langchainrust

A LangChain-inspired Rust framework for building LLM applications. Provides abstractions for agents, chains, memory, RAG pipelines, and tool-calling workflows.

一个受 LangChain 启发的 Rust 框架，用于构建 LLM 应用。提供 Agent、Chain、Memory、RAG 和工具调用等核心抽象。

English | 中文文档

中文文档

✨ 核心特性

关键优势

• 🚀 完全异步 - 基于 Tokio 的 async/await 支持
• 🔒 类型安全 - 利用 Rust 类型系统确保代码可靠性
• 📦 零成本抽象 - 高性能设计
• 🎯 简洁 API - 直观易用的接口
• 🔌 易于扩展 - 方便添加自定义工具和组件

📦 安装

在 Cargo.toml 中添加：

[dependencies]

langchainrust = "0.1.2"

tokio = { version = "1.0", features = ["full"] }

🚀 快速开始

基础对话

use langchainrust::{OpenAIChat, OpenAIConfig, BaseChatModel};
use langchainrust::schema::Message;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let config = OpenAIConfig {
        api_key: std::env::var("OPENAI_API_KEY")?,
        base_url: "https://api.openai.com/v1".to_string(),
        model: "gpt-3.5-turbo".to_string(),
        streaming: false,
        temperature: Some(0.7),
        max_tokens: Some(500),
        ..Default::default()
    };
    
    let llm = OpenAIChat::new(config);
    
    let messages = vec![
        Message::system("你是一个友好的助手。"),
        Message::human("什么是 Rust 语言？"),
    ];
    
    let response = llm.chat(messages, None).await?;
    println!("{}", response.content);
    
    Ok(())
}

提示词模板

use langchainrust::prompts::{PromptTemplate, ChatPromptTemplate};
use langchainrust::schema::Message;
use std::collections::HashMap;

// 字符串模板
let template = PromptTemplate::new("你好，{name}！今天是{day}。");
let mut vars = HashMap::new();
vars.insert("name", "小明");
vars.insert("day", "星期一");
let prompt = template.format(&vars)?;

// 聊天模板
let chat_template = ChatPromptTemplate::new(vec![
    Message::system("你是一个{role}，专精于{domain}。"),
    Message::human("你好，我是{name}。"),
    Message::human("{question}"),
]);

let mut vars = HashMap::new();
vars.insert("role", "Rust 专家");
vars.insert("domain", "系统编程");
vars.insert("name", "小红");
vars.insert("question", "解释 Rust 的所有权机制");

let messages = chat_template.format(&vars)?;

Agent 与工具调用

use langchainrust::{
    ReActAgent, AgentExecutor, BaseAgent, BaseTool,
    Calculator, DateTimeTool, SimpleMathTool,
};
use std::sync::Arc;

let tools: Vec<Arc<dyn BaseTool>> = vec![
    Arc::new(Calculator::new()),
    Arc::new(DateTimeTool::new()),
    Arc::new(SimpleMathTool::new()),
];

let agent = ReActAgent::new(llm, tools.clone(), None);
let executor = AgentExecutor::new(Arc::new(agent) as Arc<dyn BaseAgent>, tools)
    .with_max_iterations(5);

let result = executor.invoke("计算 37 + 48".to_string()).await?;
println!("结果: {}", result);

对话记忆

use langchainrust::{ChatMessageHistory, Message};

let mut history = ChatMessageHistory::new();

// 添加消息
history.add_message(Message::human("你好！"));
history.add_message(Message::ai("你好！很高兴见到你！"));

// 获取历史
for msg in history.messages() {
    println!("{:?}: {}", msg.message_type, msg.content);
}

Chain 工作流

use langchainrust::{LLMChain, SequentialChain, BaseChain};
use std::sync::Arc;
use std::collections::HashMap;
use serde_json::Value;

// 单步 Chain
let chain1 = LLMChain::new(llm1, "分析以下主题: {topic}");

// 多步顺序 Chain
let chain2 = LLMChain::new(llm2, "根据分析生成总结: {analysis}");

let pipeline = SequentialChain::new()
    .add_chain(Arc::new(chain1), vec!["topic"], vec!["analysis"])
    .add_chain(Arc::new(chain2), vec!["analysis"], vec!["summary"]);

let mut inputs = HashMap::new();
inputs.insert("topic".to_string(), Value::String("2024年人工智能发展".to_string()));

let results = pipeline.invoke(inputs).await?;

RAG 检索增强生成

use langchainrust::{
    Document, InMemoryVectorStore, MockEmbeddings,
    SimilarityRetriever, RetrieverTrait, 
    RecursiveCharacterSplitter, TextSplitter,
};
use std::sync::Arc;

// 创建文档
let docs = vec![
    Document::new("Rust 是一门系统编程语言..."),
];

// 文档分割
let splitter = RecursiveCharacterSplitter::new(200, 50);
let chunks = splitter.split_document(&docs[0]);

// 创建检索器
let store = Arc::new(InMemoryVectorStore::new());
let embeddings = Arc::new(MockEmbeddings::new(128));
let retriever = SimilarityRetriever::new(store.clone(), embeddings);

