a workshop for learning LangGraph, in public

Forging intelligent agents, one graph at a time.

GraphForge is a hands-on workspace for learning how AI agents actually work — notebooks, agent implementations, and MCP experiments, built by following one idea: learn by building, not by reading about it.

View on GitHub See the roadmap
Python · uv LangChain · LangGraph MIT licensed
agent looplanggraph.json
start Agent reasons + calls tools Tool executes + returns end tool call loop back with result

Why this repo exists

Don't just use agents — understand how they work.

Every notebook, experiment, and agent here is another piece of the puzzle toward understanding how reliable AI systems get designed and built. The repository grows the same way the understanding does.

Learn→ Build→ Experiment→ Break→ Debug→ Understand→ Forge

Learning path

Five stages, worked through in order

Each stage builds on the last — the repository's notebooks and agents are organized to follow this same progression.

01

Foundations

LangChain basicsLangGraph fundamentalsModelsMessagesToolsStateGraphsNodes & edges
02

Agent workflows

Conditional routingAgent loopsTool executionState managementStreamingCheckpointsMemory
03

Advanced agents

Multi-agent systemsHuman-in-the-loopMiddlewareOrchestrationLong-running workflowsError handlingDynamic workflows
04

MCP

Model Context ProtocolConnecting external toolsShared resources
05

Production

ObservabilityPersistenceEvaluationDebuggingDeploymentScalable architectures

Inside the repo

Where everything lives

GraphForge/
├── agents/          # agent implementations
├── notebooks/       # learning notebooks and experiments
├── mcp/             # Model Context Protocol experiments
├── utils/           # shared utilities and helpers
├── images/          # supporting assets
├── langgraph.json     # LangGraph configuration
├── pyproject.toml     # project config + dependencies
└── .env.example       # environment variable template

Built with

The tools doing the work

Python
Primary language for every agent and notebook
LangChain
LLM application framework underneath the agents
LangGraph
Stateful, graph-based agent workflows
MCP
Standardized tool and context integration
uv
Python package and environment management
Jupyter
Where the experiments actually happen
LangGraph Studio
Visualizing and debugging agent graphs

Run it yourself

From clone to a running agent

Dependencies are managed with uv. Once they're installed and your environment file is filled in, any agent wired up in langgraph.json is one command away.

bash~/GraphForge
# clone and enter
git clone https://github.com/ATLURI0001/GraphForge.git
cd GraphForge

# install dependencies with uv
pip install uv
uv sync
source .venv/bin/activate

# set up environment variables
cp .env.example .env
# → add your API keys before running anything

# start the LangGraph dev server
langgraph dev

Where things stand

Progress so far

This checklist moves as the repository grows — it's the actual state of the project, not a roadmap promise.

Project setup
LangGraph environment
LangGraph fundamentals
Agent workflows
Tool calling
State & memory
Human-in-the-loop
Multi-agent systems
MCP integrations
Advanced agent architectures
Production workflows

Reading list

Where the ideas come from

LangChain documentationdocs.langchain.com LangGraph documentationdocs.langchain.com Model Context Protocolmodelcontextprotocol.io LangGraph Studiodocs.langchain.com LangChain Academyacademy.langchain.com

MIT licensed — fork it and start forging your own.

Open an issue, follow along with the notebooks, or use it as a template for your own LangGraph experiments.

View on GitHub Browse notebooks