[MAF的Agent管道详解-03]连接LLM的IChatClient对象

博客园 - Artech

Artech · 2026-05-28 · via 博客园 - Artech

在IChatClient管道的最末端是一个与LLM进行交互的IChatClient对象，这个对象负责将最终的请求发送给LLM并返回响应结果。这个IChatClient对象的具体类型取决于我们使用的是什么模型以及模型的部署方式。系统提供了很多这样的IChatClient实现来支持不同的模型和部署方式。对于目前主流的LLM，我们都可以直接利用其客户端来创建一个对应的IChatClient对象.

1. 为三种OpenAI客户端创建IChatClient对象

OpenAIClient和AzureOpenAIClient是一个与OpenAI的API进行交互的客户端，我们可以指定模型名称调用其GetChatClient方法来获取一个对应的ChatClient对象。虽然名字雷同，但是这个ChatClient类型可没有实现IChatClient接口，我们需要调用为它定义的扩展方法AsIChatClient来将它转换成一个实现了IChatClient接口的对象。

public class AzureOpenAIClient
{
    public override ChatClient GetChatClient(string deploymentName);
    public override ResponsesClient GetResponsesClient();
}

public class OpenAIClient
{
    public virtual ChatClient GetChatClient(string model);
    public virtual ResponsesClient GetResponsesClient();
}

public static class OpenAIClientExtensions
{
    public static IChatClient AsIChatClient(this ChatClient chatClient);
    public static IChatClient AsIChatClient(this ResponsesClient responseClient, string? defaultModelId = null);
}

前面说过，GetChatClient返回的ChatClient对象采用基于文本补全的无状态的Completion API来与模型进行交互，如果需要采用有状态的Responses API，需要调用GetResponsesClient方法来获取一个ResponsesClient对象。系统依然为ResponsesClient对象定义了一个AsIChatClient的扩展方法来将它转换成一个实现了IChatClient接口的对象。

如果使用的是基于Microsoft Foundry的AIProjectClient客户端。由于它的基类是ClientConnectionProviderExtensions，我们可以调用其扩展方法GetProjectOpenAIClient得到一个ProjectOpenAIClient对象。由于ProjectOpenAIClient继承自OpenAIClient，我们同样可以调用为它定义的AsIChatClient扩展方法来将它转换成一个实现了IChatClient接口的对象。

public class AIProjectClient : ClientConnectionProvider

public static class ClientConnectionProviderExtensions
{
    public static ProjectOpenAIClient GetProjectOpenAIClient(
        this ClientConnectionProvider connectionProvider, 
        ProjectOpenAIClientOptions options = null);
}

public class ProjectOpenAIClient : OpenAIClient

2. 模拟Agent的ReAct循环

接下来我们看看一个利用OpenAIClient创建的IChatClient对象在调用LLM的时候，提供的请求和响应内容是什么样子的。下面的代码模拟了一个Agent内部的执行流程（ReAct循环），我们使用这个Agent来根据苏州的天气给出一些着装建议。我们根据OpenAIClient创建了对应的IChatClient对象，整个流程涉及两次针对它的调用。两次调用使用同一个ChatOptions对象，我们为这个ChatOptions设置了系统指令（你是一个深谙养身之道的时尚顾问）并注册了一个用于查询天气的工具GetWeather。

using dotenv.net;
using Microsoft.Extensions.AI;
using OpenAI;
using System.ClientModel;
using System.ComponentModel;

DotEnv.Load();
var model = Environment.GetEnvironmentVariable("MODEL")!;
var apiKey = Environment.GetEnvironmentVariable("API_KEY")!;
var openAIUrl = Environment.GetEnvironmentVariable("OPENAI_URL")!;
var openAIClient = new OpenAIClient(
    credential: new ApiKeyCredential(key: apiKey),
    options: new OpenAIClientOptions
    {
        Endpoint = new Uri(openAIUrl)
    });

var chatClient = openAIClient.GetResponsesClient().AsIChatClient(defaultModelId:model);
var options = new ChatOptions
{   
    Instructions = "你是一个深谙养身之道的时尚顾问。",
    Tools = [AIFunctionFactory.Create(GetWeather)]
};
var message = new ChatMessage(role: ChatRole.User, content: "根据苏州的天气给我一些着装建议。");
List<ChatMessage> messages = [message];

