LangChain.Net 使用案例

梁723

312人浏览 · 2025-09-16 07:10:44

梁723 · 2025-09-16 07:10:44 发布

1.1、DeepSeek 连接

案例一：

using LangChain.Providers;
using LangChain.Providers.DeepSeek;
using LangChain.Providers.DeepSeek.Predefined;

string apiKey = "sk-30c3xxxxx179e";
var model = new DeepSeekChatModel(new DeepSeekProvider(new DeepSeekConfiguration()
{
    ApiKey = apiKey,
    Endpoint= "https://api.deepseek.com"
}));
var result = await model.GenerateAsync("详细介绍事件总线，在项目中为什么要使用，如何使用");
Console.WriteLine(result);
Console.ReadLine();

案例二：设置历史数据

using LangChain.Providers;
using LangChain.Providers.DeepSeek;
using LangChain.Providers.DeepSeek.Predefined;

string apiKey = "sk-30c33cxxxxxxxxxxx6179e";
var model = new DeepSeekChatModel(new DeepSeekProvider(new DeepSeekConfiguration()
{
    ApiKey = apiKey,
    Endpoint= "https://api.deepseek.com"
}));
var result = await model.GenerateAsync(new ChatRequest()
{
    Messages = new List<Message> {
        new Message(){
            Content="你是一名资深的系统架构师",
            Role=MessageRole.System
        },
        new Message(){
            Content="详细介绍事件总线，在项目中为什么要使用，如何使用",
            Role=MessageRole.Human
        }
    }
});
Console.WriteLine(result);
Console.ReadLine();

案例三：流式输出

using LangChain.Providers;
using LangChain.Providers.DeepSeek;
using LangChain.Providers.DeepSeek.Predefined;

string apiKey = "sk-30c33xxxxxxxx592f3d6179e";
var model = new DeepSeekChatModel(new DeepSeekProvider(new DeepSeekConfiguration()
{
    ApiKey = apiKey,
    Endpoint= "https://api.deepseek.com"
}));
await foreach (var chunk in model.GenerateAsync(new ChatRequest()
{
    Messages = new List<Message> {
        new Message(){
            Content="你是一名资深的系统架构师",
            Role=MessageRole.System
        },
        new Message(){
            Content="详细介绍事件总线，在项目中为什么要使用，如何使用",
            Role=MessageRole.Human
        }
    }
}, new ChatSettings(){ UseStreaming = true }))
{
    Console.Write(chunk);
}

1.2、模板

案例一：增加模板

using LangChain.Chains.LLM;
using LangChain.Prompts;
using LangChain.Providers;
using LangChain.Providers.DeepSeek;
using LangChain.Providers.DeepSeek.Predefined;


string apiKey = "sk-30c3xxxxxxxxxxxxxxx592f3d6179e";
var model = new DeepSeekChatModel(new DeepSeekProvider(new DeepSeekConfiguration()
{
    ApiKey = apiKey,
    Endpoint= "https://api.deepseek.com"
}));
var propmpt = new PromptTemplate(new PromptTemplateInput(
        template:"请详细介绍{product}",
        inputVariables: ["product"]
    ));
var chain = new LlmChain(new LlmChainInput(model, prompt: propmpt));
//var result = await chain.CallAsync(new ChainValues(new Dictionary<string, object>() {
//    { "product","CAP" }
//}));
//Console.WriteLine(result.Value["text"]);
var result2 = await chain.RunAsync(input:new Dictionary<string, object>() {
    { "product","MySql索引" }
});
Console.WriteLine(result2);

