langchain-ai/langgraph-101

What it solves

Building LLM agents that reliably handle real-world tasks is hard because you often need fine-grained control over their behavior. This repository exists to teach developers how to build such agents using LangChain, LangGraph, and Deep Agents. LangGraph provides a framework for adding precision and control into agent workflows, and Deep Agents adds built-in tools, context management, and skills for complex multi-step tasks. The tutorials show how to move from simple agents to production-grade systems with human oversight and multi-agent collaboration.

How it works

The project is organized into two learning tracks. The 101 track covers fundamentals: building a first agent with models, tools, memory, and streaming, then exploring middleware, human-in-the-loop patterns, and guardrails. The 201 track goes deeper into production patterns: a stateful email triage agent, multi-agent systems with supervisors and specialized sub-agents, a deep research agent with parallel sub-researchers, and Deep Agents built from scratch with AGENTS.md, skills, backends, long-term memory, and human-in-the-loop interactions. In addition to notebooks, the repository includes standalone agent implementations (weather, email, music store, researcher, deep agent) that can be run locally via langgraph dev and LangGraph Studio. All models are configured centrally in utils/models.py, supporting OpenAI, Azure OpenAI, AWS Bedrock, and Google Vertex AI.

Who it's for

This is aimed at developers who want to learn how to build LLM agents in a structured, hands-on way. It is especially useful for those interested in the LangChain ecosystem, including LangGraph and Deep Agents. The material is a condensed version of LangChain Academy, designed to be run in a session with a LangChain engineer, but it also works for self-paced study if you are comfortable setting up a Python environment and API keys.

Highlights

• Two clear learning paths: 101 for fundamentals, 201 for production patterns
• Hands-on notebooks that build agents incrementally with modern primitives like create_agent() and create_deep_agent()
• Standalone agent examples you can run in LangGraph Studio: weather, email triage, multi-agent music store, deep researcher, and a DeepAgents research agent
• Covers advanced topics: middleware, human-in-the-loop, multi-agent supervision, long-term memory, skills, and parallel sub-agents
• Single configuration for multiple LLM providers (OpenAI, Azure, Bedrock, Vertex AI) to switch easily
• Uses langgraph dev for local development with hot-reloading and a visual debugger