// First turn: user -> assistant (with function call)
var response = await chatClient.GetResponseAsync(
    messages: messages,
    options: options);
messages.AddRange(response.Messages);

var functionCall = response.Messages.Last().Contents.OfType<FunctionCallContent>().Single();
var tool = options.Tools.Single(t => t.Name == functionCall.Name);
var toolResult = await ((AIFunction)tool).InvokeAsync(new AIFunctionArguments(functionCall.Arguments));
var toolResultMessage = new ChatMessage(ChatRole.Tool, [new FunctionResultContent(functionCall.CallId, toolResult)]);
messages.Add(toolResultMessage);

// Second turn: user -> assistant (with tool result)
response = await chatClient.GetResponseAsync(
    messages: messages,
    options: options);

Console.WriteLine(response.Messages.Last().Text);
static string GetWeather([Description("Location for weather query")] string location) => $"{location} 当前晴朗，气温为25°C。";

我们指定查询（根据苏州的天气给我一些着装建议）和ChatOptions调用IChatClient对象。LLM经过推理任务需要调用工具函数GetWeather来获取苏州的天气信息，所以响应消息的内容列表会包含一个FunctionCallContent。在手工将响应消息添加到消息列表中后，我们利用FunctionCallContent从注册的工具列表中找到对应的工具。

我们将LLM提供的输入参数从FunctionCallContent提取出来后，调用工具函数GetWeather得到对应的结果。接下来我们针对工具的返回结果创建一个角色为Tool的ChatMessage对象，并将它添加到消息列表中。最后我们再次调用IChatClient对象来获取LLM的最终回复。此时LLM就可以根据工具的返回结果来生成最终如下所示的答案：

好的，我们就顺着苏州此刻**25°C、晴朗**的状态，从**养身 + 时尚**两个角度来搭配。

---

## 🌤️ 今日苏州着装总思路
**关键词：清爽透气、遮阳不闷、早晚微调**

25°C 属于非常舒适的温度，但苏州湿度通常不低，**选对面料比堆叠衣服更重要**。

---

## 👕 上装建议
- **首选**：
  - 棉麻衬衫（浅色系：米白、浅灰、雾蓝）
  - 薄款针织或天丝T恤
- **养身理由**：
  - 棉麻、天丝透气吸湿，减少湿热闷汗，对皮肤和气血运行更友好
- **小技巧**：
  - 避免紧身、化纤材质，容易“闷火生湿”

---

## 👖 下装建议
- **推荐**：
  - 九分直筒裤 / 轻薄阔腿裤
  - 膝下A字裙或真丝半裙
- **颜色**：
  - 浅卡其、灰绿、烟粉色，有“降燥感”
- **养身点**：
  - 不勒腹、不裹腿，有助于脾胃与下肢血液循环

---

## 👟 鞋履选择
- **白色/浅色透气运动鞋**
- **软底乐福鞋 / 平底凉鞋（包后跟更养脚）**
- 避免全天穿完全平底或过硬的鞋，对足底经络不友好

---

## 🧥 随身加一件（很关键）
- **薄开衫 / 防晒衬衫**
  - 室内空调 + 早晚微风时护住肩颈
  - 肩颈保暖 = 少落枕、少疲劳

---

## 🕶️ 配饰与养身小细节
- **帽子或遮阳伞**：防晒就是防“耗气”
- **天然材质包袋**：帆布、草编，更符合当下季节气场
- **配色不宜过于浓烈**：春夏交替，宜“柔不宜躁”

---

如果你愿意告诉我：
- 是**上班 / 休闲 / 约会 / 出游**
- 或偏**中性、优雅、运动风**

我可以直接帮你搭一整套「今天就能穿出门」的苏州限定穿搭 🌿

这是第一轮调用LLM提供的请求和得到的响应内容：

{
  "model": "gpt-5.2-chat",
  "tools": [
    {
      "type": "function",
      "name": "_Main_g_GetWeather_0_1",
      "description": "",
      "parameters": {
        "type": "object",
        "required": [
          "location"
        ],
        "properties": {
          "location": {
            "description": "Location for weather query",
            "type": "string"
          }
        },
        "additionalProperties": false
      },
      "strict": null
    }
  ],
  "input": [
    {
      "type": "message",
      "role": "user",
      "content": [
        {
          "type": "input_text",
          "text": "根据苏州的天气给我一些着装建议。"
        }
      ]
    }
  ],
  "instructions": "你是一个深谙养身之道的时尚顾问。"
}

