LangGraph Course Repo — Build LLM Agents Hands-On

What does this agent do, and when should you use it?

This repository is the hands-on companion to the Udemy course "LangGraph – Develop LLM-Powered AI Agents". Work is organized as project branches: project/agentic-rag (a RAG pipeline with document grading, web search and adaptive routing), project/ReAct-agent (a reason-and-act loop built in LangGraph), project/reflection and project/reflection-agent (agents that critique and revise their own output), and project/reflexion-agent (an agent that learns from previous runs). Inside each branch, individual commits map one-to-one to lessons, so you can watch a graph grow with git log --oneline. The runnable entry point is main.py, dependencies are managed with Poetry, and configuration lives in a .env file with OPENAI_API_KEY plus optional TAVILY_API_KEY and LANGCHAIN_API_KEY/LANGCHAIN_TRACING_V2. It is teaching material rather than a packaged product: there is no hosted service or deployment manifest, and behaviour depends on the model keys and local setup you provide.

The repo contains several LangGraph sample projects, each driven by main.py, checked out by branch. The Agentic RAG branch walks the full chain in commits: project setup, folder scaffolding, an ingestion pipeline that loads and embeds data, a Graph State that carries memory between nodes, a Retrieve Node that fetches context, a Grade Docs Node that filters for structured relevance, a Web Search Node that calls the Tavily API, and a Generation Node that handles prompting and LLM calls. The graph is then wired with fan-in, fan-out and conditional edges, extended with Self-RAG so the LLM critiques itself, and finished with an Adaptive Router that selects tools dynamically. The ReAct branch implements the classic reasoning-plus-acting loop; the reflection and reflection-agent branches let an agent critique and revise its own output; the reflexion-agent branch feeds lessons from previous runs back into later decisions. Every path reaches external model and search services through keys in .env, with optional LangSmith tracing.

1. Developers learning LangGraph who want to check out project/agentic-rag and cherry-pick commits to see how a RAG graph is assembled step by step.
2. Engineers who need to demo a self-correcting retrieval QA flow and want to reuse the Grade Docs Node and Self-RAG ideas in a prototype.
3. Students or researchers comparing ReAct against reflection/reflexion loops, who can run main.py on each branch and diff the behaviour.
4. Teams already on the LangChain stack that want a reference for wiring the Tavily web-search tool into multi-step question answering.
5. Instructors preparing a lesson plan who can map each video to a commit listed in the repository's lesson table.
6. Developers who want to observe graph execution locally by enabling LANGCHAIN_TRACING_V2 with LANGCHAIN_API_KEY.

How do you install or deploy this agent?

The repo expects a Python environment with Poetry. Per the README's Quick Start (note the README spells the repo name langgaph-course; use the real clone URL):

# 1. Clone & enter
$ git clone https://github.com/emarco177/langgaph-course.git
$ cd langgaph-course

# 2. Choose a project branch
$ git checkout project/agentic-rag  # for example

# 3. Install deps (Poetry)
$ poetry install

# 4. Run
$ poetry run python main.py

Then create a .env file in the repository root:

OPENAI_API_KEY=...
TAVILY_API_KEY=...          # optional – for web-search lessons
LANGCHAIN_API_KEY=...       # optional – for LangSmith tracing
LANGCHAIN_TRACING_V2=true   # optional
PYTHONPATH=$(pwd)

How do you use this agent?

Once installed, switch to the branch for the project you want and run the entry script:

$ git checkout project/ReAct-agent
$ poetry run python main.py

Use project/agentic-rag for the full RAG pipeline, project/ReAct-agent for the reasoning-and-acting loop, project/reflection and project/reflection-agent for self-critique and revision, and project/reflexion-agent for learning from past runs. To follow the course lesson by lesson, inspect and rewind the history:

$ git log --oneline
$ git checkout <hash>

The web-search lessons additionally require TAVILY_API_KEY; LangSmith tracing requires LANGCHAIN_API_KEY and LANGCHAIN_TRACING_V2=true.

What are this agent's strengths and limitations?

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

How does FollowAgents rate this agent?

FAQ