deepagents-in-action
📚 《Deep Agents 实战》—— LangChain 官方大使出品,基于 LangChain / LangGraph 生态,从零构建生产级 AI Agent 的完整指南
What is deepagents-in-action?
📚 《Deep Agents 实战》—— LangChain 官方大使出品,基于 LangChain / LangGraph 生态,从零构建生产级 AI Agent 的完整指南
Overview of deepagents-in-action
What it is
Deep Agents in Action (《Deep Agents 实战》) is a Chinese-language open-source course authored by 沧海九粟, a LangChain official ambassador and LCAE-certified engineer. The course is built on the LangChain / LangGraph ecosystem and provides a complete guide to building production-grade AI agents from scratch. It includes 17 chapters with text, experiment templates, and ongoing video supplements. The course starts with the first agent and progressively covers task planning, file system operations, sub-agent collaboration, and context management. The current recommended version is Deep Agents 0.7.x. The course is hosted on GitHub under the DataWhale organization and has a companion website for browsing chapters.
Key features
- 17 chapters with detailed text and experiment templates, with videos being added continuously.
- Based on the LangChain / LangGraph ecosystem, covering both framework and harness concepts.
- Covers a wide range of topics: virtual file system, task planning and decomposition, sub-agents (synchronous and asynchronous), Skills, long-term memory, human-in-the-loop, sandbox execution, file system permissions, MCP protocol, grading rubrics, streaming, interpreters, dynamic sub-agents, and decision model harness.
- Provides ready-to-use experiment templates for each chapter, created via the AgentSeek tool.
- Model inference is supported by SiliconFlow (硅基流动), a one-stop large model cloud service platform.
- Open-source and free to use; no license is explicitly stated in the repository metadata.
Use cases
This course is designed for developers and AI enthusiasts who want to learn how to build production-grade AI agents using the LangChain / LangGraph ecosystem. It is suitable for:
- Understanding the architecture and execution loop of agent frameworks.
- Implementing task planning, file system management, and sub-agent delegation.
- Adding advanced capabilities such as long-term memory, human approval workflows, and sandboxed code execution.
- Experimenting with streaming, interpreters, and dynamic agent orchestration.
- Using MCP (Model Context Protocol) to extend tool ecosystems.
Getting started
To start with the course, install AgentSeek and create a template for the desired chapter. Example commands: bash uv tool install --upgrade agentseek agentseek create deepagents/default --checkout main --no-input After entering the generated directory, view available tasks, install dependencies, configure .env, then check the environment and start the application: bash agentseek task --list # After completing dependency installation and .env configuration agentseek doctor agentseek dev For local development of the course website itself (if you want to contribute or preview), Node.js ≥ 22.12.0 is required: bash npm install npm run dev See the repository README for detailed instructions on each chapter's template and requirements.
License and pricing
The repository is marked as open source (pricing_type: open_source). No specific license is stated in the metadata or README. The course is free to use. Model inference costs may apply when using SiliconFlow; the course provides a new user coupon and an experiment quota subsidy pool as described in the README.
Based on the project's public README and repository metadata.