{
  "id": "resp_08fd9fcf3071918b006a000a00f53081938f105b04d924cb63",
  "object": "response",
  "created_at": 1778387456,
  "status": "completed",
  "background": false,
  "completed_at": 1778387457,
  "content_filters": [
    {
      "blocked": false,
      "source_type": "prompt",
      "content_filter_raw": [],
      "content_filter_results": {
        "hate": {
          "filtered": false,
          "severity": "safe"
        },
        "sexual": {
          "filtered": false,
          "severity": "safe"
        },
        "violence": {
          "filtered": false,
          "severity": "safe"
        },
        "self_harm": {
          "filtered": false,
          "severity": "safe"
        }
      },
      "content_filter_offsets": {
        "start_offset": 0,
        "end_offset": 49,
        "check_offset": 0
      }
    },
    {
      "blocked": false,
      "source_type": "completion",
      "content_filter_raw": [],
      "content_filter_results": {
        "hate": {
          "filtered": false,
          "severity": "safe"
        },
        "sexual": {
          "filtered": false,
          "severity": "safe"
        },
        "violence": {
          "filtered": false,
          "severity": "safe"
        },
        "self_harm": {
          "filtered": false,
          "severity": "safe"
        }
      },
      "content_filter_offsets": {
        "start_offset": 0,
        "end_offset": 1170,
        "check_offset": 0
      }
    }
  ],
  "error": null,
  "frequency_penalty": 0.0,
  "incomplete_details": null,
  "instructions": "你是一个深谙养身之道的时尚顾问。",
  "max_output_tokens": null,
  "max_tool_calls": null,
  "model": "gpt-5.2-chat",
  "output": [
    {
      "id": "rs_08fd9fcf3071918b006a000a0152f88193b91826c5aa30181a",
      "type": "reasoning",
      "summary": []
    },
    {
      "id": "fc_08fd9fcf3071918b006a000a01c6548193b850d9b453ce47f8",
      "type": "function_call",
      "status": "completed",
      "arguments": "{\"location\":\"苏州\"}",
      "call_id": "call_kYGZgvSLCPipLqtmiIqfnIDT",
      "name": "_Main_g_GetWeather_0_1"
    }
  ],
  "parallel_tool_calls": true,
  "presence_penalty": 0.0,
  "previous_response_id": null,
  "prompt_cache_key": null,
  "prompt_cache_retention": null,
  "reasoning": {
    "effort": "medium",
    "summary": null
  },
  "safety_identifier": null,
  "service_tier": "default",
  "store": true,
  "temperature": 1.0,
  "text": {
    "format": {
      "type": "text"
    },
    "verbosity": "medium"
  },
  "tool_choice": "auto",
  "tools": [
    {
      "type": "function",
      "description": null,
      "name": "_Main_g_GetWeather_0_1",
      "parameters": {
        "type": "object",
        "required": [
          "location"
        ],
        "properties": {
          "location": {
            "description": "Location for weather query",
            "type": "string"
          }
        },
        "additionalProperties": false
      },
      "strict": false
    }
  ],
  "top_logprobs": 0,
  "top_p": 0.85,
  "truncation": "disabled",
  "usage": {
    "input_tokens": 83,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens": 43,
    "output_tokens_details": {
      "reasoning_tokens": 0
    },
    "total_tokens": 126
  },
  "user": null,
  "metadata": {}
}

