Published: 2026-07-31

freeCodeCamp's free agentic AI course: LangGraph, memory, RAG and deployment

Chapters / key moments (click to jump — plays here on the page)

freeCodeCamp has released a full-length, free course on building production-ready multi-agent systems with LangChain and LangGraph. It runs in seven phases — from what actually separates an agent from a plain LLM app, through async Python and Pydantic validation, LangGraph state graphs, memory, RAG and human-in-the-loop controls, to Dockerising the result and deploying it on AWS and Render. It closes with three end-to-end projects rather than isolated snippets.

Source video

"Agentic AI – Complete Course for Beginners" by freeCodeCamp — Watch on YouTube →

Key Takeaways

Seven phases, deliberately sequenced. Agentic fundamentals → asynchronous Python and Pydantic → LangChain → LangGraph → memory, monitoring and autonomy → deployment and production engineering → three real projects. The framing throughout is that agents differ from earlier LLM and RAG apps because the model decomposes a goal into tasks and chooses its own tools.
LangChain is taught first on purpose. The instructor's argument is that LangGraph is a LangChain product and you still use LangChain for the surrounding work — loading models, prompt templates, utility code — while LangGraph handles only the agent workflow graph. Skipping LangChain leaves gaps you hit later.
The LangGraph section is the core. It covers state management, nodes and edges, conditional edges, start/end nodes, graph compilation, state graphs, checkpointers and message state, then works through sequential, parallel, conditional and iterative workflows before building an agentic chatbot.
Async is treated as a prerequisite, not an optimisation. The course spends a phase on asynchronous programming specifically because LangGraph relies on it internally — that is what lets agents run in parallel — alongside Pydantic for data and model validation.
Production phase is unusually complete for a free course. Dockerising the agents, a FastAPI backend, database setup, GitHub Actions CI/CD, AWS and Render deployment, environment variables, production folder structure, logging and monitoring — plus agent tracing and debugging with LangSmith in the preceding phase.
Three capstone projects. An end-to-end agentic chatbot (LangGraph, database, LangSmith, tools, RAG, human-in-the-loop, AWS and Render); a ChatGPT-style agent (LangGraph, FastAPI, Chroma, SQLAlchemy, AWS); and "TripMate AI", a multi-agent trip planner on Groq, LangGraph, PostgreSQL and FastAPI.
Stated prerequisites. Advanced Python and generative-AI basics, with Anaconda, VS Code, Git and GitHub, Docker Desktop and Postman installed before starting.

How this relates to agent platforms we cover

This course builds agents from framework primitives rather than using a packaged agent runtime. If you want the same capabilities — persistent memory, tool use, human approval gates — without writing the graph yourself, that is roughly the job that OpenClaw, Hermes and Kilo Code do out of the box. See our comparison guide for where a framework build makes more sense than an off-the-shelf agent.

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Configuration shown

Prerequisites and tools the course lists

Comfort with advanced Python and basic generative-AI knowledge (what a large language model is). Software: Anaconda, VS Code, Git and GitHub, Docker Desktop, Postman.

Course map (seven phases)

1 Agentic AI fundamentals (reasoning, planning, memory, tools, architecture, human feedback). 2 Async Python and Pydantic validation. 3 LangChain (model loading, prompt templates, utilities). 4 LangGraph (state, nodes, edges, conditional edges, start and end nodes, graph compilation, StateGraph, checkpointer; sequential, parallel, conditional and iterative workflows). 5 Memory, tools, RAG with a vector database, human-in-the-loop, tracing and debugging with LangSmith. 6 Deployment: Docker, FastAPI backend, database, GitHub Actions CI/CD, AWS and Render, environment variables, logging. 7 Three projects.

The three projects

1) End-to-end agentic chatbot with LangGraph, a database, LangSmith, tools, RAG and human-in-the-loop, deployed on AWS and Render. 2) A custom ChatGPT-style agent with an LLM, LangGraph, FastAPI, Chroma, SQLAlchemy and AWS. 3) TripMate AI: a multi-agent trip planner with Groq, LangGraph, PostgreSQL and FastAPI.

Template shown in the video. Adapt to your project — do not copy verbatim without reviewing each line.

Reproducible steps

Why agents instead of plain RAG, as the instructor explains it
A model has a knowledge cutoff (he shows the GPT-4.1 model page listing January 1, 2024). RAG fixes stale knowledge by adding a vector database the model searches, but it fails on constantly changing data such as news or weather, since someone has to keep filling the database. An agent instead gets tools, such as a web search tool, and decides which to call, so it can fetch live information.
Agent definition used in the course
An AI that takes a goal, plans, acts through tools, adapts to change and asks for help only when necessary, with minimal human guidance.

Gotchas

The transcript is only the course introduction and the first lesson; no code or commands appear in it.
Free course from freeCodeCamp, taught by Pappy (the captions garble the instructor's name).