\u7ed9 Model\u3002<\/li>\n
  • \u62ff\u56de\u7edf\u4e00\u7684 ModelResponse \u6216\u7edf\u4e00\u7684 Responses \u98ce\u683c stream events\u3002<\/li>\n
  • \u7ee7\u7eed\u6267\u884c\u5de5\u5177\u3001handoff\u3001\u8f93\u51fa\u6821\u9a8c\u548c\u4e0b\u4e00\u8f6e\u3002<\/li>\n

    \u771f\u6b63\u77e5\u9053\u201c\u600e\u4e48\u8c03\u7528\u67d0\u4e2a\u540e\u7aef\u201d\u7684，\u662f model adapter：<\/p>\n\nAdapter\u8d1f\u8d23\u4ec0\u4e48<\/tr>\n\n\n\n\n\n\n
    OpenAIResponsesModel<\/td>\n\u8c03\u7528 OpenAI Responses API<\/td>\n<\/tr>\n
    OpenAIResponsesWSModel<\/td>\n\u901a\u8fc7 WebSocket \u8c03\u7528 OpenAI Responses API<\/td>\n<\/tr>\n
    OpenAIChatCompletionsModel<\/td>\n\u8c03\u7528 Chat Completions API，\u5e76\u505a\u683c\u5f0f\u8f6c\u6362<\/td>\n<\/tr>\n
    LitellmModel<\/td>\n\u901a\u8fc7 LiteLLM \u8c03\u7528\u7b2c\u4e09\u65b9 provider<\/td>\n<\/tr>\n
    AnyLLMModel<\/td>\n\u901a\u8fc7 any-llm \u8c03\u7528 Responses \u6216 Chat Completions \u98ce\u683c provider<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u56e0\u6b64，\u6a21\u578b\u9002\u914d\u5c42\u7684\u6838\u5fc3\u8bbe\u8ba1\u662f：<\/p>\n

    Runner \u53ea\u770b Model \u62bd\u8c61；
    \nProvider \u8d1f\u8d23\u628a\u540d\u5b57\u89e3\u6790\u6210 Model；
    \nAdapter \u8d1f\u8d23 provider-specific \u8bf7\u6c42\u3001\u80fd\u529b\u6821\u9a8c\u3001\u9519\u8bef\u548c usage \u5f52\u4e00\u5316\u3002<\/p>\n

    \u7b2c\u5341\u4e8c\u7bc7\u5173\u6ce8\u7684\u6e90\u7801\u5165\u53e3<\/h3>\n

    \u672c\u7bc7\u4e3b\u8981\u6d89\u53ca\u8fd9\u4e9b\u6587\u4ef6：<\/p>\n\n\u6587\u4ef6\u4f5c\u7528<\/tr>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
    src\/agents\/models\/interface.py<\/td>\nModel\u3001ModelProvider\u3001ModelTracing \u62bd\u8c61<\/td>\n<\/tr>\n
    src\/agents\/models\/openai_provider.py<\/td>\nOpenAI \u9ed8\u8ba4 provider<\/td>\n<\/tr>\n
    src\/agents\/models\/multi_provider.py<\/td>\n\u524d\u7f00\u8def\u7531 provider<\/td>\n<\/tr>\n
    src\/agents\/models\/openai_responses.py<\/td>\nOpenAI Responses adapter<\/td>\n<\/tr>\n
    src\/agents\/models\/openai_chatcompletions.py<\/td>\nChat Completions adapter<\/td>\n<\/tr>\n
    src\/agents\/models\/chatcmpl_converter.py<\/td>\nResponses items \u4e0e Chat messages \u8f6c\u6362<\/td>\n<\/tr>\n
    src\/agents\/model_settings.py<\/td>\nModelSettings \u5b9a\u4e49\u548c\u5408\u5e76\u903b\u8f91<\/td>\n<\/tr>\n
    src\/agents\/models\/default_models.py<\/td>\n\u9ed8\u8ba4\u6a21\u578b\u548c\u9ed8\u8ba4 settings<\/td>\n<\/tr>\n
    src\/agents\/run_internal\/turn_preparation.py<\/td>\nrun loop \u89e3\u6790\u6a21\u578b\u548c settings<\/td>\n<\/tr>\n
    src\/agents\/run_internal\/run_loop.py<\/td>\n\u8c03\u7528 model.get_response() \/ stream_response()<\/td>\n<\/tr>\n
    src\/agents\/run_internal\/model_retry.py<\/td>\nrunner-managed retry<\/td>\n<\/tr>\n
    src\/agents\/extensions\/models\/litellm_model.py<\/td>\nLiteLLM adapter<\/td>\n<\/tr>\n
    src\/agents\/extensions\/models\/any_llm_model.py<\/td>\nAny-LLM adapter<\/td>\n<\/tr>\n
    docs\/models\/index.md<\/td>\n\u6a21\u578b\u914d\u7f6e\u6587\u6863<\/td>\n<\/tr>\n
    examples\/model_providers<\/td>\n\u7b2c\u4e09\u65b9 provider \u793a\u4f8b<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u5efa\u8bae\u9605\u8bfb\u987a\u5e8f：<\/p>\n

  • \u5148\u8bfb interface.py，\u7406\u89e3 SDK \u5bf9\u6a21\u578b\u7684\u6700\u4f4e\u8981\u6c42\u3002<\/li>\n
  • \u518d\u8bfb turn_preparation.py，\u770b Runner \u5982\u4f55\u89e3\u6790\u6a21\u578b\u3002<\/li>\n
  • \u7136\u540e\u8bfb openai_provider.py，\u770b\u9ed8\u8ba4 provider \u5982\u4f55\u9009\u62e9 Responses \u6216 Chat Completions\u3002<\/li>\n
  • \u6700\u540e\u5bf9\u6bd4 openai_responses.py \u4e0e openai_chatcompletions.py\u3002<\/li>\n

    ModelTracing：\u6a21\u578b\u8c03\u7528\u7684 trace \u5f00\u5173<\/h3>\n

    ModelTracing \u662f\u6a21\u578b adapter \u63a5\u6536\u7684 tracing \u914d\u7f6e：<\/p>\n

    class<\/span> ModelTracing<\/span>(<\/span>enum.<\/span>Enum)<\/span>:<\/span>
    \n DISABLED =<\/span> 0<\/span>
    \n ENABLED =<\/span> 1<\/span>
    \n ENABLED_WITHOUT_DATA =<\/span> 2<\/span><\/p>\n

    def<\/span> is_disabled<\/span>(<\/span>self)<\/span> –<\/span>><\/span> bool<\/span>:<\/span>
    \n return<\/span> self ==<\/span> ModelTracing.<\/span>DISABLED<\/p>\n

    def<\/span> include_data<\/span>(<\/span>self)<\/span> –<\/span>><\/span> bool<\/span>:<\/span>
    \n return<\/span> self ==<\/span> ModelTracing.<\/span>ENABLED<\/p>\n

    \u5b83\u6765\u81ea RunConfig：<\/p>\n

    def<\/span> get_model_tracing_impl<\/span>(<\/span>
    \n tracing_disabled:<\/span> bool<\/span>,<\/span>
    \n trace_include_sensitive_data:<\/span> bool<\/span>,<\/span>
    \n)<\/span> –<\/span>><\/span> ModelTracing:<\/span>
    \n if<\/span> tracing_disabled:<\/span>
    \n return<\/span> ModelTracing.<\/span>DISABLED
    \n if<\/span> trace_include_sensitive_data:<\/span>
    \n return<\/span> ModelTracing.<\/span>ENABLED
    \n return<\/span> ModelTracing.<\/span>ENABLED_WITHOUT_DATA<\/p>\n

    \u542b\u4e49\u5f88\u660e\u786e：<\/p>\n\n\u914d\u7f6e\u884c\u4e3a<\/tr>\n\n\n\n\n
    tracing_disabled=True<\/td>\n\u6a21\u578b span \u7981\u7528<\/td>\n<\/tr>\n
    trace_include_sensitive_data=True<\/td>\nspan \u8bb0\u5f55\u8f93\u5165\u8f93\u51fa<\/td>\n<\/tr>\n
    trace_include_sensitive_data=False<\/td>\nspan \u5b58\u5728，\u4f46\u4e0d\u8bb0\u5f55\u8f93\u5165\u8f93\u51fa<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u8fd9\u5c31\u662f\u7b2c\u5341\u4e00\u7bc7\u63d0\u5230\u7684\u654f\u611f\u6570\u636e\u8fb9\u754c\u5728\u6a21\u578b\u5c42\u7684\u843d\u70b9\u3002<\/p>\n

    Model \u63a5\u53e3<\/h3>\n

    Model \u662f\u6240\u6709\u6a21\u578b adapter \u7684\u516c\u5171\u63a5\u53e3：<\/p>\n

    class<\/span> Model<\/span>(<\/span>abc.<\/span>ABC)<\/span>:<\/span>
    \n async<\/span> def<\/span> get_response<\/span>(<\/span>
    \n self,<\/span>
    \n system_instructions,<\/span>
    \n input<\/span>,<\/span>
    \n model_settings,<\/span>
    \n tools,<\/span>
    \n output_schema,<\/span>
    \n handoffs,<\/span>
    \n tracing,<\/span>
    \n *<\/span>,<\/span>
    \n previous_response_id,<\/span>
    \n conversation_id,<\/span>
    \n prompt,<\/span>
    \n )<\/span> –<\/span>><\/span> ModelResponse:<\/span>
    \n .<\/span>.<\/span>.<\/span><\/p>\n

    stream_response(…) \u53c2\u6570\u51e0\u4e4e\u76f8\u540c，\u53ea\u662f\u8fd4\u56de\u5f02\u6b65\u4e8b\u4ef6\u6d41：<\/p>\n

    def<\/span> stream_response<\/span>(<\/span>.<\/span>.<\/span>.<\/span>)<\/span> –<\/span>><\/span> AsyncIterator[<\/span>TResponseStreamEvent]<\/span>:<\/span>
    \n .<\/span>.<\/span>.<\/span><\/p>\n

    \u6ce8\u610f\u51e0\u4e2a\u8bbe\u8ba1\u70b9：<\/p>\n

  • input \u4f7f\u7528 Responses \u98ce\u683c\u7684 SDK canonical items\u3002<\/li>\n
  • tools \u662f SDK \u7edf\u4e00\u7684 Tool \u5bf9\u8c61\u3002<\/li>\n
  • handoffs \u4e5f\u4ea4\u7ed9 model adapter \u8f6c\u6362\u6210\u6a21\u578b\u53ef\u89c1\u5de5\u5177\u3002<\/li>\n
  • output_schema \u7531 adapter \u8f6c\u6362\u4e3a provider \u652f\u6301\u7684\u7ed3\u6784\u5316\u8f93\u51fa\u914d\u7f6e\u3002<\/li>\n
  • previous_response_id\u3001conversation_id\u3001prompt \u662f Responses \u7279\u6027，\u4f46\u63a5\u53e3\u7edf\u4e00\u66b4\u9732\u3002<\/li>\n

    \u8fd9\u610f\u5473\u7740 Chat Completions adapter \u4e5f\u4f1a\u6536\u5230\u8fd9\u4e9b\u53c2\u6570，\u4f46\u5b83\u4e0d\u4e00\u5b9a\u652f\u6301\u3002<\/p>\n

    \u5b83\u7684\u5904\u7406\u7b56\u7565\u662f：<\/p>\n

  • \u9ed8\u8ba4\u517c\u5bb9\u6a21\u5f0f\u4e0b\u8bb0\u5f55 warning \u5e76\u5ffd\u7565\u4e0d\u652f\u6301\u7684\u7279\u6027\u3002<\/li>\n
  • strict_feature_validation=True \u65f6\u76f4\u63a5\u629b UserError\u3002<\/li>\n