这是第二轮调用LLM提供的请求和得到的响应内容：

{
  "model": "gpt-5.2-chat",
  "tools": [
    {
      "type": "function",
      "name": "_Main_g_GetWeather_0_1",
      "description": "",
      "parameters": {
        "type": "object",
        "required": [
          "location"
        ],
        "properties": {
          "location": {
            "description": "Location for weather query",
            "type": "string"
          }
        },
        "additionalProperties": false
      },
      "strict": null
    }
  ],
  "input": [
    {
      "type": "message",
      "role": "user",
      "content": [
        {
          "type": "input_text",
          "text": "根据苏州的天气给我一些着装建议。"
        }
      ]
    },
    {
      "type": "reasoning",
      "id": "rs_08fd9fcf3071918b006a000a0152f88193b91826c5aa30181a",
      "summary": []
    },
    {
      "type": "function_call",
      "id": "fc_08fd9fcf3071918b006a000a01c6548193b850d9b453ce47f8",
      "status": "completed",
      "call_id": "call_kYGZgvSLCPipLqtmiIqfnIDT",
      "name": "_Main_g_GetWeather_0_1",
      "arguments": "{\"location\":\"苏州\"}"
    },
    {
      "type": "function_call_output",
      "call_id": "call_kYGZgvSLCPipLqtmiIqfnIDT",
      "output": "\"苏州 当前晴朗，气温为25°C。\""
    }
  ],
  "instructions": "你是一个深谙养身之道的时尚顾问。"
}

{
  "id": "resp_08fd9fcf3071918b006a000a025fbc8193a1219305fbea8789",
  "object": "response",
  "created_at": 1778387458,
  "status": "completed",
  "background": false,
  "completed_at": 1778387467,
  "content_filters": [
    {
      "blocked": false,
      "source_type": "completion",
      "content_filter_raw": [],
      "content_filter_results": {
        "hate": {
          "filtered": false,
          "severity": "safe"
        },
        "sexual": {
          "filtered": false,
          "severity": "safe"
        },
        "violence": {
          "filtered": false,
          "severity": "safe"
        },
        "self_harm": {
          "filtered": false,
          "severity": "safe"
        }
      },
      "content_filter_offsets": {
        "start_offset": 0,
        "end_offset": 1912,
        "check_offset": 0
      }
    }
  ],
  "error": null,
  "frequency_penalty": 0.0,
  "incomplete_details": null,
  "instructions": "你是一个深谙养身之道的时尚顾问。",
  "max_output_tokens": null,
  "max_tool_calls": null,
  "model": "gpt-5.2-chat",
  "output": [
    {
      "id": "msg_08fd9fcf3071918b006a000a02c53c819385543898398da88e",
      "type": "message",
      "status": "completed",
      "content": [
        {
          "type": "output_text",
          "annotations": [],
          "logprobs": [],
          "text": "...（同上面展示的LLM最终回复内容）..."
        }
      ],
      "role": "assistant"
    }
  ],
  "parallel_tool_calls": true,
  "presence_penalty": 0.0,
  "previous_response_id": null,
  "prompt_cache_key": null,
  "prompt_cache_retention": null,
  "reasoning": {
    "effort": "medium",
    "summary": null
  },
  "safety_identifier": null,
  "service_tier": "default",
  "store": true,
  "temperature": 1.0,
  "text": {
    "format": {
      "type": "text"
    },
    "verbosity": "medium"
  },
  "tool_choice": "auto",
  "tools": [
    {
      "type": "function",
      "description": null,
      "name": "_Main_g_GetWeather_0_1",
      "parameters": {
        "type": "object",
        "required": [
          "location"
        ],
        "properties": {
          "location": {
            "description": "Location for weather query",
            "type": "string"
          }
        },
        "additionalProperties": false
      },
      "strict": false
    }
  ],
  "top_logprobs": 0,
  "top_p": 0.85,
  "truncation": "disabled",
  "usage": {
    "input_tokens": 157,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens": 612,
    "output_tokens_details": {
      "reasoning_tokens": 0
    },
    "total_tokens": 769
  },
  "user": null,
  "metadata": {}
}

此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。原文来自 — 版权归原作者所有。

推荐订阅源

博客园 - Artech

1. 为三种OpenAI客户端创建IChatClient对象

2. 模拟Agent的ReAct循环