模板二：模板多参数

using LangChain.Chains.LLM;
using LangChain.Prompts;
using LangChain.Providers;
using LangChain.Providers.DeepSeek;
using LangChain.Providers.DeepSeek.Predefined;
using LangChain.Schema;


string apiKey = "sk-30cxxxxxxxxxxxxx592f3d6179e";
var model = new DeepSeekChatModel(new DeepSeekProvider(new DeepSeekConfiguration()
{
    ApiKey = apiKey,
    Endpoint= "https://api.deepseek.com"
}));
var prompt = ChatPromptTemplate.FromPromptMessages(new List<BaseMessagePromptTemplate>() { 
     SystemMessagePromptTemplate.FromTemplate("请将{input}翻译为{output}"),
     HumanMessagePromptTemplate.FromTemplate("{text}")
});
var chat = new LlmChain(new LlmChainInput(model, prompt) { 
    Verbose = true,
});
var result = await chat.CallAsync(new ChainValues(new Dictionary<string, object>(3) {
    { "input","中文"},
    { "output","英文"},
    { "text","早上好，很高兴见到你" }
}));
Console.WriteLine(result.Value["text"]);
Console.ReadLine();

1.3、历史记录

using LangChain.Memory;
using LangChain.Providers;
using LangChain.Providers.DeepSeek;
using LangChain.Providers.DeepSeek.Predefined;
using static LangChain.Chains.Chain;

internal class Program
{
    private static async Task Main(string[] args)
    {
        string apiKey = "sk-30xxxxxxxxxxxxxxxf3d6179e";
        var model = new DeepSeekChatModel(new DeepSeekProvider(new DeepSeekConfiguration()
        {
            ApiKey = apiKey,
            Endpoint = "https://api.deepseek.com"
        }));
        var template = @"以下是人类和人工智能之间的友好对话。
            {history}
            Human: {input}
            AI：
        ";
        var memory = PickMemoryStrategy(model);
        var chain = LoadMemory(memory, outputKey: "history")
            | Template(template)
            | LLM(model)
            | UpdateMemory(memory, requestKey: "input", responseKey: "text");
        while (true)
        {
            Console.WriteLine();
            Console.Write("Human: ");
            var input = Console.ReadLine() ?? string.Empty;
            if (input == "exit")
            {
                break;
            }
            //通过将用户的输入添加到原始链中来构建新链
            var currentChain = Set(input, "input")
                | chain;
            // 从AI获得响应
            var response = await currentChain.RunAsync("text"); 
            Console.Write("AI: ");
            Console.WriteLine(response);
        }

    }
     
    /// <summary>
    /// 获取其他历史记录
    /// </summary>
    /// <returns></returns>
    private static BaseChatMessageHistory GetChatMessageHistory()
    {
        //获取其他历史记录
        return new ChatMessageHistory();
    }

    private static BaseChatMemory PickMemoryStrategy(IChatModel model)
    {
        MessageFormatter messageFormatter = new MessageFormatter {
            AiPrefix = "AI",
            HumanPrefix = "Human"
        };
        BaseChatMessageHistory chatHistory= GetChatMessageHistory();
        string memoryClassName = PromptForChoice(new[]
        {
            nameof(ConversationBufferMemory),
            nameof(ConversationWindowBufferMemory),
            nameof(ConversationSummaryMemory),
            nameof(ConversationSummaryBufferMemory)
        });
        switch (memoryClassName)
        {
            case nameof(ConversationBufferMemory):
                return GetConversationBufferMemory(chatHistory, messageFormatter);

            case nameof(ConversationWindowBufferMemory):
                return GetConversationWindowBufferMemory(chatHistory, messageFormatter);

            case nameof(ConversationSummaryMemory):
                return GetConversationSummaryMemory(chatHistory, messageFormatter, model);

            case nameof(ConversationSummaryBufferMemory):
                return GetConversationSummaryBufferMemory(chatHistory, messageFormatter, (IChatModelWithTokenCounting)model);

            default:
                throw new InvalidOperationException($"Unexpected memory class name: '{memoryClassName}'");
        }
    }