    ModelProvider \u63a5\u53e3<\/h3>\n

    ModelProvider \u53ea\u505a\u4e00\u4ef6\u4e8b：\u628a\u6a21\u578b\u540d\u89e3\u6790\u6210 Model \u5b9e\u4f8b\u3002<\/p>\n

    class<\/span> ModelProvider<\/span>(<\/span>abc.<\/span>ABC)<\/span>:<\/span>
    \n @abc.<\/span>abstractmethod<\/span>
    \n def<\/span> get_model<\/span>(<\/span>self,<\/span> model_name:<\/span> str<\/span> |<\/span> None<\/span>)<\/span> –<\/span>><\/span> Model:<\/span>
    \n .<\/span>.<\/span>.<\/span><\/p>\n

    async<\/span> def<\/span> aclose<\/span>(<\/span>self)<\/span> –<\/span>><\/span> None<\/span>:<\/span>
    \n return<\/span> None<\/span><\/p>\n

    \u8fd9\u4e2a\u63a5\u53e3\u7684\u8fb9\u754c\u5f88\u91cd\u8981：<\/p>\n

  • Model \u8d1f\u8d23\u4e00\u6b21\u5177\u4f53\u6a21\u578b\u8c03\u7528\u3002<\/li>\n
  • ModelProvider \u8d1f\u8d23\u6a21\u578b\u540d\u89e3\u6790\u3001\u7f13\u5b58\u3001\u8fde\u63a5\u590d\u7528\u548c\u8d44\u6e90\u5173\u95ed\u3002<\/li>\n

    \u4f8b\u5982 OpenAI WebSocket Responses \u6a21\u578b\u53ef\u80fd\u6301\u6709\u6301\u4e45\u8fde\u63a5，\u6240\u4ee5 OpenAIProvider.aclose() \u8981\u5173\u95ed\u7f13\u5b58\u7684 websocket model\u3002<\/p>\n

    \u666e\u901a HTTP Responses model \u4e0d\u9700\u8981\u5728 provider \u91cc\u7f13\u5b58\u5b9e\u4f8b\u3002<\/p>\n

    Runner \u5982\u4f55\u9009\u62e9\u6a21\u578b<\/h3>\n

    turn_preparation.py \u4e2d\u7684 get_model(…) \u662f\u6a21\u578b\u89e3\u6790\u5165\u53e3：<\/p>\n

    def<\/span> get_model<\/span>(<\/span>agent:<\/span> Agent[<\/span>Any]<\/span>,<\/span> run_config:<\/span> RunConfig)<\/span> –<\/span>><\/span> Model:<\/span>
    \n if<\/span> isinstance<\/span>(<\/span>run_config.<\/span>model,<\/span> Model)<\/span>:<\/span>
    \n return<\/span> run_config.<\/span>model
    \n elif<\/span> isinstance<\/span>(<\/span>run_config.<\/span>model,<\/span> str<\/span>)<\/span>:<\/span>
    \n return<\/span> run_config.<\/span>model_provider.<\/span>get_model(<\/span>run_config.<\/span>model)<\/span>
    \n elif<\/span> isinstance<\/span>(<\/span>agent.<\/span>model,<\/span> Model)<\/span>:<\/span>
    \n return<\/span> agent.<\/span>model<\/p>\n

    return<\/span> run_config.<\/span>model_provider.<\/span>get_model(<\/span>agent.<\/span>model)<\/span><\/p>\n

    \u4f18\u5148\u7ea7\u662f：<\/p>\n\n\u4f18\u5148\u7ea7\u6765\u6e90<\/tr>\n\n\n\n\n\n
    1<\/td>\nRunConfig.model \u662f\u5177\u4f53 Model \u5b9e\u4f8b<\/td>\n<\/tr>\n
    2<\/td>\nRunConfig.model \u662f\u5b57\u7b26\u4e32，\u901a\u8fc7 RunConfig.model_provider \u89e3\u6790<\/td>\n<\/tr>\n
    3<\/td>\nAgent.model \u662f\u5177\u4f53 Model \u5b9e\u4f8b<\/td>\n<\/tr>\n
    4<\/td>\nAgent.model \u662f\u5b57\u7b26\u4e32\u6216 None，\u901a\u8fc7 RunConfig.model_provider \u89e3\u6790<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u8fd9\u8bf4\u660e RunConfig.model \u53ef\u4ee5\u8986\u76d6 Agent \u81ea\u5df1\u7684\u6a21\u578b\u3002<\/p>\n

    \u4f8b\u5982：<\/p>\n

    agent =<\/span> Agent(<\/span>name=<\/span>"Assistant"<\/span>,<\/span> model=<\/span>"gpt-5.4-mini"<\/span>)<\/span><\/p>\n

    result =<\/span> await<\/span> Runner.<\/span>run(<\/span>
    \n agent,<\/span>
    \n "\u4f60\u597d"<\/span>,<\/span>
    \n run_config=<\/span>RunConfig(<\/span>model=<\/span>"gpt-5.6-sol"<\/span>)<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u672c\u6b21 run \u5b9e\u9645\u4f7f\u7528 gpt-5.6-sol\u3002<\/p>\n

    ModelSettings \u5982\u4f55\u89e3\u6790<\/h3>\n

    \u6a21\u578b settings \u4e0d\u662f\u7b80\u5355\u62ff Agent \u4e0a\u7684\u914d\u7f6e\u3002<\/p>\n

    \u6e90\u7801\u4e2d\u6709\u4e00\u5c42\u9ed8\u8ba4 settings \u5bf9\u9f50\u903b\u8f91：<\/p>\n

    def<\/span> get_model_settings<\/span>(<\/span>agent:<\/span> Agent[<\/span>Any]<\/span>,<\/span> run_config:<\/span> RunConfig)<\/span> –<\/span>><\/span> ModelSettings:<\/span>
    \n model_settings =<\/span> agent.<\/span>model_settings
    \n if<\/span> model_settings ==<\/span> _implicit_model_settings_for_agent(<\/span>agent)<\/span>:<\/span>
    \n model_settings =<\/span> _model_settings_for_resolved_name(<\/span>agent,<\/span> run_config)<\/span>
    \n return<\/span> model_settings.<\/span>resolve(<\/span>run_config.<\/span>model_settings)<\/span><\/p>\n

    \u542b\u4e49\u662f：<\/p>\n

  • \u5982\u679c Agent \u4f7f\u7528\u7684\u662f\u9690\u5f0f\u9ed8\u8ba4 settings，\u4e14 run-level model \u8986\u76d6\u4e86\u6a21\u578b\u540d，\u9ed8\u8ba4 settings \u8981\u8ddf\u7740\u65b0\u7684\u6a21\u578b\u540d\u91cd\u65b0\u8ba1\u7b97\u3002<\/li>\n
  • \u5982\u679c Agent \u660e\u786e\u8bbe\u7f6e\u8fc7 ModelSettings，\u5219\u4fdd\u6301\u7528\u6237\u663e\u5f0f\u914d\u7f6e\u3002<\/li>\n
  • \u6700\u540e\u7528 run_config.model_settings \u8986\u76d6 Agent settings\u3002<\/li>\n

    ModelSettings.resolve(…) \u7684\u8986\u76d6\u89c4\u5219\u662f：<\/p>\n

    changes =<\/span> {<\/span>
    \n field.<\/span>name:<\/span> getattr<\/span>(<\/span>override,<\/span> field.<\/span>name)<\/span>
    \n for<\/span> field in<\/span> fields(<\/span>self)<\/span>
    \n if<\/span> getattr<\/span>(<\/span>override,<\/span> field.<\/span>name)<\/span> is<\/span> not<\/span> None<\/span>
    \n}<\/span><\/p>\n

    \u5373\u975e None \u5b57\u6bb5\u8986\u76d6\u3002<\/p>\n

    \u4e24\u4e2a\u7279\u6b8a\u5408\u5e76：<\/p>\n

  • extra_args \u4f1a\u5408\u5e76 dict，\u800c\u4e0d\u662f\u6574\u4f53\u66ff\u6362\u3002<\/li>\n
  • retry \u4f1a\u6df1\u5ea6\u5408\u5e76 backoff \u7b49\u5b57\u6bb5\u3002<\/li>\n

    \u9ed8\u8ba4\u6a21\u578b\u548c\u9ed8\u8ba4 Settings<\/h3>\n

    \u5f53\u524d\u6e90\u7801\u4e2d\u9ed8\u8ba4\u6a21\u578b\u6765\u81ea：<\/p>\n

    OPENAI_DEFAULT_MODEL_ENV_VARIABLE_NAME =<\/span> "OPENAI_DEFAULT_MODEL"<\/span><\/p>\n

    def<\/span> get_default_model<\/span>(<\/span>)<\/span> –<\/span>><\/span> str<\/span>:<\/span>
    \n return<\/span> os.<\/span>getenv(<\/span>OPENAI_DEFAULT_MODEL_ENV_VARIABLE_NAME,<\/span> "gpt-5.4-mini"<\/span>)<\/span>.<\/span>lower(<\/span>)<\/span><\/p>\n

    \u9ed8\u8ba4\u662f：<\/p>\n

    gpt-5.4-mini<\/p>\n

    GPT-5 \u7cfb\u5217\u4f1a\u6709\u9ed8\u8ba4 ModelSettings：<\/p>\n

    _GPT_5_NONE_DEFAULT_MODEL_SETTINGS =<\/span> ModelSettings(<\/span>
    \n reasoning=<\/span>Reasoning(<\/span>effort=<\/span>"none"<\/span>)<\/span>,<\/span>
    \n verbosity=<\/span>"low"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u4f8b\u5982 gpt-5.4-mini \u9ed8\u8ba4\u662f：<\/p>\n

    ModelSettings(<\/span>
    \n reasoning=<\/span>Reasoning(<\/span>effort=<\/span>"none"<\/span>)<\/span>,<\/span>
    \n verbosity=<\/span>"low"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u975e GPT-5 \u6a21\u578b\u9ed8\u8ba4\u662f\u7a7a ModelSettings()\u3002<\/p>\n

    \u8fd9\u89e3\u91ca\u4e86\u4e00\u4e2a\u5e38\u89c1\u73b0\u8c61：<\/p>\n

    Agent(<\/span>name=<\/span>"A"<\/span>)<\/span><\/p>\n

    \u4e0d\u4f20\u6a21\u578b\u65f6，\u5e76\u4e0d\u662f\u5b8c\u5168\u6ca1\u6709\u6a21\u578b\u914d\u7f6e\u3002\u9ed8\u8ba4\u6a21\u578b\u5982\u679c\u662f GPT-5 \u7cfb\u5217，SDK \u4f1a\u7ed9\u5b83\u8865\u4e00\u7ec4\u4f4e\u5ef6\u8fdf\u9ed8\u8ba4 settings\u3002<\/p>\n