// 索引文档
retriever.add_documents(chunks).await?;

// 检索
let relevant_docs = retriever.retrieve("什么是 Rust？", 3).await?;

文档加载器

LangChainRust 现在支持从多种格式加载文档，包括 PDF 和 CSV 文件。

PDF Loader

use langchainrust::retrieval::{PDFLoader, DocumentLoader};

// 加载 PDF 文件
let pdf_loader = PDFLoader::new("path/to/document.pdf");
let documents = pdf_loader.load().await?;

// 提取的文档包含文本内容和元数据
for doc in documents {
    println!("Content: {}", &doc.content[..100.min(doc.content.len())]);
    println!("Metadata: {:?}", doc.metadata);
}

CSV Loader

use langchainrust::retrieval::{CSVLoader, DocumentLoader};

// 加载 CSV 文件，指定内容列为"description"
let csv_loader = CSVLoader::new("path/to/data.csv", "description");
let documents = csv_loader.load().await?;

// 每一行数据转换为单独的文档，具有对应元数据
for doc in documents {
    println!("Content: {}", doc.content);
    println!("Row Metadata: {:?}", doc.metadata);
}

📚 完整示例

查看 examples/ 目录：

基础示例

• hello_llm - 基础 LLM 对话
• streaming - 流式输出
• prompt_template - 提示词模板
• tools - 内置工具

中级示例

• agent_with_tools - Agent 工具调用
• memory_conversation - 多轮对话记忆
• chain_pipeline - Chain 工作流

高级示例

• rag_demo - 完整 RAG 流程
• multi_tool_agent - 多工具 Agent
• full_pipeline - 完整 AI 应用

运行示例：

# 无需 API Key

cargo run --example prompt_template

cargo run --example tools


# 需要 API Key

export OPENAI_API_KEY="your-key"

cargo run --example hello_llm

cargo run --example agent_with_tools

🧪 测试

# 运行所有测试

cargo test


# 运行特定模块测试

cargo test prompts:: --lib -- --nocapture


# 显示测试输出

cargo test -- --nocapture

📁 项目结构

src/
├── core/                # 核心抽象
│   ├── language_models/ # 基础 LLM trait
│   ├── runnables/       # Runnable trait
│   └── tools/           # Tool trait
├── language_models/     # LLM 实现
│   └── openai/          # OpenAI 客户端
├── agents/              # Agent 框架
│   └── react/           # ReActAgent
├── prompts/             # 提示词模板
├── memory/              # 记忆管理
├── chains/              # 链式调用
├── retrieval/           # RAG 组件
├── embeddings/          # 文本嵌入
├── vector_stores/       # 向量存储
├── tools/               # 内置工具
└── schema/              # 数据结构

🔧 配置

环境变量

export OPENAI_API_KEY="your-api-key"

export OPENAI_BASE_URL="https://api.openai.com/v1"  # 可选：自定义端点

OpenAIConfig 配置项

🔐 安全提示

• 切勿将 API Key 提交到版本控制
• 使用环境变量存储密钥
• 支持代理/自定义端点

📖 文档

• API 文档
• 示例代码
• 贡献指南

🤝 贡献

欢迎贡献代码！请查看 CONTRIBUTING.md 了解详情。

📄 许可证

Apache License, Version 2.0 或 MIT License，任选其一。

🙏 致谢

本项目受 LangChain 启发，使用 Rust 实现。

English Documentation

✨ Features

Key Benefits

• 🚀 Fully Async - Tokio-based async/await support
• 🔒 Type-Safe - Leverage Rust's type system
• 📦 Zero-Cost Abstractions - High-performance design
• 🎯 Simple API - Intuitive interfaces
• 🔌 Extensible - Easy to add custom tools

📦 Installation

Add to Cargo.toml:

[dependencies]

langchainrust = "0.1.2"

tokio = { version = "1.0", features = ["full"] }

🚀 Quick Start

Basic Chat

use langchainrust::{OpenAIChat, OpenAIConfig, BaseChatModel};
use langchainrust::schema::Message;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let config = OpenAIConfig {
        api_key: std::env::var("OPENAI_API_KEY")?,
        base_url: "https://api.openai.com/v1".to_string(),
        model: "gpt-3.5-turbo".to_string(),
        streaming: false,
        temperature: Some(0.7),
        max_tokens: Some(500),
        ..Default::default()
    };
    
    let llm = OpenAIChat::new(config);
    
    let messages = vec![
        Message::system("You are a helpful assistant."),
        Message::human("What is Rust?"),
    ];
    
    let response = llm.chat(messages, None).await?;
    println!("{}", response.content);
    
    Ok(())
}

Prompt Templates

use langchainrust::prompts::{PromptTemplate, ChatPromptTemplate};
use langchainrust::schema::Message;
use std::collections::HashMap;

// String template
let template = PromptTemplate::new("Hello, {name}! Today is {day}.");
let mut vars = HashMap::new();
vars.insert("name", "Alice");
vars.insert("day", "Monday");
let prompt = template.format(&vars)?;