    private static string PromptForChoice(string[] choiceTexts)
    {
        while (true)
        {
            Console.Clear();
            Console.WriteLine("从以下选项中选择:");
            int choiceNumber = 1;
            foreach (string choiceText in choiceTexts)
            {
                Console.WriteLine($"    {choiceNumber}: {choiceText}");
                choiceNumber++;
            }
            Console.WriteLine();
            Console.Write("Enter choice: ");
            string choiceEntry = Console.ReadLine() ?? string.Empty;
            if (int.TryParse(choiceEntry, out int choiceIndex))
            {
                string choiceText = choiceTexts[choiceIndex];
                Console.WriteLine();
                Console.WriteLine($"You selected '{choiceText}'");
                return choiceText;
            }
        }
    }

    private static BaseChatMemory GetConversationBufferMemory(BaseChatMessageHistory chatHistory, MessageFormatter messageFormatter)
    {
        return new ConversationBufferMemory(chatHistory)
        {
            Formatter = messageFormatter
        };
    }

    private static BaseChatMemory GetConversationWindowBufferMemory(BaseChatMessageHistory chatHistory, MessageFormatter messageFormatter)
    {
        return new ConversationWindowBufferMemory(chatHistory)
        {
            WindowSize = 3,
            Formatter = messageFormatter
        };
    }

    private static BaseChatMemory GetConversationSummaryMemory(BaseChatMessageHistory chatHistory, MessageFormatter messageFormatter, IChatModel model)
    {
        return new ConversationSummaryMemory(model, chatHistory)
        {
            Formatter = messageFormatter
        };
    }
   
    private static BaseChatMemory GetConversationSummaryBufferMemory(BaseChatMessageHistory chatHistory, MessageFormatter messageFormatter, IChatModelWithTokenCounting model)
    {
        return new ConversationSummaryBufferMemory(model, chatHistory)
        {
            MaxTokenCount = 25,
            Formatter = messageFormatter
        };
    }

}

1.4、多模态

using LangChain.Abstractions.Chains.Base;
using LangChain.Chains.LLM;
using LangChain.Chains.Sequentials;
using LangChain.Prompts;
using LangChain.Providers;
using LangChain.Providers.DeepSeek;
using LangChain.Providers.DeepSeek.Predefined;
using LangChain.Schema;

string apiKey = "sk-30c3xxxxxxxxxxxxxxx92f3d6179e";
var model = new DeepSeekChatModel(new DeepSeekProvider(new DeepSeekConfiguration()
{
    ApiKey = apiKey,
    Endpoint = "https://api.deepseek.com"
}));
var firstTemplate = "What is a good name for a company that makes {product}?";
var firstPrompt = new PromptTemplate(new PromptTemplateInput(firstTemplate,new List<string> { "product" }));
// 创建第一个 LLM 链，配置如下：
// - 使用上面定义的 OpenAI 模型和提示模板
// - 开启详细日志输出（Verbose = true）
// - 指定输出键为 "company_name"（便于后续链获取结果）
var chainOne = new LlmChain(new LlmChainInput(model, firstPrompt) { 
    Verbose=true,
    OutputKey= "company_name"
});

// 第二个提示模板：用于生成公司描述，包含占位符 {company_name}
var secongTemplate = "Write a 50 words description for the following company:{company_name}";
// company_name : 声明需要从上游链获取的参数
var secondPrompt = new PromptTemplate(new PromptTemplateInput(secongTemplate, new List<string> { "company_name" }));
var chainTwo = new LlmChain(new LlmChainInput(model, secondPrompt));

// - chains: 按顺序执行的链数组（chainOne → chainTwo）
// - inputVariables: 整个链的初始输入参数（此处为 product）
// - outputVariables: 最终输出的结果键（此处包含 chainOne 输出的 company_name 和 chainTwo 输出的 text）
var overallChain = new SequentialChain(new SequentialChainInput(
    new IChain[]{ chainOne, chainTwo },
    new[] { "product" },       // 初始输入参数
    new[] { "company_name", "text" } // 最终输出值对应的键
));