    OpenAIProvider：\u9ed8\u8ba4\u6a21\u578b Provider<\/h3>\n

    OpenAIProvider \u662f\u9ed8\u8ba4 provider\u3002<\/p>\n

    \u5b83\u7684\u6784\u9020\u53c2\u6570\u5305\u62ec：<\/p>\n

    OpenAIProvider(<\/span>
    \n api_key=<\/span>None<\/span>,<\/span>
    \n base_url=<\/span>None<\/span>,<\/span>
    \n websocket_base_url=<\/span>None<\/span>,<\/span>
    \n openai_client=<\/span>None<\/span>,<\/span>
    \n organization=<\/span>None<\/span>,<\/span>
    \n project=<\/span>None<\/span>,<\/span>
    \n use_responses=<\/span>None<\/span>,<\/span>
    \n use_responses_websocket=<\/span>None<\/span>,<\/span>
    \n strict_feature_validation=<\/span>False<\/span>,<\/span>
    \n responses_websocket_options=<\/span>None<\/span>,<\/span>
    \n buffer_streamed_tool_calls=<\/span>False<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u51e0\u4e2a\u5173\u952e\u70b9：<\/p>\n

  • \u5982\u679c\u4f20\u5165 openai_client，\u4e0d\u80fd\u518d\u4f20 api_key\u3001base_url\u3001websocket_base_url\u3002<\/li>\n
  • \u5982\u679c\u6ca1\u6709\u4f20 client，\u4f1a\u61d2\u52a0\u8f7d AsyncOpenAI\u3002<\/li>\n
  • HTTP client \u662f\u5171\u4eab\u7684，\u907f\u514d\u6bcf\u6b21\u8bf7\u6c42\u65b0\u5efa\u8fde\u63a5\u6c60\u3002<\/li>\n
  • \u9ed8\u8ba4\u4f7f\u7528 Responses API\u3002<\/li>\n
  • \u53ef\u4ee5\u5207\u6362\u5230 Chat Completions\u3002<\/li>\n
  • Responses \u8fd8\u53ef\u4ee5\u5207\u6362 HTTP \u6216 websocket transport\u3002<\/li>\n

    \u9ed8\u8ba4 client \u89e3\u6790：<\/p>\n

    self.<\/span>_client =<\/span> _openai_shared.<\/span>get_default_openai_client(<\/span>)<\/span> or<\/span> AsyncOpenAI(<\/span>
    \n api_key=<\/span>self.<\/span>_stored_api_key or<\/span> _openai_shared.<\/span>get_default_openai_key(<\/span>)<\/span>,<\/span>
    \n base_url=<\/span>self.<\/span>_stored_base_url or<\/span> os.<\/span>getenv(<\/span>"OPENAI_BASE_URL"<\/span>)<\/span>,<\/span>
    \n websocket_base_url=<\/span>(<\/span>
    \n self.<\/span>_stored_websocket_base_url or<\/span> os.<\/span>getenv(<\/span>"OPENAI_WEBSOCKET_BASE_URL"<\/span>)<\/span>
    \n )<\/span>,<\/span>
    \n http_client=<\/span>shared_http_client(<\/span>)<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u8fd9\u8bf4\u660e\u4f60\u53ef\u4ee5\u7528\u73af\u5883\u53d8\u91cf\u6216\u4ee3\u7801\u5168\u5c40\u914d\u7f6e OpenAI client\u3002<\/p>\n

    \u5207\u6362\u9ed8\u8ba4 OpenAI API<\/h3>\n

    \u9ed8\u8ba4\u662f Responses：<\/p>\n

    set_default_openai_api(<\/span>"responses"<\/span>)<\/span><\/p>\n

    \u4e5f\u53ef\u4ee5\u5207\u6362\u4e3a Chat Completions：<\/p>\n

    from<\/span> agents import<\/span> set_default_openai_api<\/p>\n

    set_default_openai_api(<\/span>"chat_completions"<\/span>)<\/span><\/p>\n

    \u5207\u6362\u540e，OpenAIProvider().get_model("…") \u4f1a\u8fd4\u56de\u4e0d\u540c adapter：<\/p>\n\n\u914d\u7f6e\u8fd4\u56de<\/tr>\n\n\n\n\n
    Responses HTTP<\/td>\nOpenAIResponsesModel<\/td>\n<\/tr>\n
    Responses WebSocket<\/td>\nOpenAIResponsesWSModel<\/td>\n<\/tr>\n
    Chat Completions<\/td>\nOpenAIChatCompletionsModel<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u5982\u679c\u53ea\u60f3\u672c\u6b21 run \u4f7f\u7528 Chat Completions，\u66f4\u63a8\u8350\u76f4\u63a5\u4f20 provider：<\/p>\n

    provider =<\/span> OpenAIProvider(<\/span>use_responses=<\/span>False<\/span>)<\/span><\/p>\n

    result =<\/span> await<\/span> Runner.<\/span>run(<\/span>
    \n agent,<\/span>
    \n "\u4f60\u597d"<\/span>,<\/span>
    \n run_config=<\/span>RunConfig(<\/span>model_provider=<\/span>provider)<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u8fd9\u6837\u4e0d\u4f1a\u5f71\u54cd\u8fdb\u7a0b\u5185\u5176\u4ed6 run\u3002<\/p>\n

    OpenAI Responses Model<\/h3>\n

    OpenAIResponsesModel \u7684\u975e\u6d41\u5f0f\u8c03\u7528\u662f：<\/p>\n

    with<\/span> response_span(<\/span>disabled=<\/span>tracing.<\/span>is_disabled(<\/span>)<\/span>)<\/span> as<\/span> span_response:<\/span>
    \n response =<\/span> await<\/span> self.<\/span>_fetch_response(<\/span>.<\/span>.<\/span>.<\/span>)<\/span><\/p>\n

    \u5b83\u6700\u7ec8\u8c03\u7528：<\/p>\n

    response =<\/span> await<\/span> client.<\/span>responses.<\/span>create(<\/span>**<\/span>create_kwargs)<\/span><\/p>\n

    _build_response_create_kwargs(…) \u8d1f\u8d23\u6784\u9020 Responses \u8bf7\u6c42\u3002<\/p>\n

    \u6838\u5fc3\u5b57\u6bb5\u5305\u62ec：<\/p>\n

    create_kwargs =<\/span> {<\/span>
    \n "previous_response_id"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "conversation"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "instructions"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "model"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "input"<\/span>:<\/span> list_input,<\/span>
    \n "include"<\/span>:<\/span> include,<\/span>
    \n "tools"<\/span>:<\/span> tools_param,<\/span>
    \n "prompt"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "temperature"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "top_p"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "truncation"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "max_output_tokens"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "tool_choice"<\/span>:<\/span> tool_choice_param,<\/span>
    \n "parallel_tool_calls"<\/span>:<\/span> parallel_tool_calls,<\/span>
    \n "text"<\/span>:<\/span> response_format,<\/span>
    \n "store"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "prompt_cache_options"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "reasoning"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "metadata"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "context_management"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n}<\/span><\/p>\n

    \u8fd9\u6761\u8def\u5f84\u5929\u7136\u652f\u6301 Responses \u7684\u80fd\u529b：<\/p>\n

  • previous_response_id<\/li>\n
  • conversation<\/li>\n
  • prompt<\/li>\n
  • Responses tool surfaces<\/li>\n
  • hosted tools<\/li>\n
  • response include<\/li>\n
  • context management<\/li>\n
  • Responses structured output<\/li>\n
  • Responses streaming terminal events<\/li>\n

    \u6240\u4ee5 OpenAI-only \u5e94\u7528\u9ed8\u8ba4\u63a8\u8350\u8d70 Responses\u3002<\/p>\n

    Responses \u5de5\u5177\u8f6c\u6362<\/h3>\n

    Responses adapter \u4f1a\u628a SDK tools \u548c handoffs \u8f6c\u6210 Responses \u5de5\u5177\u53c2\u6570\u3002<\/p>\n

    converted_tools =<\/span> Converter.<\/span>convert_tools(<\/span>
    \n tools,<\/span>
    \n handoffs,<\/span>
    \n model=<\/span>effective_computer_tool_model,<\/span>
    \n tool_choice=<\/span>model_settings.<\/span>tool_choice,<\/span>
    \n)<\/span><\/p>\n

    Converter.convert_tool_choice(…) \u652f\u6301\u7684\u9009\u62e9\u5305\u62ec：<\/p>\n

  • auto<\/li>\n
  • required<\/li>\n
  • none<\/li>\n
  • function tool name<\/li>\n
  • file_search<\/li>\n
  • web_search<\/li>\n
  • computer<\/li>\n
  • image_generation<\/li>\n
  • code_interpreter<\/li>\n
  • mcp<\/li>\n
  • MCPToolChoice<\/li>\n

    \u540c\u65f6\u5b83\u4f1a\u6821\u9a8c\u4e0d\u5408\u6cd5\u7ec4\u5408\u3002\u4f8b\u5982 deferred-loading function tools \u6ca1\u6709 ToolSearchTool() \u65f6，tool_choice="required" \u53ef\u80fd\u4f1a\u88ab\u62d2\u7edd\u3002<\/p>\n

    \u8fd9\u7c7b\u6821\u9a8c\u653e\u5728 adapter \u5c42\u662f\u5408\u7406\u7684：\u53ea\u6709 adapter \u77e5\u9053\u76ee\u6807 API \u652f\u6301\u4ec0\u4e48\u3002<\/p>\n

    Responses \u8f93\u51fa\u683c\u5f0f<\/h3>\n

    \u7ed3\u6784\u5316\u8f93\u51fa\u901a\u8fc7 text \u5b57\u6bb5\u4f20\u7ed9 Responses API：<\/p>\n

    {<\/span>
    \n "format"<\/span>:<\/span> {<\/span>
    \n "type"<\/span>:<\/span> "json_schema"<\/span>,<\/span>
    \n "name"<\/span>:<\/span> "final_output"<\/span>,<\/span>
    \n "schema"<\/span>:<\/span> output_schema.<\/span>json_schema(<\/span>)<\/span>,<\/span>
    \n "strict"<\/span>:<\/span> output_schema.<\/span>is_strict_json_schema(<\/span>)<\/span>,<\/span>
    \n }<\/span>
    \n}<\/span><\/p>\n

    \u5982\u679c\u6ca1\u6709\u7ed3\u6784\u5316\u8f93\u51fa，\u8fd4\u56de omit\u3002<\/p>\n

    \u5982\u679c\u8bbe\u7f6e\u4e86 ModelSettings.verbosity，adapter \u4f1a\u628a\u5b83\u653e\u5230 response format \u76f8\u5173\u7ed3\u6784\u4e2d：<\/p>\n

    if<\/span> model_settings.<\/span>verbosity is<\/span> not<\/span> None<\/span>:<\/span>
    \n if<\/span> response_format is<\/span> not<\/span> omit:<\/span>
    \n response_format[<\/span>"verbosity"<\/span>]<\/span> =<\/span> model_settings.<\/span>verbosity
    \n else<\/span>:<\/span>
    \n response_format =<\/span> {<\/span>"verbosity"<\/span>:<\/span> model_settings.<\/span>verbosity}<\/span><\/p>\n

    \u8fd9\u4e5f\u662f\u4e3a\u4ec0\u4e48 ModelSettings \u4e0d\u662f\u7b80\u5355\u7167\u642c\u5230 API \u53c2\u6570，\u800c\u662f\u9700\u8981 adapter \u9010\u9879\u89e3\u91ca\u3002<\/p>\n

    OpenAI Chat Completions Model<\/h3>\n

    Chat Completions adapter \u7684\u6838\u5fc3\u5de5\u4f5c\u662f\u8f6c\u6362\u3002<\/p>\n

    SDK \u5185\u90e8\u8f93\u5165\u662f Responses \u98ce\u683c items，\u4f46 Chat Completions API \u8981\u7684\u662f messages：<\/p>\n

    converted_messages =<\/span> Converter.<\/span>items_to_messages(<\/span>
    \n input<\/span>,<\/span>
    \n model=<\/span>self.<\/span>model,<\/span>
    \n base_url=<\/span>str<\/span>(<\/span>self.<\/span>_client.<\/span>base_url)<\/span>,<\/span>
    \n strict_feature_validation=<\/span>self.<\/span>_strict_feature_validation,<\/span>
    \n)<\/span><\/p>\n