// Chat template
let chat_template = ChatPromptTemplate::new(vec![
    Message::system("You are a {role} specializing in {domain}."),
    Message::human("Hello, I'm {name}."),
    Message::human("{question}"),
]);

let mut vars = HashMap::new();
vars.insert("role", "Rust expert");
vars.insert("domain", "systems programming");
vars.insert("name", "Bob");
vars.insert("question", "Explain ownership in Rust");

let messages = chat_template.format(&vars)?;

Agent with Tools

use langchainrust::{
    ReActAgent, AgentExecutor, BaseAgent, BaseTool,
    Calculator, DateTimeTool, SimpleMathTool,
};
use std::sync::Arc;

let tools: Vec<Arc<dyn BaseTool>> = vec![
    Arc::new(Calculator::new()),
    Arc::new(DateTimeTool::new()),
    Arc::new(SimpleMathTool::new()),
];

let agent = ReActAgent::new(llm, tools.clone(), None);
let executor = AgentExecutor::new(Arc::new(agent) as Arc<dyn BaseAgent>, tools)
    .with_max_iterations(5);

let result = executor.invoke("What is 37 + 48?".to_string()).await?;
println!("Answer: {}", result);

Memory

use langchainrust::{ChatMessageHistory, Message};

let mut history = ChatMessageHistory::new();

// Add messages
history.add_message(Message::human("Hello!"));
history.add_message(Message::ai("Hi there!"));

// Retrieve messages
for msg in history.messages() {
    println!("{:?}: {}", msg.message_type, msg.content);
}

Chain Pipelines

use langchainrust::{LLMChain, SequentialChain, BaseChain};
use std::sync::Arc;
use std::collections::HashMap;
use serde_json::Value;

// Single chain
let chain1 = LLMChain::new(llm1, "Analyze this topic: {topic}");

// Sequential chains
let chain2 = LLMChain::new(llm2, "Summarize: {analysis}");

let pipeline = SequentialChain::new()
    .add_chain(Arc::new(chain1), vec!["topic"], vec!["analysis"])
    .add_chain(Arc::new(chain2), vec!["analysis"], vec!["summary"]);

let mut inputs = HashMap::new();
inputs.insert("topic".to_string(), Value::String("AI in 2024".to_string()));

let results = pipeline.invoke(inputs).await?;

RAG Pipeline

use langchainrust::{
    Document, InMemoryVectorStore, MockEmbeddings,
    SimilarityRetriever, RetrieverTrait, 
    RecursiveCharacterSplitter, TextSplitter,
};
use std::sync::Arc;

// Create documents
let docs = vec![
    Document::new("Rust is a systems programming language..."),
];

// Split documents
let splitter = RecursiveCharacterSplitter::new(200, 50);
let chunks = splitter.split_document(&docs[0]);

// Create retriever
let store = Arc::new(InMemoryVectorStore::new());
let embeddings = Arc::new(MockEmbeddings::new(128));
let retriever = SimilarityRetriever::new(store.clone(), embeddings);

// Index documents
retriever.add_documents(chunks).await?;

// Search
let relevant_docs = retriever.retrieve("What is Rust?", 3).await?;

📚 Examples

See examples/ for complete code:

Basic

• hello_llm - Basic LLM chat
• streaming - Streaming output
• prompt_template - Using templates
• tools - Built-in tools

Intermediate

• agent_with_tools - Agent with tool calling
• memory_conversation - Multi-turn conversations
• chain_pipeline - Chain workflows

Advanced

• rag_demo - Full RAG pipeline
• multi_tool_agent - Agent with multiple tools
• full_pipeline - Complete AI application

Run examples:

# Without API key

cargo run --example prompt_template

cargo run --example tools


# With API key

export OPENAI_API_KEY="your-key"

cargo run --example hello_llm

cargo run --example agent_with_tools

🧪 Testing

# Run all tests

cargo test


# Run specific module

cargo test prompts:: --lib -- --nocapture


# Show test output

cargo test -- --nocapture

📁 Project Structure

src/
├── core/                # Core abstractions
│   ├── language_models/ # Base LLM traits
│   ├── runnables/       # Runnable trait
│   └── tools/           # Tool trait
├── language_models/     # LLM implementations
│   └── openai/          # OpenAI client
├── agents/              # Agent framework
│   └── react/           # ReActAgent
├── prompts/             # Prompt templates
├── memory/              # Memory management
├── chains/              # Chain workflows
├── retrieval/           # RAG components
├── embeddings/          # Text embeddings
├── vector_stores/       # Vector databases
├── tools/               # Built-in tools
└── schema/              # Data structures

🔧 Configuration

Environment Variables

export OPENAI_API_KEY="your-api-key"

export OPENAI_BASE_URL="https://api.openai.com/v1"  # Optional: custom endpoint

OpenAIConfig Options

🔐 Security

• Never commit API keys to version control
• Use environment variables for secrets
• Support for proxy/custom endpoints

📖 Documentation

• API Documentation
• Examples
• Contributing

🤝 Contributing

Contributions are welcome! See CONTRIBUTING.md for details.

📄 License

Licensed under either of:

• Apache License, Version 2.0
• MIT License

🙏 Acknowledgments

Inspired by LangChain, implemented in Rust.