This repository presents an intelligent customer support agent powered by LangGraph—a robust tool designed for building intricate workflows with language models. The agent is programmed to classify customer queries, evaluate sentiment, and deliver suitable responses or escalate inquiries when necessary. In this version, the agent leverages the browser-use library to consult online resources and uses a Gradio interface for testing.

Efficient and accurate customer support is essential in today’s competitive business landscape. By automating initial customer interactions, response times can be significantly shortened, thereby enhancing customer satisfaction. This project demonstrates how advanced language models combined with a graph-based workflow can be used to create an effective and scalable support system capable of handling diverse customer concerns.

• State Management: Utilizes TypedDict to track and manage the state of each customer interaction.
• Query Categorization: Classifies queries into categories such as Technical, Billing, or General.
• Sentiment Analysis: Assesses the emotional tone of customer queries.
• Response Generation: Constructs responses based on query category and sentiment.
• Escalation Protocol: Automatically escalates queries with negative sentiment to a human agent.
• Browser-Based Assistance: Employs the browser-use library to consult online sources for informed, up-to-date responses.
• Workflow Graph: Leverages LangGraph to build an adaptable and extendable workflow.

1. Setup: Install the necessary libraries and prepare the Python environment.
2. State Design: Create a structure to store query data (category, sentiment, response).
3. Node Functions: Define functions for categorization, sentiment analysis, and generating responses.
4. Graph Design: Use StateGraph to establish the workflow, incorporating nodes and edges for the complete support process.
5. Routing Logic: Conditionally route the query based on its category and sentiment.
6. Graph Compilation: Compile the workflow into an executable application.
7. Execution: Process customer queries through the workflow and present results via the Gradio UI.

The following diagram illustrates the workflow of the customer support agent—from query categorization and sentiment analysis to routing and escalation:

This project showcases the versatility and power of LangGraph in creating AI-driven workflows. By merging natural language processing with graph-based techniques and augmenting capabilities with browser-based data fetching, we’ve built a customer support agent capable of efficiently addressing a wide range of queries. The system is easily extendable and can be integrated into existing support pipelines or databases.

Follow these steps to set up the project and run the customer support agent application.

Clone this repository to your local machine:

git clone https://github.com/alphatechlogics/LangGraphAgent.git
cd LangGraphAgent

We recommend using uv to set up your Python environment:

uv venv --python 3.11
source .venv/bin/activate

Note: The prompt should now show your virtual environment (e.g., (LangGraphAgent)).

source venv/bin/activate

Install the necessary packages. We use uv pip to ensure packages are installed within the virtual environment:

uv pip install langgraph langchain_openai gradio
uv pip install --pre browser-use
playwright install

Create a .env file in the project root directory and add your OpenAI API key. This is required to interact with the OpenAI models:

OPENAI_API_KEY=your_openai_api_key_here

After installing the dependencies and setting up the .env file, you can run the application:

python customer_support_gradio.py

Once the app is running, open your browser and navigate to the URL shown in the terminal (typically http://127.0.0.1:7860) to interact with the customer support agent.