    \u5982\u679c\u6709 system instructions，\u4f1a\u63d2\u5230 messages \u5f00\u5934：<\/p>\n

    if<\/span> system_instructions:<\/span>
    \n converted_messages.<\/span>insert(<\/span>
    \n 0<\/span>,<\/span>
    \n {<\/span>"content"<\/span>:<\/span> system_instructions,<\/span> "role"<\/span>:<\/span> "system"<\/span>}<\/span>,<\/span>
    \n )<\/span><\/p>\n

    \u6700\u7ec8\u8c03\u7528：<\/p>\n

    ret =<\/span> await<\/span> self.<\/span>_get_client(<\/span>)<\/span>.<\/span>chat.<\/span>completions.<\/span>create(<\/span>**<\/span>create_kwargs)<\/span><\/p>\n

    \u6838\u5fc3\u5b57\u6bb5\u5305\u62ec：<\/p>\n

    create_kwargs =<\/span> {<\/span>
    \n "model"<\/span>:<\/span> self.<\/span>model,<\/span>
    \n "messages"<\/span>:<\/span> converted_messages,<\/span>
    \n "tools"<\/span>:<\/span> tools_param,<\/span>
    \n "temperature"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "top_p"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "frequency_penalty"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "presence_penalty"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "max_tokens"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "tool_choice"<\/span>:<\/span> tool_choice,<\/span>
    \n "response_format"<\/span>:<\/span> response_format,<\/span>
    \n "parallel_tool_calls"<\/span>:<\/span> parallel_tool_calls,<\/span>
    \n "stream"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "stream_options"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "store"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "reasoning_effort"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "verbosity"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n "top_logprobs"<\/span>:<\/span> .<\/span>.<\/span>.<\/span>,<\/span>
    \n}<\/span><\/p>\n

    Chat Completions \u8def\u5f84\u4f1a\u628a\u8fd4\u56de\u7684 ChatCompletionMessage \u518d\u8f6c\u6362\u56de Responses \u98ce\u683c output items：<\/p>\n

    items =<\/span> Converter.<\/span>message_to_output_items(<\/span>message)<\/span><\/p>\n

    \u8fd9\u4fdd\u8bc1 run loop \u540e\u7eed\u4ecd\u7136\u5904\u7406\u7edf\u4e00\u7684 SDK item\u3002<\/p>\n

    Chat Completions \u4e0d\u652f\u6301\u7684\u80fd\u529b<\/h3>\n

    Chat Completions adapter \u4f1a\u6536\u5230\u63a5\u53e3\u4e2d\u7684 Responses-only \u53c2\u6570，\u4f8b\u5982：<\/p>\n

  • previous_response_id<\/li>\n
  • conversation_id<\/li>\n
  • prompt<\/li>\n
  • reasoning.mode<\/li>\n
  • reasoning.context<\/li>\n
  • \u4e00\u4e9b Responses-only tool surface<\/li>\n

    \u9ed8\u8ba4\u517c\u5bb9\u6a21\u5f0f\u4e0b，\u5b83\u4f1a warning \u5e76\u5ffd\u7565\u3002<\/p>\n

    \u5982\u679c\u542f\u7528 strict validation：<\/p>\n

    provider =<\/span> OpenAIProvider(<\/span>
    \n use_responses=<\/span>False<\/span>,<\/span>
    \n strict_feature_validation=<\/span>True<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u8fd9\u4e9b\u4e0d\u5339\u914d\u4f1a\u53d8\u6210 UserError\u3002<\/p>\n

    \u5f00\u53d1\u671f\u5efa\u8bae\u6253\u5f00 strict validation，\u907f\u514d\u67d0\u4e9b\u5b57\u6bb5\u88ab\u9759\u9ed8\u964d\u7ea7\u3002<\/p>\n

    \u751f\u4ea7\u671f\u662f\u5426\u6253\u5f00，\u8981\u770b\u4f60\u662f\u5426\u4f9d\u8d56\u517c\u5bb9\u8001 provider \u7684\u5bb9\u5fcd\u884c\u4e3a\u3002<\/p>\n

    Responses \u4e0e Chat Completions \u5bf9\u6bd4<\/h3>\n\n\u7ef4\u5ea6ResponsesChat Completions<\/tr>\n\n\n\n\n\n\n\n\n\n\n\n
    SDK \u9ed8\u8ba4\u63a8\u8350<\/td>\n\u662f<\/td>\n\u5426<\/td>\n<\/tr>\n
    OpenAI hosted tools<\/td>\n\u652f\u6301\u66f4\u591a<\/td>\n\u652f\u6301\u6709\u9650<\/td>\n<\/tr>\n
    previous_response_id<\/td>\n\u652f\u6301<\/td>\n\u4e0d\u652f\u6301<\/td>\n<\/tr>\n
    conversation_id<\/td>\n\u652f\u6301<\/td>\n\u4e0d\u652f\u6301<\/td>\n<\/tr>\n
    prompt-managed request<\/td>\n\u652f\u6301<\/td>\n\u4e0d\u652f\u6301<\/td>\n<\/tr>\n
    Responses tool search<\/td>\n\u652f\u6301<\/td>\n\u4e0d\u652f\u6301<\/td>\n<\/tr>\n
    \u8f93\u5165\u683c\u5f0f<\/td>\nResponses items<\/td>\nchat messages<\/td>\n<\/tr>\n
    SDK \u8f6c\u6362\u6210\u672c<\/td>\n\u4f4e<\/td>\n\u9ad8<\/td>\n<\/tr>\n
    \u7b2c\u4e09\u65b9 provider \u517c\u5bb9\u6027<\/td>\n\u53d6\u51b3\u4e8e provider<\/td>\n\u901a\u5e38\u66f4\u597d<\/td>\n<\/tr>\n
    stream terminal \u8bed\u4e49<\/td>\n\u539f\u751f Responses events<\/td>\n\u9700\u8981\u8f6c\u6362\u6210 Responses \u98ce\u683c events<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u5b9e\u7528\u5efa\u8bae：<\/p>\n

  • OpenAI-only \u65b0\u9879\u76ee\u4f18\u5148 Responses\u3002<\/li>\n
  • OpenAI-compatible \u7b2c\u4e09\u65b9 provider \u901a\u5e38\u5148\u8bd5 Chat Completions\u3002<\/li>\n
  • \u5982\u679c\u4f9d\u8d56 server-managed conversation\u3001Responses hosted tools\u3001tool search，\u7528 Responses\u3002<\/li>\n
  • \u5982\u679c\u5fc5\u987b\u6df7\u7528\u4e24\u79cd adapter，\u9010\u9879\u9a8c\u8bc1\u5de5\u5177\u3001\u7ed3\u6784\u5316\u8f93\u51fa\u3001streaming \u548c retry \u884c\u4e3a\u3002<\/li>\n

    Streaming \u7ec8\u6b62\u8bed\u4e49<\/h3>\n

    Responses stream \u7684\u7ec8\u6b62\u4e8b\u4ef6\u4e0d\u662f\u7b80\u5355\u7684\u201c\u6d41\u7ed3\u675f\u5c31\u6210\u529f\u201d\u3002<\/p>\n

    Responses adapter \u4f1a\u68c0\u67e5：<\/p>\n

  • response.completed<\/li>\n
  • response.failed<\/li>\n
  • response.incomplete<\/li>\n
  • error<\/li>\n
  • response.error<\/li>\n

    \u5982\u679c\u6536\u5230 failed \u6216 incomplete terminal，\u4f1a\u8f6c\u6362\u6210 ModelBehaviorError\u3002<\/p>\n

    Chat Completions stream \u5219\u9700\u8981 ChatCmplStreamHandler \u628a chat chunks \u8f6c\u6362\u6210 Responses \u98ce\u683c events\u3002<\/p>\n

    \u8fd9\u4e5f\u662f adapter \u5c42\u7684\u804c\u8d23：\u4e0d\u540c provider \u7684 stream \u534f\u8bae\u4e0d\u540c，\u4f46 run loop \u53ea\u6d88\u8d39\u7edf\u4e00\u7684 TResponseStreamEvent\u3002<\/p>\n

    ModelResponse：\u7edf\u4e00\u8fd4\u56de\u503c<\/h3>\n

    \u65e0\u8bba\u5e95\u5c42\u662f Responses \u8fd8\u662f Chat Completions，\u6700\u7ec8\u90fd\u8981\u8fd4\u56de：<\/p>\n

    ModelResponse(<\/span>
    \n output=<\/span>response.<\/span>output,<\/span>
    \n usage=<\/span>usage,<\/span>
    \n response_id=<\/span>response.<\/span>id<\/span>,<\/span>
    \n request_id=<\/span>getattr<\/span>(<\/span>response,<\/span> "_request_id"<\/span>,<\/span> None<\/span>)<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u6216\u7531 Chat Completions \u8f6c\u6362\u540e\u7684：<\/p>\n

    ModelResponse(<\/span>
    \n output=<\/span>items,<\/span>
    \n usage=<\/span>usage,<\/span>
    \n response_id=<\/span>None<\/span>,<\/span>
    \n)<\/span><\/p>\n

    output \u662f SDK \u540e\u7eed\u5904\u7406\u7684\u5173\u952e：<\/p>\n

  • message output \u8fdb\u5165\u6700\u7ec8\u8f93\u51fa\u5224\u65ad\u3002<\/li>\n
  • function call \u8fdb\u5165\u5de5\u5177\u6267\u884c\u3002<\/li>\n
  • handoff tool call \u8fdb\u5165 Agent \u5207\u6362\u3002<\/li>\n
  • reasoning item \u8fdb\u5165\u5386\u53f2\u548c\u8ffd\u8e2a\u3002<\/li>\n

    \u53ea\u8981 adapter \u5f52\u4e00\u5316\u6b63\u786e，run loop \u4e0d\u9700\u8981\u77e5\u9053\u5e95\u5c42 provider\u3002<\/p>\n

    ModelSettings \u5230\u8bf7\u6c42\u53c2\u6570\u7684\u6620\u5c04<\/h3>\n

    ModelSettings \u662f SDK \u7edf\u4e00\u6a21\u578b\u914d\u7f6e：<\/p>\n\n\u5b57\u6bb5\u5e38\u89c1\u7528\u9014<\/tr>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
    temperature<\/td>\n\u91c7\u6837\u6e29\u5ea6<\/td>\n<\/tr>\n
    top_p<\/td>\nnucleus sampling<\/td>\n<\/tr>\n
    frequency_penalty<\/td>\n\u9891\u7387\u60e9\u7f5a<\/td>\n<\/tr>\n
    presence_penalty<\/td>\n\u5b58\u5728\u60e9\u7f5a<\/td>\n<\/tr>\n
    tool_choice<\/td>\n\u5de5\u5177\u9009\u62e9<\/td>\n<\/tr>\n
    parallel_tool_calls<\/td>\n\u662f\u5426\u5141\u8bb8\u5e76\u884c\u5de5\u5177\u8c03\u7528<\/td>\n<\/tr>\n
    truncation<\/td>\nResponses \u622a\u65ad\u7b56\u7565<\/td>\n<\/tr>\n
    max_tokens<\/td>\n\u6700\u5927\u8f93\u51fa token<\/td>\n<\/tr>\n
    reasoning<\/td>\n\u63a8\u7406\u6a21\u578b\u914d\u7f6e<\/td>\n<\/tr>\n
    verbosity<\/td>\n\u8f93\u51fa\u8be6\u7565<\/td>\n<\/tr>\n
    metadata<\/td>\nprovider metadata<\/td>\n<\/tr>\n
    store<\/td>\n\u662f\u5426\u670d\u52a1\u7aef\u5b58\u50a8\u54cd\u5e94<\/td>\n<\/tr>\n
    response_include<\/td>\nResponses include<\/td>\n<\/tr>\n
    top_logprobs<\/td>\ntop token logprobs<\/td>\n<\/tr>\n
    retry<\/td>\nrunner-managed retry<\/td>\n<\/tr>\n
    context_management<\/td>\nResponses context management<\/td>\n<\/tr>\n
    prompt_cache_options<\/td>\nprompt cache \u914d\u7f6e<\/td>\n<\/tr>\n
    extra_args<\/td>\n\u76f4\u63a5\u4f20\u7ed9 provider \u7684\u989d\u5916\u53c2\u6570<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u4f46\u4e0d\u662f\u6bcf\u4e2a adapter \u90fd\u652f\u6301\u6240\u6709\u5b57\u6bb5\u3002<\/p>\n