// 执行链式调用，传入初始参数 product="colourful socks"
var result = await overallChain.CallAsync(new ChainValues(
        new Dictionary<string, object> {
            { "product", "colourful socks" } // 设置输入参数
        }
    ));
// 输出第二个链生成的公司描述（通过键 "text" 获取结果）
Console.WriteLine(result.Value["text"]);
Console.WriteLine("SequentialChain sample finished.");
Console.ReadLine();

1.5、导入向量库

案例一：导入文档

using LangChain.Databases.Sqlite;
using LangChain.DocumentLoaders;
using LangChain.Extensions;
using LangChain.Providers;
using LangChain.Providers.Ollama;
using LangChain.Splitters.Text;


var provider =new OllamaProvider();
var embeddingModel = new OllamaEmbeddingModel(provider,id: "all-minilm:latest");
var llm = new OllamaChatModel(provider,id: "llama3:latest");
var vectorDatabase = new SqLiteVectorDatabase(dataSource: "vectors.db");
// 使用文本分割器
var textSplitter = new RecursiveCharacterTextSplitter(
    chunkSize: 1000,
    chunkOverlap: 200
);
var vectorCollection = await vectorDatabase.AddDocumentsFromAsync<PdfPigPdfLoader>(
    embeddingModel,
    dimensions:1384,
    dataSource:DataSource.FromPath("D:\\123.pdf"),
    //dataSource: DataSource.FromUrl("https://canonburyprimaryschool.co.uk/wp-content/uploads/2016/01/Joanne-K.-Rowling-Harry-Potter-Book-1-Harry-Potter-and-the-Philosophers-Stone-EnglishOnlineClub.com_.pdf"),
    collectionName: "harrypotter",
    textSplitter: textSplitter,
    behavior: AddDocumentsToDatabaseBehavior.AlwaysAddDocuments);
var question = "介绍菜品管理";
var similarDocuments = await vectorCollection.GetSimilarDocuments(embeddingModel, question,amount:10);
var answer = await llm.GenerateAsync(
   $"""
     使用以下上下文来回答最后的问题。
     如果答案与上下文不符，那么就说你不知道，不要试图编造答案。
     请确保答案尽可能简短。所有内容中文回答

     {similarDocuments.AsString()}

     问题: {question}
     回答:
     """);

Console.WriteLine($"LLM answer: {answer}"); 
Console.ReadLine();

案例二：读取向量库

using LangChain.Databases.Sqlite;
using LangChain.Extensions;
using LangChain.Providers;
using LangChain.Providers.Ollama;

var provider = new OllamaProvider();
var embeddingModel = new OllamaEmbeddingModel(provider, id: "all-minilm:latest");
var llm = new OllamaChatModel(provider, id: "llama3:latest");

// 连接到现有的向量数据库
var vectorDatabase = new SqLiteVectorDatabase(dataSource: "vectors.db");

// 获取现有的集合（不需要添加新文档）
var vectorCollection = await vectorDatabase.GetCollectionAsync("harrypotter");

var question = "介绍菜品管理";
var similarDocuments = await vectorCollection.GetSimilarDocuments(embeddingModel, question, amount: 5); // 减少数量以提高精度

// 输出相似文档用于调试
//Console.WriteLine($"检索到的相关文档: {similarDocuments.AsString()}");

var answer = await llm.GenerateAsync(
   $"""
     使用以下上下文来回答最后的问题。
     如果答案与上下文不符，那么就说你不知道，不要试图编造答案。
     请确保答案尽可能简短。所有内容中文回答

     {similarDocuments.AsString()}

     问题: {question}
     回答:
     """);

Console.WriteLine($"LLM 回答: {answer}");
Console.ReadLine();

