easy-langent: A Structured LangChain and LangGraph Tutorial from Basics to Multi-Agent Systems

What easy-langent is and who it is for

easy-langent is a structured tutorial from the Datawhale China open-source community. The name combines lang (from LangChain and LangGraph) with agent. Its stated goal is to bridge the gap between conceptual understanding and practical development: the README describes the target user as someone who either cannot start working with agents despite understanding the concepts, or who has learned a framework but cannot apply it to a real project.

The tutorial is organized as a three-part sequence. Part one covers LangChain and LangGraph framework recognition. Part two covers LangChain components and practical exercises. Part three covers LangGraph components and advanced multi-agent workflows.

The prerequisites listed in the README are Python programming basics, a fundamental understanding of large language models, and a basic understanding of what AI agents are. The README points readers who lack these foundations to two earlier Datawhale tutorials: Happy-llm for LLM fundamentals and Hello-Agents for basic agent concepts.

The tutorial is hosted online at datawhalechina.github.io/easy-langent/ with a mirrored version at easy-langent.datawhale.cc for users in mainland China. Both are free and require no account.

Chapter structure and what the curriculum covers

The tutorial has eight numbered chapters plus a preface and an epilogue. All chapters and exercises are marked as completed in the README's status table.

Chapter 1 introduces LangChain and LangGraph frameworks: the README describes the chapter as covering framework overview, environment setup, and a first hands-on experience with the lang ecosystem. Chapter 2 covers LangChain's core components: model invocation, prompt templates, and output parsers. Chapter 3 covers LangChain's advanced components: memory, tool use, and combined practice. Chapter 4 covers application-level system design and RAG practice, building retrieval-augmented generation pipelines with document loaders (using docx2txt and pdfplumber for file handling).

Chapter 5 is a mid-course comprehensive exercise where students design and implement an agent application.

Chapter 6 opens the LangGraph section: stateful workflows, nodes, edges, and state management. Chapter 7 covers multi-agent collaboration and complex flow control. Chapter 8 is a comprehensive practical exercise building a 'who is the spy' social deduction game agent.

The pyproject.toml lists langchain 1.2.13 or newer, langgraph 1.1.3 or newer, langchain-openai 1.1.12 or newer, and langchain-community 0.4.1 or newer as the main dependencies. Document processing uses pdfplumber, pypdf, and docx2txt.

Setting up the environment to run code examples

The project supports two setup approaches. For quick dependency installation:

pip install -r requirements.txt

The requirements.txt pins langchain 1.0.0 or newer, langgraph 1.0.0 or newer, langchain-community 0.4.1 or newer, langchain-openai 1.1.12 or newer, and ipykernel 7.2.0 or newer. The pyproject.toml includes additional document processing packages.

To run the VitePress documentation site locally for offline reading:

npm run docs:dev

Before running any chapter's notebook code, copy the .env.example to .env and fill in your API credentials. The .env.example documents five providers: DeepSeek (platform.deepseek.com), SiliconFlow (cloud.siliconflow.cn), Alibaba Cloud Tongyi Qianwen (Aliyun Bailian console), ZhipuAI (bigmodel.cn), and OpenAI. Each section shows the API_KEY and BASE_URL format specific to that provider.

The project requires Python 3.11 or newer as specified in the pyproject.toml and .python-version file. The repository also uses uv as the package manager, with a uv.lock file present.

Community project examples and what students build

A distinct feature of easy-langent is the collection of community-built projects from two cohort learning runs. These are submitted by students at the end of each organized study group.

From the internal testing cohort, one project is a Werewolf game (from a god's-eye view) built with LangGraph. From the April 2026 cohort, eleven projects are listed. These include a script murder game (four-player version using LangGraph), an intelligent knowledge base Q&A system using LangChain, a medical RAG diagnostic system, an AI interview agent using LangGraph, a customer service ticket processing system using LangGraph, an AI debate competition system, and a smart assistant with memory using LangChain.

These projects illustrate the range of applications the course prepares students for. They also serve as reference implementations: each has its own README in the project/ directory explaining the design and how to run it.

The project list demonstrates the course's emphasis on LangGraph for multi-agent and game-style systems and LangChain for retrieval and data pipeline tasks. Students combining both frameworks appear in several projects.

Limitations and what the tutorial does not cover

The tutorial is written primarily in Chinese. All lesson text, exercises, and README documentation are in Simplified Chinese. The code comments, API calls, and package names are framework-standard English, so reading the code is accessible to English readers, but following the prose explanations requires Chinese proficiency.

The tutorial depends on LangChain and LangGraph specifically. It does not cover alternative agent frameworks such as LlamaIndex, PydanticAI, or AutoGen. Teams that have already standardized on a different framework will not find directly applicable guidance.

LangChain Academy is a well-known alternative for English-speaking developers who want structured LangChain and LangGraph instruction. LangChain Academy is the official course series from the LangChain team. The difference: easy-langent is community-written in Chinese with a more linear chapter structure and explicit mid-course projects, while LangChain Academy is English-first with a different progression and more focus on the latest framework capabilities. easy-langent is better suited for Chinese-speaking learners who want chapter-by-chapter progression with exercises; LangChain Academy is better for English-speaking developers who want material directly from the framework maintainers.

License, maintenance, and attribution

The last push was on 2026-09-09. The project appears actively maintained within the Datawhale community study group model, which organizes periodic cohort runs and accepts project submissions after each run.

The license situation has two components. The pyproject.toml declares Apache 2.0 for the code. The README ends with a Creative Commons BY-NC-SA 4.0 notice for the content: attribution required, non-commercial use only, and derivative works must share under the same license. The CC BY-NC-SA 4.0 restriction applies to the lesson text, documentation, and course materials. Commercial use of the course content requires permission from the authors.

The README credits the project lead as muxxiong (muxiaoxiong on GitHub), a Datawhale member, with a contributing author kemling (Datawhale member). The project is part of the Datawhale open-source education initiative, which produces Chinese-language AI learning materials and runs community study group cohorts where participants work through courses together and submit projects at the end. The repository includes issue and pull request guidance pointing to the Datawhale support team for contributors who need follow-up on proposed changes.

Editorial conclusion

Python developers who have basic LLM familiarity and want a structured path through LangChain components, LangGraph stateful workflows, and multi-agent system design will find easy-langent well-organized: each chapter builds on the previous and ends with an exercise, and the community project examples show realistic outputs from completing the course. The tutorial is written primarily in Chinese, so English-only readers will find the prose less accessible even though the API calls and code are language-independent. Before running the notebook examples, confirm your API provider credentials match the .env.example format; the project supports multiple providers including DeepSeek, SiliconFlow, Alibaba Cloud, ZhipuAI, and OpenAI. The last push was on 2026-09-09.

Frequently asked questions

Official sources

1. datawhalechina/easy-langent on GitHub
2. Issues
3. License: Apache-2.0
4. Project website
5. README