This repository contains the implementation of the ReActXen framework, focused on agent-based design and interaction.
It has been referred to as ReAct++, Enhanced ReAct, etc.

📘 Tutorial PDF

Our work has been accepted at EMNLP 2025 Industry Track 🎉

ReAct Meets Industrial IoT: Language Agents for Data Access
James T. Rayfield, Shuxin Lin, Nianjun Zhou, Dhaval C. Patel

📑 Read the Paper (OpenReview PDF)

If you use this work, please cite:

@inproceedings{patel2025react,
  title     = {ReAct Meets Industrial IoT: Language Agents for Data Access},
  author    = {James T. Rayfield and Shuxin Lin and Nianjun Zhou and Dhaval C. Patel},
  booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track},
  year      = {2025},
  url       = {https://openreview.net/forum?id=luETrQw0j6}
}

The ReActXen framework is built on the Agent-Family that has multiple helper agents as outlined in the following diagram. The ReAct agent supports both text generation and code generation based on the specified action.

• Project Setup Instructions
• Setting Up the LLM
• Getting Ready to use ReActXen
• Hello Word ReActXen
• API Functions
• Publications

To get started with this project, follow these steps:

1. Clone the repository:

   git clone git@github.ibm.com:GenAIApps/ReActXen.git
   cd reactxen

2. Set up a Python virtual environment (Python 3.11 or greater is required):

   python3.11 -m venv reactxen
   source reactxen/bin/activate  # On macOS/Linux
   # .\\reactxen\\Scripts\\activate  # On Windows

3. Install the required dependencies: Simply run the following command to install the package and its dependencies:

   pip install .

4. If you need to clean or remove the Python virtual environment (reactxen), follow these steps:

   deactivate
   rm -rf reactxen

1. Setting Up Environment Variables. Copy the .env_template file to .env:

   cp env/.env_template .env

2. Edit the .env file and replace the placeholders with your actual values:

   WATSONX_APIKEY=your_watsonx_apikey
   WATSONX_URL=https://your-watsonx-url.com
   WATSONX_PROJECT_ID=your_project_id

In the hello_world_math.py demo, the ReActXen framework is used to solve the following mathematical equation:

Question: Find the (r) that satisfies the equation:

[ \log_{16} (r+16) = \frac{5}{4} ]

The question asks the agent to find the value of (r) that satisfies the logarithmic equation.

1. Execute the demo script (id 15 = granite 3.2, ibm/granite-3-2-8b-instruct)

   cd src/reactxen/demo
   python hello_world_math.py --mode code --model_id 15 # for code model
   python hello_world_math.py --mode text --model_id 15 # for text model

• create_reactxen_agent
  Description: Initializes and configures a ReActXen agent.
  Example Usage:

  agent = create_reactxen_agent(question="Find the r...", key="", ...)

• run_reactxen_agent Description: Runs the agent and returns the result. Example Usage:

  agent.run()

  You can view the review agent output here: Review

• export_benchmark_metric Description: Exports benchmark metrics from the agent's run. Example Usage:

  agent.export_benchmark_metric()

  You can view the benchmark metric here: Benchmark

• export_trajectory Description: Exports the trajectory of the agent's decisions. Example Usage:

  agent.export_trajectory()

  You can view the sample exported trajectory here: Trajectory