    \u4f8b\u5982：<\/p>\n

  • Responses \u4f7f\u7528 max_output_tokens\u3002<\/li>\n
  • Chat Completions \u4f7f\u7528 max_tokens\u3002<\/li>\n
  • Responses \u4f7f\u7528 text \u8868\u793a\u7ed3\u6784\u5316\u8f93\u51fa\u3002<\/li>\n
  • Chat Completions \u4f7f\u7528 response_format\u3002<\/li>\n
  • Chat Completions \u53ea\u4f7f\u7528 reasoning.effort，\u4e0d\u652f\u6301 reasoning.mode \u548c reasoning.context\u3002<\/li>\n

    \u6240\u4ee5 ModelSettings \u662f\u7edf\u4e00\u5165\u53e3，\u4e0d\u662f\u7edf\u4e00 wire format\u3002<\/p>\n

    extra_args \u7684\u91cd\u590d\u952e\u4fdd\u62a4<\/h3>\n

    \u4e24\u4e2a OpenAI adapter \u90fd\u4f1a\u68c0\u67e5 extra_args \u4e0e\u663e\u5f0f\u53c2\u6570\u662f\u5426\u91cd\u590d\u3002<\/p>\n

    \u4f8b\u5982 Responses：<\/p>\n

    duplicate_extra_arg_keys =<\/span> sorted<\/span>(<\/span>
    \n k
    \n for<\/span> k in<\/span> extra_args
    \n if<\/span> k in<\/span> create_kwargs and<\/span> not<\/span> _is_openai_omitted_value(<\/span>create_kwargs[<\/span>k]<\/span>)<\/span>
    \n)<\/span><\/p>\n

    \u5982\u679c\u91cd\u590d，\u4f1a\u629b：<\/p>\n

    responses.create() got multiple values for keyword argument '…'<\/p>\n

    \u8fd9\u53ef\u4ee5\u907f\u514d\u540c\u4e00\u4e2a\u5b57\u6bb5\u65e2\u901a\u8fc7 ModelSettings.temperature \u4f20，\u53c8\u901a\u8fc7 extra_args={"temperature": …} \u4f20，\u9020\u6210\u884c\u4e3a\u4e0d\u786e\u5b9a\u3002<\/p>\n

    \u5efa\u8bae：<\/p>\n

  • SDK \u5df2\u6709\u5b57\u6bb5，\u4f18\u5148\u7528 ModelSettings \u76f4\u63a5\u5b57\u6bb5\u3002<\/li>\n
  • SDK \u6682\u672a\u66b4\u9732\u7684\u65b0 provider \u53c2\u6570，\u518d\u7528 extra_args\u3002<\/li>\n
  • \u4e0d\u8981\u540c\u65f6\u4f7f\u7528\u4e24\u6761\u8def\u5f84\u4f20\u540c\u4e00\u4e2a\u53c2\u6570\u3002<\/li>\n

    Runner-managed Retry<\/h3>\n

    \u6a21\u578b\u91cd\u8bd5\u4e0d\u662f\u9ed8\u8ba4\u5f00\u542f\u7684\u3002<\/p>\n

    \u9700\u8981\u663e\u5f0f\u8bbe\u7f6e：<\/p>\n

    from<\/span> agents import<\/span> ModelRetrySettings,<\/span> ModelSettings,<\/span> retry_policies<\/p>\n

    model_settings =<\/span> ModelSettings(<\/span>
    \n retry=<\/span>ModelRetrySettings(<\/span>
    \n max_retries=<\/span>2<\/span>,<\/span>
    \n policy=<\/span>retry_policies.<\/span>network_error(<\/span>)<\/span>,<\/span>
    \n )<\/span>
    \n)<\/span><\/p>\n

    run loop \u8c03\u7528\u6a21\u578b\u65f6\u4f1a\u5305\u4e00\u5c42：<\/p>\n

    new_response =<\/span> await<\/span> get_response_with_retry(<\/span>
    \n get_response=<\/span>lambda<\/span>:<\/span> model.<\/span>get_response(<\/span>.<\/span>.<\/span>.<\/span>)<\/span>,<\/span>
    \n rewind=<\/span>rewind_model_request,<\/span>
    \n retry_settings=<\/span>model_settings.<\/span>retry,<\/span>
    \n get_retry_advice=<\/span>model.<\/span>get_retry_advice,<\/span>
    \n previous_response_id=<\/span>previous_response_id,<\/span>
    \n conversation_id=<\/span>conversation_id,<\/span>
    \n)<\/span><\/p>\n

    \u91cd\u8bd5\u9700\u8981\u8003\u8651 replay safety\u3002<\/p>\n

    \u7279\u522b\u662f：<\/p>\n

  • \u4f7f\u7528 previous_response_id\u3002<\/li>\n
  • \u4f7f\u7528 conversation_id\u3002<\/li>\n
  • \u6d41\u5f0f\u54cd\u5e94\u5df2\u7ecf\u5f00\u59cb\u8f93\u51fa\u3002<\/li>\n
  • \u53ef\u80fd\u5df2\u7ecf\u53d1\u751f\u5de5\u5177\u526f\u4f5c\u7528\u3002<\/li>\n

    \u8fd9\u4e9b\u573a\u666f\u4e0d\u80fd\u7b80\u5355\u6309 HTTP 500 \u6216\u7f51\u7edc\u9519\u8bef\u91cd\u653e\u3002<\/p>\n

    Adapter \u53ef\u4ee5\u901a\u8fc7 get_retry_advice(…) \u63d0\u4f9b provider-specific \u5efa\u8bae\u3002OpenAI\u3001LiteLLM\u3001Any-LLM adapter \u90fd\u4f1a\u590d\u7528 OpenAI \u98ce\u683c retry advice \u63d0\u53d6\u903b\u8f91\u3002<\/p>\n

    Provider-managed retry \u4e0e Runner retry<\/h3>\n

    OpenAI SDK\u3001LiteLLM\u3001any-llm \u81ea\u8eab\u4e5f\u53ef\u80fd\u6709\u91cd\u8bd5\u673a\u5236\u3002<\/p>\n

    Runner-managed retry \u5728\u91cd\u653e\u65f6\u4f1a\u901a\u8fc7 contextvar \u7981\u7528 provider-managed retries：<\/p>\n

    provider_managed_retries_disabled(<\/span>disabled=<\/span>True<\/span>)<\/span><\/p>\n

    adapter \u91cc\u4f1a\u68c0\u67e5：<\/p>\n

    if<\/span> should_disable_provider_managed_retries(<\/span>)<\/span>:<\/span>
    \n return<\/span> client.<\/span>with_options(<\/span>max_retries=<\/span>0<\/span>)<\/span><\/p>\n

    \u8fd9\u6837\u907f\u514d\u4e24\u5c42 retry \u53e0\u52a0，\u5bfc\u81f4\u8bf7\u6c42\u6b21\u6570\u96be\u4ee5\u63a7\u5236\u3002<\/p>\n

    \u6587\u7ae0\u91cc\u7684\u5b9e\u8df5\u5efa\u8bae\u662f：\u5982\u679c\u4f60\u9700\u8981\u53ef\u89e3\u91ca\u3001\u53ef\u5ba1\u8ba1\u7684\u91cd\u8bd5\u7b56\u7565，\u4f18\u5148\u4f7f\u7528 runner-managed retry，\u5e76\u660e\u786e policy\u3002<\/p>\n

    MultiProvider：\u57fa\u4e8e\u524d\u7f00\u7684\u8def\u7531<\/h3>\n

    MultiProvider \u662f\u9ed8\u8ba4 RunConfig.model_provider\u3002<\/p>\n

    \u5b83\u652f\u6301\u6309\u6a21\u578b\u540d\u524d\u7f00\u8def\u7531：<\/p>\n\n\u6a21\u578b\u540d\u9ed8\u8ba4\u8def\u7531<\/tr>\n\n\n\n\n\n
    gpt-5.4-mini<\/td>\nOpenAI provider<\/td>\n<\/tr>\n
    openai\/gpt-4.1<\/td>\nOpenAI provider，\u9ed8\u8ba4\u53bb\u6389 openai\/<\/td>\n<\/tr>\n
    litellm\/openrouter\/openai\/gpt-5.4-mini<\/td>\nLitellmProvider<\/td>\n<\/tr>\n
    any-llm\/openrouter\/openai\/gpt-5.4-mini<\/td>\nAnyLLMProvider<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u6e90\u7801\u4e2d\u9ed8\u8ba4\u8bf4\u660e：<\/p>\n

    class<\/span> MultiProvider<\/span>(<\/span>ModelProvider)<\/span>:<\/span>
    \n """Maps model based on prefix."""<\/span><\/p>\n

    \u5982\u679c\u9047\u5230\u672a\u77e5\u524d\u7f00，\u9ed8\u8ba4\u629b UserError\u3002<\/p>\n

    \u5982\u679c\u4f60\u4f7f\u7528 OpenAI-compatible endpoint，\u5e76\u4e14\u540e\u7aef\u671f\u671b\u5b8c\u6574 namespaced model id，\u53ef\u4ee5\u914d\u7f6e：<\/p>\n

    provider =<\/span> MultiProvider(<\/span>
    \n openai_base_url=<\/span>"https:\/\/openrouter.ai\/api\/v1"<\/span>,<\/span>
    \n openai_api_key=<\/span>"…"<\/span>,<\/span>
    \n openai_prefix_mode=<\/span>"model_id"<\/span>,<\/span>
    \n unknown_prefix_mode=<\/span>"model_id"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u8fd9\u6837 openai\/gpt-4.1 \u4e0d\u4f1a\u88ab\u5f53\u6210 OpenAI alias \u53bb\u6389\u524d\u7f00\u3002<\/p>\n

    OpenAI-compatible endpoint<\/h3>\n

    \u6700\u5c0f\u63a5\u5165\u65b9\u5f0f\u662f\u81ea\u5b9a\u4e49 AsyncOpenAI：<\/p>\n

    from<\/span> openai import<\/span> AsyncOpenAI
    \nfrom<\/span> agents import<\/span> OpenAIChatCompletionsModel<\/p>\n

    client =<\/span> AsyncOpenAI(<\/span>
    \n api_key=<\/span>"provider_api_key"<\/span>,<\/span>
    \n base_url=<\/span>"https:\/\/provider.example.com\/v1"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    model =<\/span> OpenAIChatCompletionsModel(<\/span>
    \n model=<\/span>"provider-model-name"<\/span>,<\/span>
    \n openai_client=<\/span>client,<\/span>
    \n)<\/span><\/p>\n