1.6、HuggingFace

using LangChain.Providers;
using LangChain.Providers.HuggingFace;
using LangChain.Providers.HuggingFace.Predefined;

var apiKey = "hf_LDCUmQqaHxxxxxxxxxxxoMVMQhceoS";
using var client = new HttpClient();
var provider = new HuggingFaceProvider(apiKey:apiKey,client);
var gpt2Model = new Gpt2Model(provider);
var gp2ModelResponse = await gpt2Model.GenerateAsync("给一家生产彩色袜子的公司起个好名字是什么?");
Console.WriteLine("### GP2 Response");
Console.WriteLine(gp2ModelResponse);


const string imageToTextModel = "Salesforce/blip-image-captioning-base";
var model = new HuggingFaceImageToTextModel(provider, imageToTextModel);
var path = Path.Combine(Path.GetTempPath(), "solar_system.png");
var imageData = await File.ReadAllBytesAsync(path);
var binaryData = new BinaryData(imageData, "image/jpg");
var imageToTextResponse = await model.GenerateTextFromImageAsync(new ImageToTextRequest
{
    Image = binaryData
});
Console.WriteLine("\n\n### ImageToText Response");
Console.WriteLine(imageToTextResponse.Text);

Console.ReadLine();

1.7、Web Api请求大模型

Program.cs

using LangChain.Extensions.DependencyInjection;

var builder = WebApplication.CreateBuilder(args);
builder.Services.AddControllers();
builder.Services.AddEndpointsApiExplorer();
builder.Services.AddSwaggerGen();
builder.Services.AddOpenAi();
builder.Services.AddAnthropic();

var app = builder.Build();
if (app.Environment.IsDevelopment())
{
    app.UseSwagger();
    app.UseSwaggerUI();
}
app.UseHttpsRedirection();
app.UseAuthorization();
app.MapControllers();
app.Run();

api:

[ApiController]
[Route("[controller]")]
public class OpenAiSampleController : ControllerBase
{
    private readonly OpenAiProvider _openAi;
    public OpenAiSampleController(OpenAiProvider openAi)
    {
        _openAi = openAi;
    }

    [HttpGet(Name = "GetOpenAiResponse")]
    public async Task<string> Get()
    {
        var llm = new OpenAiChatModel(_openAi, id: ChatClient.LatestFastModel);
        var response = await llm.GenerateAsync("What is a good name for a company that sells colourful socks?");
        return response.LastMessageContent;
    }
}


[ApiController]
[Route("[controller]")]
public class AnthropicSampleController : ControllerBase
{
    private readonly AnthropicChatModel _anthropicModel;  
    public AnthropicSampleController(AnthropicChatModel anthropicModel)
    {
        _anthropicModel = anthropicModel;
    }

    [HttpGet(Name = "GetAnthropicResponse")]
    public async Task<string> Get()
    {
        var response = await _anthropicModel.GenerateAsync("What is a good name for a company that sells colourful socks?");
        return response.LastMessageContent;
    }
}

智能体开发者社区

中国智能体开发者社区，聚焦智能体与大模型开发，提供前沿资讯、实用工具链、开源项目及行业案例。通过技术沙龙、开发者大赛等活动，促进经验交流与协作，助力开发者快速构建创新智能应用。

更多推荐

从一张商品图到正式上架：实测店匠 Shoplazza Athena，AI Agent 不只是聊天框

智能体开发者社区

2026 AI Agent 落地观察：从演示到生产，智能体正在穿越规模化拐点

智能体开始批量走进生产环境。行业调研数据显示，超过半数使用 AI 的企业已将 Agent 部署至生产环境，2026 年被普遍认为是智能体应用的爆发元年——从单点实验走向系统协作，从模型能力竞争转向协议生态竞争。这篇文章梳理近期 Agent 落地的几个关键信号：编程智能体的工作台化、协议层的标准化、企业级 AgentOps 的成型，以及各行业的真实案例，最后聊聊落地里真实的坑。—## 一、格局之变：