用 LangChain、LangGraph 和 Groq 构建 AI 智能体的 TypeScript 实战项目

A hands-on TypeScript project for learning how to build AI agents with LangChain, LangGraph, and Groq. Each exercise is a runnable example of an agentic workflow.

Exercises

Joke Improver Agent

src/Joke-Improver-Agent-Exercise/ contains an agent that evaluates a joke, improves it when necessary, and evaluates the revised version again. It uses a LangGraph feedback loop, structured LLM output, and an in-memory cache.

Weather Agent

src/Weather-Agent-Exercise/ contains a weather assistant that lets the LLM choose between tools for current conditions and tomorrow's forecast. It uses WeatherAPI for weather data, LangGraph's ToolNode for tool execution, and in-memory caching for LLM and tool calls.

Prerequisites

Node.js 18 or later
npm
A Groq API key
A WeatherAPI key for the Weather Agent exercise

Setup

npm install

Create a .env file in the project root:

GROQ_API_KEY=your_groq_api_key
WEATHERAPI_API_KEY=your_weatherapi_key

GROQ_API_KEY is required by both exercises. WEATHERAPI_API_KEY is required only by the Weather Agent.

Run an Exercise

Run the Joke Improver Agent:

npm run dev -- src/Joke-Improver-Agent-Exercise/index.ts

Run the Weather Agent:

npm run dev -- src/Weather-Agent-Exercise/index.ts

Type-check the project:

npm run build

Project Structure

ai-agent-practice/
├── src/
│   ├── Joke-Improver-Agent-Exercise/
│   │   ├── index.ts                 # Joke evaluation and improvement graph
│   │   └── instruction.md           # Exercise requirements and guidance
│   └── Weather-Agent-Exercise/
│       ├── getWeatherInfo.ts        # WeatherAPI request helpers
│       ├── index.ts                 # Tool-calling weather-agent graph
│       ├── instruction.md           # Exercise requirements and guidance
│       ├── prompts.ts               # System and sample user prompts
│       └── urls.ts                  # WeatherAPI URL builders
├── .env                             
├── .gitignore                       
├── package.json                     
├── package-lock.json                
├── tsconfig.json                    
└── README.md

Generated or local-only directories such as node_modules/ and dist/ are intentionally excluded from version control.

Tech Stack

LangGraph for stateful agent workflows
LangChain for LLM integrations and tools
Groq for language-model inference
WeatherAPI for weather data
TypeScript and tsx for development and execution

Notes

The examples use Groq's openai/gpt-oss-120b model.
Both exercises run sample requests defined directly in their index.ts files.
Read each exercise's instruction.md before modifying its implementation.

License

ISC