    \u7136\u540e\u653e\u5230 Agent：<\/p>\n

    agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Assistant"<\/span>,<\/span>
    \n instructions=<\/span>"\u4f60\u662f\u4e00\u4e2a\u7b80\u6d01\u52a9\u624b\u3002"<\/span>,<\/span>
    \n model=<\/span>model,<\/span>
    \n)<\/span><\/p>\n

    \u5982\u679c\u8981\u6309 run \u751f\u6548，\u53ef\u4ee5\u5199 ModelProvider：<\/p>\n

    class<\/span> CustomModelProvider<\/span>(<\/span>ModelProvider)<\/span>:<\/span>
    \n def<\/span> get_model<\/span>(<\/span>self,<\/span> model_name:<\/span> str<\/span> |<\/span> None<\/span>)<\/span> –<\/span>><\/span> Model:<\/span>
    \n return<\/span> OpenAIChatCompletionsModel(<\/span>
    \n model=<\/span>model_name or<\/span> "provider-model-name"<\/span>,<\/span>
    \n openai_client=<\/span>client,<\/span>
    \n )<\/span><\/p>\n

    \u518d\u4f20\u5165：<\/p>\n

    result =<\/span> await<\/span> Runner.<\/span>run(<\/span>
    \n agent,<\/span>
    \n "\u4f60\u597d"<\/span>,<\/span>
    \n run_config=<\/span>RunConfig(<\/span>model_provider=<\/span>CustomModelProvider(<\/span>)<\/span>)<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u8fd9\u9002\u5408\u4e00\u4e2a run \u5185\u6240\u6709 Agent \u90fd\u8d70\u540c\u4e00\u4e2a\u81ea\u5b9a\u4e49 provider \u7684\u573a\u666f\u3002<\/p>\n

    LiteLLM adapter<\/h3>\n

    LitellmModel \u901a\u8fc7 LiteLLM \u8c03\u7528\u7b2c\u4e09\u65b9 provider\u3002<\/p>\n

    \u76f4\u63a5\u4f7f\u7528：<\/p>\n

    from<\/span> agents.<\/span>extensions.<\/span>models.<\/span>litellm_model import<\/span> LitellmModel<\/p>\n

    agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Assistant"<\/span>,<\/span>
    \n model=<\/span>LitellmModel(<\/span>
    \n model=<\/span>"openrouter\/openai\/gpt-5.4-mini"<\/span>,<\/span>
    \n api_key=<\/span>"…"<\/span>,<\/span>
    \n )<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u6216\u4f7f\u7528 MultiProvider \u524d\u7f00：<\/p>\n

    agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Assistant"<\/span>,<\/span>
    \n model=<\/span>"litellm\/openrouter\/openai\/gpt-5.4-mini"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    LiteLLM adapter \u57fa\u672c\u8d70 Chat Completions \u98ce\u683c：<\/p>\n

  • \u628a SDK items \u8f6c\u6210 chat messages\u3002<\/li>\n
  • \u628a SDK tools \u8f6c\u6210 chat completion tools\u3002<\/li>\n
  • \u8c03\u7528 litellm.acompletion(…)\u3002<\/li>\n
  • \u628a\u8fd4\u56de message \u8f6c\u6210 SDK output items\u3002<\/li>\n
  • \u4f7f\u7528 generation_span \u505a tracing\u3002<\/li>\n

    \u6e90\u7801\u4e2d\u8c03\u7528\u5f62\u6001：<\/p>\n

    ret =<\/span> await<\/span> litellm.<\/span>acompletion(<\/span>
    \n model=<\/span>self.<\/span>model,<\/span>
    \n messages=<\/span>converted_messages,<\/span>
    \n tools=<\/span>converted_tools or<\/span> None<\/span>,<\/span>
    \n temperature=<\/span>model_settings.<\/span>temperature,<\/span>
    \n tool_choice=<\/span>self.<\/span>_remove_not_given(<\/span>tool_choice)<\/span>,<\/span>
    \n response_format=<\/span>self.<\/span>_remove_not_given(<\/span>response_format)<\/span>,<\/span>
    \n parallel_tool_calls=<\/span>parallel_tool_calls,<\/span>
    \n api_key=<\/span>self.<\/span>api_key,<\/span>
    \n base_url=<\/span>self.<\/span>base_url,<\/span>
    \n **<\/span>extra_kwargs,<\/span>
    \n)<\/span><\/p>\n

    \u9002\u5408：<\/p>\n

  • \u9700\u8981\u63a5\u5165\u591a\u4e2a\u7b2c\u4e09\u65b9 provider\u3002<\/li>\n
  • provider \u5df2\u88ab LiteLLM \u652f\u6301\u3002<\/li>\n
  • \u80fd\u63a5\u53d7 Chat Completions \u98ce\u683c\u7684\u80fd\u529b\u8fb9\u754c\u3002<\/li>\n

    Any-LLM adapter<\/h3>\n

    AnyLLMModel \u7684\u7279\u70b9\u662f\u53ef\u4ee5\u6309 provider \u80fd\u529b\u9009\u62e9 Responses \u6216 Chat Completions：<\/p>\n

    def<\/span> _selected_api<\/span>(<\/span>self)<\/span> –<\/span>><\/span> Literal[<\/span>"responses"<\/span>,<\/span> "chat_completions"<\/span>]<\/span>:<\/span>
    \n if<\/span> self.<\/span>api is<\/span> not<\/span> None<\/span>:<\/span>
    \n .<\/span>.<\/span>.<\/span>
    \n return<\/span> "responses"<\/span> if<\/span> self.<\/span>_supports_responses(<\/span>)<\/span> else<\/span> "chat_completions"<\/span><\/p>\n

    \u76f4\u63a5\u4f7f\u7528：<\/p>\n

    from<\/span> agents.<\/span>extensions.<\/span>models.<\/span>any_llm_model import<\/span> AnyLLMModel<\/p>\n

    agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Assistant"<\/span>,<\/span>
    \n model=<\/span>AnyLLMModel(<\/span>
    \n model=<\/span>"openrouter\/openai\/gpt-5.4-mini"<\/span>,<\/span>
    \n api_key=<\/span>"…"<\/span>,<\/span>
    \n )<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u6216\u4f7f\u7528 MultiProvider \u524d\u7f00：<\/p>\n

    agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Assistant"<\/span>,<\/span>
    \n model=<\/span>"any-llm\/openrouter\/openai\/gpt-5.4-mini"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u5982\u679c provider \u652f\u6301 Responses，\u5b83\u4f1a\u8d70 _get_response_via_responses(…)；\u5426\u5219\u8d70 _get_response_via_chat(…)\u3002<\/p>\n

    \u8fd9\u6bd4 LiteLLM \u591a\u4e86\u4e00\u4e2a\u9009\u62e9\u5c42：<\/p>\n\nAdapterAPI \u98ce\u683c<\/tr>\n\n\n\n
    LiteLLM<\/td>\nChat Completions \u98ce\u683c<\/td>\n<\/tr>\n
    Any-LLM<\/td>\nResponses \u6216 Chat Completions，\u53d6\u51b3\u4e8e provider \u80fd\u529b<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u4f46\u9009\u62e9\u66f4\u591a\u4e5f\u610f\u5473\u7740\u66f4\u8981\u6d4b\u8bd5 provider \u80fd\u529b\u5dee\u5f02\u3002<\/p>\n

    \u975e OpenAI Provider \u4e0e Tracing<\/h3>\n

    \u7b2c\u4e09\u65b9 provider \u793a\u4f8b\u91cc\u901a\u5e38\u4f1a\u8c03\u7528：<\/p>\n

    set_tracing_disabled(<\/span>disabled=<\/span>True<\/span>)<\/span><\/p>\n

    \u539f\u56e0\u662f\u9ed8\u8ba4 tracing \u4f1a\u4e0a\u4f20\u5230 OpenAI traces backend，\u9700\u8981 OpenAI API key\u3002<\/p>\n

    \u5982\u679c\u4f60\u7684\u6a21\u578b provider \u4e0d\u662f OpenAI，\u4f46\u4ecd\u60f3\u4f7f\u7528 OpenAI tracing，\u53ef\u4ee5\u5355\u72ec\u8bbe\u7f6e tracing export key：<\/p>\n

    from<\/span> agents import<\/span> set_tracing_export_api_key<\/p>\n

    set_tracing_export_api_key(<\/span>"sk-…"<\/span>)<\/span><\/p>\n

    \u6216\u8005\u63a5\u5165\u81ea\u5b9a\u4e49 tracing processor\u3002<\/p>\n

    \u751f\u4ea7\u5efa\u8bae：<\/p>\n

  • \u975e OpenAI provider \u9ed8\u8ba4\u5148\u5173\u95ed OpenAI tracing\u3002<\/li>\n
  • \u5982\u9700 trace，\u63a5\u5165\u4f01\u4e1a\u5185\u90e8 tracing processor\u3002<\/li>\n
  • \u4e0d\u8981\u628a\u7b2c\u4e09\u65b9 provider \u7684 API key \u5199\u8fdb trace metadata\u3002<\/li>\n

    \u6df7\u7528\u6a21\u578b\u65f6\u7684\u8fb9\u754c<\/h3>\n

    Agents SDK \u5141\u8bb8\u4e00\u4e2a workflow \u4e2d\u4e0d\u540c Agent \u4f7f\u7528\u4e0d\u540c\u6a21\u578b\u3002<\/p>\n

    \u4f8b\u5982：<\/p>\n

    triage_agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Triage agent"<\/span>,<\/span>
    \n model=<\/span>"gpt-5.4-mini"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    expert_agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Expert agent"<\/span>,<\/span>
    \n model=<\/span>OpenAIChatCompletionsModel(<\/span>
    \n model=<\/span>"provider-model"<\/span>,<\/span>
    \n openai_client=<\/span>custom_client,<\/span>
    \n )<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u4f46\u6df7\u7528\u6a21\u578b\u8981\u6ce8\u610f：<\/p>\n

  • \u5de5\u5177\u80fd\u529b\u662f\u5426\u4e00\u81f4\u3002<\/li>\n
  • handoff \u662f\u5426\u80fd\u88ab\u76ee\u6807 adapter \u8f6c\u6210\u5de5\u5177\u3002<\/li>\n
  • output schema \u662f\u5426\u88ab\u76ee\u6807 provider \u652f\u6301\u3002<\/li>\n
  • reasoning items \u662f\u5426\u80fd\u6b63\u786e replay\u3002<\/li>\n
  • stream event \u662f\u5426\u80fd\u5f52\u4e00\u5316\u3002<\/li>\n
  • usage \u662f\u5426\u5b8c\u6574\u3002<\/li>\n
  • retry \u662f\u5426 replay-safe\u3002<\/li>\n

    \u6587\u6863\u4e5f\u5efa\u8bae：\u540c\u4e00\u4e2a workflow \u5c3d\u91cf\u4f7f\u7528\u540c\u4e00\u79cd model shape\u3002\u53ea\u6709\u786e\u5b9e\u9700\u8981\u65f6\u518d\u6df7\u7528\u3002<\/p>\n

    \u5b9e\u8df5\u4e00：\u663e\u5f0f\u4f7f\u7528 OpenAI Responses<\/h3>\n

    \u9ed8\u8ba4\u60c5\u51b5\u4e0b OpenAIProvider \u5df2\u7ecf\u8d70 Responses\u3002<\/p>\n

