LangGraph Email Agent

A learning project that builds an email-processing agent with LangGraph.

What it demonstrates

• A model node that reasons over MessagesState.
• Tool calling with bind_tools() and ToolNode.
• Explicit ReAct-style routing: model -> tools -> model.
• Human approval with interrupt() before side-effecting actions.
• Durable SQLite checkpoints with SqliteSaver.
• A nested notice-extraction graph with its own domain state and approval step.

Project layout

chains/     LangChain structured-output pipelines
graphs/     Main email graph and notice subgraph
utils/      Simulated email and legal-ticket integrations

Setup

Use Python 3.11 or newer and create a virtual environment:

python -m venv venv
.\venv\Scripts\python.exe -m pip install -r requirements.txt

Create a .env file with the model settings expected by ChatOpenAI:

LLM_MODEL=your-model
LLM_API_KEY=your-api-key
BASE_URL=https://your-openai-compatible-endpoint/v1

Run

From the Langgraph directory:

.\venv\Scripts\python.exe -m graphs.run_email_agent

The runner processes the first sample email, pauses for approval when needed, resumes using the same thread_id, and prints the checkpointed state.

Architecture

email_agent -> approval -> email_tools -> email_agent
                            |
                            +-> extract_notice_data -> notice extraction graph

Graph Visualizations

Email Agent Graph

Notice Extraction Graph

The SQLite checkpoint files, .env, and virtual environment are local runtime artifacts and are excluded from Git.