AI Agent Router

AI-Agent-Router is a demonstration of a multi‑agent conversational router that uses a large language model to classify user messages and route them to specialized agents for a contextual response. It shows how to orchestrate multiple LLM-powered components in a single workflow using LangGraph.

Problem statement

When building chatbots or AI assistants with multiple skills (e.g., therapy vs. logic), it is hard to route each user request to the appropriate agent. Without routing, a generic model may answer inappropriately or inefficiently.

This project demonstrates how to classify user messages into different categories (emotional vs. logical) and invoke the corresponding agent to generate a better response.

Main features

• Message classification: uses an LLM to determine whether a user message is emotional or logical.
• Specialized agents: defines separate agents (therapist and logic) to handle different kinds of queries.
• Graph orchestration: uses LangGraph to build a state graph that routes messages based on classification.
• Extensible architecture: easy to add more agents and classification categories.

Architecture / Workflow

1. Load models – A chat model is loaded through init_chat_model (e.g., Anthropic Claude) using your API key.
2. Define message classifier – A Pydantic model describes the classification result.
3. Build the routing graph – The router uses LangGraph's StateGraph to set up nodes for classification and different agents, with edges based on classification results.
4. Process a message – The chat_once function takes a message, runs the graph, and returns the agent’s response.

For more details, see the comments in main.py.

Tech stack

• Python
• LangChain / LangGraph – for building LLM workflows and the state graph.
• Anthropic Claude (or any chat model) via langchain[anthropic]
• Pydantic – type safety and structured outputs.
• dotenv – for environment variable management.

Installation

Clone the repository and install dependencies:

git clone https://github.com/tahamohmadf19-dev/AI-Agent-Router1.git
cd AI-Agent-Router1
python -m venv venv
source venv/bin/activate  # On Windows use `venv\Scripts\activate`
pip install -r requirements.txt  # or use pip install -e . to install from pyproject

Note: This project uses Python 3.10 or newer. The pyproject.toml lists the required packages.

Environment variables

Create a .env file in the project root with your API keys:

ANTHROPIC_API_KEY=sk-...

You can adjust the model name or provider in main.py (e.g., anthropic:claude-sonnet-4-5).

How to run

After installing dependencies and setting environment variables, you can test the router by starting a Python session:

from main import chat_once
response = chat_once("I feel very stressed. What should I do?")
print(response)

The function will classify the message and route it to the appropriate agent.

Usage example

Here is a sample interaction:

>>> chat_once("I'm worried about my future and need some advice.")
# Therapist agent responds with empathetic guidance.
>>> chat_once("What is the capital of France?")
# Logical agent returns a factual answer: "The capital of France is Paris."

Future improvements

• Add more classification categories and agents (e.g., coding assistant, math tutor).
• Integrate a vector database and retrieval‑augmented generation to ground responses in your documents.
• Build a Streamlit or FastAPI interface for interactive chat.
• Persist conversation histories in a secure database.

License

This project is provided for educational purposes under the MIT license.