    \u4f60\u4ecd\u7136\u53ef\u4ee5\u663e\u5f0f\u5199\u6e05\u695a：<\/p>\n

    from<\/span> agents import<\/span> Agent,<\/span> OpenAIProvider,<\/span> RunConfig,<\/span> Runner<\/p>\n

    agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Assistant"<\/span>,<\/span>
    \n instructions=<\/span>"\u4f60\u662f\u4e00\u4e2a\u7b80\u6d01\u52a9\u624b\u3002"<\/span>,<\/span>
    \n model=<\/span>"gpt-5.4-mini"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    provider =<\/span> OpenAIProvider(<\/span>use_responses=<\/span>True<\/span>)<\/span><\/p>\n

    result =<\/span> await<\/span> Runner.<\/span>run(<\/span>
    \n agent,<\/span>
    \n "\u7528\u4e00\u53e5\u8bdd\u89e3\u91ca Responses API\u3002"<\/span>,<\/span>
    \n run_config=<\/span>RunConfig(<\/span>model_provider=<\/span>provider)<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u5982\u679c\u8981\u542f\u7528 websocket transport：<\/p>\n

    provider =<\/span> OpenAIProvider(<\/span>
    \n use_responses=<\/span>True<\/span>,<\/span>
    \n use_responses_websocket=<\/span>True<\/span>,<\/span>
    \n)<\/span><\/p>\n

    WebSocket transport \u9002\u5408\u9700\u8981\u590d\u7528\u8fde\u63a5\u6216\u7279\u5b9a\u4f4e\u5ef6\u8fdf\u573a\u666f，\u4f46\u8981\u6ce8\u610f\u8fde\u63a5\u751f\u547d\u5468\u671f\u548c keepalive \u914d\u7f6e\u3002<\/p>\n

    \u5b9e\u8df5\u4e8c：\u5207\u6362\u5230 Chat Completions<\/h3>\n

    \u5982\u679c provider \u4e0d\u652f\u6301 Responses，\u53ef\u4ee5\u7528 Chat Completions：<\/p>\n

    from<\/span> openai import<\/span> AsyncOpenAI
    \nfrom<\/span> agents import<\/span> Agent,<\/span> OpenAIChatCompletionsModel<\/p>\n

    client =<\/span> AsyncOpenAI(<\/span>
    \n api_key=<\/span>"provider_api_key"<\/span>,<\/span>
    \n base_url=<\/span>"https:\/\/provider.example.com\/v1"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Assistant"<\/span>,<\/span>
    \n instructions=<\/span>"\u4f60\u662f\u4e00\u4e2a\u7b80\u6d01\u52a9\u624b\u3002"<\/span>,<\/span>
    \n model=<\/span>OpenAIChatCompletionsModel(<\/span>
    \n model=<\/span>"provider-model"<\/span>,<\/span>
    \n openai_client=<\/span>client,<\/span>
    \n strict_feature_validation=<\/span>True<\/span>,<\/span>
    \n )<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u5f00\u53d1\u671f\u5efa\u8bae\u6253\u5f00 strict_feature_validation=True\u3002<\/p>\n

    \u5982\u679c\u53d1\u73b0\u4ee3\u7801\u4f9d\u8d56 conversation_id\u3001prompt\u3001Responses tool search \u7b49\u80fd\u529b，\u5c31\u5e94\u8be5\u5207\u56de Responses \u6216\u8c03\u6574\u529f\u80fd\u8bbe\u8ba1\u3002<\/p>\n

    \u5b9e\u8df5\u4e09：\u4f7f\u7528 LiteLLM<\/h3>\n

    from<\/span> agents import<\/span> Agent,<\/span> Runner,<\/span> function_tool,<\/span> set_tracing_disabled<\/p>\n

    set_tracing_disabled(<\/span>True<\/span>)<\/span><\/p>\n

    @function_tool<\/span>
    \ndef<\/span> get_weather<\/span>(<\/span>city:<\/span> str<\/span>)<\/span> –<\/span>><\/span> str<\/span>:<\/span>
    \n return<\/span> f"<\/span>{<\/span>city}<\/span><\/span> \u5929\u6c14\u6674\u3002"<\/span><\/span><\/p>\n

    agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Assistant"<\/span>,<\/span>
    \n instructions=<\/span>"\u4f60\u53ea\u7528\u4ff3\u53e5\u56de\u7b54\u3002"<\/span>,<\/span>
    \n model=<\/span>"litellm\/openrouter\/openai\/gpt-5.4-mini"<\/span>,<\/span>
    \n tools=<\/span>[<\/span>get_weather]<\/span>,<\/span>
    \n)<\/span><\/p>\n

    result =<\/span> await<\/span> Runner.<\/span>run(<\/span>agent,<\/span> "\u4e1c\u4eac\u5929\u6c14\u600e\u4e48\u6837？"<\/span>)<\/span><\/p>\n

    \u8fd9\u91cc litellm\/ \u524d\u7f00\u4f1a\u88ab MultiProvider \u8def\u7531\u5230 LitellmProvider\u3002<\/p>\n

    \u5982\u679c\u8981\u4f20\u663e\u5f0f key，\u4f7f\u7528 LitellmModel(…) \u76f4\u63a5\u6784\u9020\u66f4\u6e05\u695a\u3002<\/p>\n

    \u5b9e\u8df5\u56db：\u4f7f\u7528 Any-LLM<\/h3>\n

    from<\/span> agents import<\/span> Agent,<\/span> Runner,<\/span> set_tracing_disabled<\/p>\n

    set_tracing_disabled(<\/span>True<\/span>)<\/span><\/p>\n

    agent =<\/span> Agent(<\/span>
    \n name=<\/span>"Assistant"<\/span>,<\/span>
    \n instructions=<\/span>"\u4f60\u662f\u4e00\u4e2a\u7b80\u6d01\u52a9\u624b\u3002"<\/span>,<\/span>
    \n model=<\/span>"any-llm\/openrouter\/openai\/gpt-5.4-mini"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    result =<\/span> await<\/span> Runner.<\/span>run(<\/span>agent,<\/span> "\u4f60\u597d"<\/span>)<\/span><\/p>\n

    Any-LLM \u4f1a\u89e3\u6790 provider：<\/p>\n

    openrouter\/openai\/gpt-5.4-mini<\/p>\n

    provider \u662f openrouter，provider model \u662f openai\/gpt-5.4-mini\u3002<\/p>\n

    \u5982\u679c\u4f60\u8981\u5f3a\u5236 API \u7c7b\u578b：<\/p>\n

    AnyLLMModel(<\/span>
    \n model=<\/span>"openrouter\/openai\/gpt-5.4-mini"<\/span>,<\/span>
    \n api=<\/span>"chat_completions"<\/span>,<\/span>
    \n)<\/span><\/p>\n

    \u8fd9\u9002\u5408 provider \u540c\u65f6\u652f\u6301\u591a\u79cd API，\u4f46\u4f60\u5e0c\u671b\u56fa\u5b9a\u884c\u4e3a\u7684\u573a\u666f\u3002<\/p>\n

    \u6d4b\u8bd5\u5efa\u8bae<\/h3>\n

    \u6a21\u578b\u9002\u914d\u5c42\u6d4b\u8bd5\u4e0d\u80fd\u53ea\u6d4b final output\u3002<\/p>\n

    \u81f3\u5c11\u8986\u76d6：<\/p>\n

  • RunConfig.model \u8986\u76d6 Agent.model\u3002<\/li>\n
  • ModelSettings run-level \u8986\u76d6 agent-level\u3002<\/li>\n
  • extra_args \u4e0e\u663e\u5f0f\u5b57\u6bb5\u91cd\u590d\u65f6\u62a5\u9519\u3002<\/li>\n
  • Responses \u8def\u5f84\u80fd\u4f20 previous_response_id \u548c conversation_id\u3002<\/li>\n
  • Chat Completions strict validation \u80fd\u62d2\u7edd Responses-only \u5b57\u6bb5\u3002<\/li>\n
  • Function tool call \u80fd\u4ece provider response \u8f6c\u6210 SDK item\u3002<\/li>\n
  • streaming \u7ec8\u6b62\u4e8b\u4ef6\u80fd\u6b63\u786e\u5904\u7406 failed\/incomplete\u3002<\/li>\n
  • usage \u80fd\u5f52\u4e00\u5316\u3002<\/li>\n
  • retry policy \u53ea\u5728 replay-safe \u573a\u666f\u91cd\u8bd5\u3002<\/li>\n
  • provider aclose() \u80fd\u91ca\u653e\u7f13\u5b58\u8d44\u6e90\u3002<\/li>\n

    \u672c\u4ed3\u5e93\u5df2\u6709\u76f8\u5173\u6d4b\u8bd5：<\/p>\n

  • tests\/models\/test_openai_responses.py<\/li>\n
  • tests\/models\/test_openai_chatcompletions.py<\/li>\n
  • tests\/models\/test_openai_chatcompletions_converter.py<\/li>\n
  • tests\/models\/test_model_retry.py<\/li>\n
  • tests\/models\/test_default_models.py<\/li>\n
  • tests\/models\/test_litellm_chatcompletions_stream.py<\/li>\n
  • tests\/models\/test_any_llm_model.py<\/li>\n
  • tests\/test_config.py<\/li>\n

    \u65b0\u589e provider \u6216 adapter \u65f6，\u5e94\u4f18\u5148\u53c2\u8003\u8fd9\u4e9b\u6d4b\u8bd5\u98ce\u683c\u3002<\/p>\n

    \u5e38\u89c1\u9519\u8bef<\/h3>\n

    \u7b2c\u4e00，\u628a ModelSettings \u5f53\u6210\u6240\u6709 provider \u90fd\u652f\u6301\u7684\u53c2\u6570\u96c6\u5408\u3002<\/p>\n

    \u5b83\u662f SDK \u7edf\u4e00\u914d\u7f6e\u5165\u53e3，\u4e0d\u4ee3\u8868\u6240\u6709 adapter \u90fd\u80fd\u539f\u6837\u652f\u6301\u6bcf\u4e2a\u5b57\u6bb5\u3002<\/p>\n

    \u7b2c\u4e8c，\u7528 Chat Completions \u4f46\u4f9d\u8d56 Responses-only \u529f\u80fd\u3002<\/p>\n

    \u4f8b\u5982 conversation_id\u3001prompt\u3001tool search\u3001deferred-loading Responses tools\u3002<\/p>\n

    \u7b2c\u4e09，\u6df7\u7528 Responses \u4e0e Chat Completions \u540e\u6ca1\u6709\u6d4b streaming\u3002<\/p>\n

    \u975e\u6d41\u5f0f\u80fd\u8dd1\u4e0d\u4ee3\u8868 streaming event \u5f52\u4e00\u5316\u6b63\u786e\u3002<\/p>\n

    \u7b2c\u56db，\u5ffd\u7565 strict validation\u3002<\/p>\n

    \u5f00\u53d1\u671f\u4e0d\u5f00 strict validation，\u5bb9\u6613\u628a\u5b57\u6bb5\u9759\u9ed8\u964d\u7ea7\u6210 warning\u3002<\/p>\n

    \u7b2c\u4e94，\u628a extra_args \u5f53\u6210\u4e07\u80fd\u51fa\u53e3\u3002<\/p>\n

    \u91cd\u590d\u4f20\u53c2\u4f1a\u62a5\u9519；\u672a\u77e5 provider \u53c2\u6570\u4e5f\u53ef\u80fd\u53ea\u5728\u8fd0\u884c\u65f6\u5931\u8d25\u3002<\/p>\n

    \u7b2c\u516d，\u7b2c\u4e09\u65b9 provider \u4e0d\u5173\u95ed tracing\u3002<\/p>\n

    \u6ca1\u6709 OpenAI tracing key \u65f6，\u9ed8\u8ba4 exporter \u53ef\u80fd\u51fa\u73b0 401 \u6216\u8df3\u8fc7\u4e0a\u62a5\u3002<\/p>\n

    \u7b2c\u4e03，\u91cd\u8bd5\u7b56\u7565\u8fc7\u4e8e\u7c97\u66b4\u3002<\/p>\n

    \u5e26 conversation_id \u6216\u5df2\u5f00\u59cb streaming \u7684\u8bf7\u6c42\u4e0d\u80fd\u7b80\u5355\u91cd\u653e\u3002<\/p>\n

    \u7b2c\u516b，\u6ca1\u6709\u5173\u95ed\u6301\u4e45 provider \u8d44\u6e90\u3002<\/p>\n

    WebSocket Responses model \u548c\u67d0\u4e9b\u7b2c\u4e09\u65b9 client \u53ef\u80fd\u6301\u6709\u8fde\u63a5，\u5e94\u5728\u751f\u547d\u5468\u671f\u7ed3\u675f\u65f6\u8c03\u7528 provider\/model \u7684 close\u3002<\/p>\n

    \u672c\u7bc7\u5c0f\u7ed3<\/h3>\n

    \u672c\u7bc7\u62c6\u89e3\u4e86 Agents SDK \u7684\u6a21\u578b\u9002\u914d\u5c42\u3002<\/p>\n

    \u6838\u5fc3\u7ed3\u8bba：<\/p>\n

  • Runner \u53ea\u4f9d\u8d56 Model \u62bd\u8c61，\u4e0d\u4f9d\u8d56\u5177\u4f53 provider \u8bf7\u6c42\u7ed3\u6784\u3002<\/li>\n
  • ModelProvider \u8d1f\u8d23\u628a\u6a21\u578b\u540d\u89e3\u6790\u6210\u5177\u4f53 Model\u3002<\/li>\n
  • RunConfig.model \u4f18\u5148\u7ea7\u9ad8\u4e8e Agent.model\u3002<\/li>\n
  • ModelSettings \u4f7f\u7528\u975e None \u8986\u76d6，\u5e76\u7279\u6b8a\u5408\u5e76 extra_args \u548c retry\u3002<\/li>\n
  • \u5f53\u524d\u6e90\u7801\u9ed8\u8ba4\u6a21\u578b\u662f gpt-5.4-mini，\u5e76\u5e26 GPT-5 \u9ed8\u8ba4 settings\u3002<\/li>\n
  • OpenAIProvider \u9ed8\u8ba4\u8d70 Responses，\u53ef\u5207\u6362 Chat Completions \u6216 Responses WebSocket\u3002<\/li>\n
  • Responses adapter \u539f\u751f\u652f\u6301 server-managed state\u3001prompt\u3001Responses tools\u3001context management \u7b49\u80fd\u529b\u3002<\/li>\n
  • Chat Completions adapter \u9700\u8981\u5728 SDK Responses items \u4e0e chat messages \u4e4b\u95f4\u8f6c\u6362\u3002<\/li>\n
  • LiteLLM \u66f4\u504f Chat Completions \u98ce\u683c，Any-LLM \u53ef\u6309 provider \u80fd\u529b\u9009\u62e9 Responses \u6216 Chat Completions\u3002<\/li>\n
  • \u65b0\u589e\u6216\u5207\u6362 provider \u65f6，\u5fc5\u987b\u9a8c\u8bc1\u5de5\u5177\u3001handoff\u3001\u7ed3\u6784\u5316\u8f93\u51fa\u3001streaming\u3001usage\u3001retry \u548c tracing\u3002<\/li>\n

    \u4e0b\u4e00\u7bc7\u4f1a\u8fdb\u5165 MCP \u96c6\u6210：\u8ba9 Agent \u4f7f\u7528\u5916\u90e8\u5de5\u5177\u751f\u6001，\u7406\u89e3 MCP server \u5982\u4f55\u6620\u5c04\u4e3a SDK tools，\u4ee5\u53ca\u8fde\u63a5\u3001\u7f13\u5b58\u3001\u8fc7\u6ee4\u548c\u5ba1\u6279\u5982\u4f55\u5de5\u4f5c\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

    \u672c\u7bc7\u5bfc\u8bfb
    \n\u524d\u5341\u4e00\u7bc7\u5df2\u7ecf\u628a Agents SDK \u7684\u8fd0\u884c\u4e3b\u7ebf\u62c6\u5f00\u4e86：
    \nAgent \u5b9a\u4e49\u89d2\u8272\u3001\u5de5\u5177\u3001handoffs \u548c\u8f93\u51fa\u7c7b\u578b\u3002Runner \u63a8\u8fdb\u4e00\u6b21 run\u3002Function Tool \u628a Python \u51fd\u6570\u66b4\u9732\u7ed9\u6a21\u578b\u3002Handoff \u548c Agent-as-tool \u7ec4\u7ec7\u591a Agent \u5de5\u4f5c\u6d41\u3002Streaming \u628a\u8fd0\u884c\u8fc7\u7a0b\u53d8\u6210\u4e8b\u4ef6\u6d41\u3002Sessions \u4fdd\u5b58\u591a\u8f6e\u5386\u53f2\u3002Guardrails \u63a7\u5236\u8f93\u5165\u3001\u8f93\u51fa\u548c\u5de5\u5177\u8fb9\u754c\u3002Tracing \u8bb0\u5f55\u4e00\u6b21 run \u7684\u53ef\u89c2\u6d4b\u94fe\u8def\u3002<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[10200,81,190],"topic":[],"class_list":["post-89825","post","type-post","status-publish","format-standard","hentry","category-server","tag-openai-agent","tag-python","tag-190"],"yoast_head":"\nopenai-agents-python-sdk \u6e90\u7801\u89e3\u6790 | \u7b2c\u5341\u4e8c\u7bc7\uff1a\u6a21\u578b\u9002\u914d\u5c42\uff1aResponses\u3001Chat Completions \u4e0e\u7b2c\u4e09\u65b9 Provider - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/89825.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"openai-agents-python-sdk \u6e90\u7801\u89e3\u6790 | \u7b2c\u5341\u4e8c\u7bc7\uff1a\u6a21\u578b\u9002\u914d\u5c42\uff1aResponses\u3001Chat Completions \u4e0e\u7b2c\u4e09\u65b9 Provider - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u672c\u7bc7\u5bfc\u8bfb \u524d\u5341\u4e00\u7bc7\u5df2\u7ecf\u628a Agents SDK \u7684\u8fd0\u884c\u4e3b\u7ebf\u62c6\u5f00\u4e86： Agent \u5b9a\u4e49\u89d2\u8272\u3001\u5de5\u5177\u3001handoffs \u548c\u8f93\u51fa\u7c7b\u578b\u3002Runner \u63a8\u8fdb\u4e00\u6b21 run\u3002Function Tool \u628a Python \u51fd\u6570\u66b4\u9732\u7ed9\u6a21\u578b\u3002Handoff \u548c Agent-as-tool \u7ec4\u7ec7\u591a Agent \u5de5\u4f5c\u6d41\u3002Streaming \u628a\u8fd0\u884c\u8fc7\u7a0b\u53d8\u6210\u4e8b\u4ef6\u6d41\u3002Sessions \u4fdd\u5b58\u591a\u8f6e\u5386\u53f2\u3002Guardrails \u63a7\u5236\u8f93\u5165\u3001\u8f93\u51fa\u548c\u5de5\u5177\u8fb9\u754c\u3002Tracing \u8bb0\u5f55\u4e00\u6b21 run \u7684\u53ef\u89c2\u6d4b\u94fe\u8def\u3002\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/89825.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-08-03T14:00:38+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"14 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/89825.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/89825.html\",\"name\":\"openai-agents-python-sdk \u6e90\u7801\u89e3\u6790 | \u7b2c\u5341\u4e8c\u7bc7\uff1a\u6a21\u578b\u9002\u914d\u5c42\uff1aResponses\u3001Chat Completions \u4e0e\u7b2c\u4e09\u65b9 Provider - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-08-03T14:00:38+00:00\",\"dateModified\":\"2026-08-03T14:00:38+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/89825.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/89825.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/89825.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"openai-agents-python-sdk \u6e90\u7801\u89e3\u6790 | \u7b2c\u5341\u4e8c\u7bc7\uff1a\u6a21\u578b\u9002\u914d\u5c42\uff1aResponses\u3001Chat Completions \u4e0e\u7b2c\u4e09\u65b9 Provider\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"openai-agents-python-sdk \u6e90\u7801\u89e3\u6790 | \u7b2c\u5341\u4e8c\u7bc7\uff1a\u6a21\u578b\u9002\u914d\u5c42\uff1aResponses\u3001Chat Completions \u4e0e\u7b2c\u4e09\u65b9 Provider - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/89825.html","og_locale":"zh_CN","og_type":"article","og_title":"openai-agents-python-sdk \u6e90\u7801\u89e3\u6790 | \u7b2c\u5341\u4e8c\u7bc7\uff1a\u6a21\u578b\u9002\u914d\u5c42\uff1aResponses\u3001Chat Completions \u4e0e\u7b2c\u4e09\u65b9 Provider - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u672c\u7bc7\u5bfc\u8bfb \u524d\u5341\u4e00\u7bc7\u5df2\u7ecf\u628a Agents SDK \u7684\u8fd0\u884c\u4e3b\u7ebf\u62c6\u5f00\u4e86： Agent \u5b9a\u4e49\u89d2\u8272\u3001\u5de5\u5177\u3001handoffs \u548c\u8f93\u51fa\u7c7b\u578b\u3002Runner \u63a8\u8fdb\u4e00\u6b21 run\u3002Function Tool \u628a Python \u51fd\u6570\u66b4\u9732\u7ed9\u6a21\u578b\u3002Handoff \u548c Agent-as-tool \u7ec4\u7ec7\u591a Agent \u5de5\u4f5c\u6d41\u3002Streaming \u628a\u8fd0\u884c\u8fc7\u7a0b\u53d8\u6210\u4e8b\u4ef6\u6d41\u3002Sessions \u4fdd\u5b58\u591a\u8f6e\u5386\u53f2\u3002Guardrails \u63a7\u5236\u8f93\u5165\u3001\u8f93\u51fa\u548c\u5de5\u5177\u8fb9\u754c\u3002Tracing \u8bb0\u5f55\u4e00\u6b21 run \u7684\u53ef\u89c2\u6d4b\u94fe\u8def\u3002","og_url":"https:\/\/www.wsisp.com\/helps\/89825.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-08-03T14:00:38+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"14 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/89825.html","url":"https:\/\/www.wsisp.com\/helps\/89825.html","name":"openai-agents-python-sdk \u6e90\u7801\u89e3\u6790 | \u7b2c\u5341\u4e8c\u7bc7\uff1a\u6a21\u578b\u9002\u914d\u5c42\uff1aResponses\u3001Chat Completions \u4e0e\u7b2c\u4e09\u65b9 Provider - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-08-03T14:00:38+00:00","dateModified":"2026-08-03T14:00:38+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/89825.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/89825.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/89825.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"openai-agents-python-sdk \u6e90\u7801\u89e3\u6790 | \u7b2c\u5341\u4e8c\u7bc7\uff1a\u6a21\u578b\u9002\u914d\u5c42\uff1aResponses\u3001Chat Completions \u4e0e\u7b2c\u4e09\u65b9 Provider"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/89825","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=89825"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/89825\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=89825"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=89825"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=89825"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=89825"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}