Current dynamic Large-Language Model (LLM)-based Multi-Agent Systems (MAS) generally assume that agents will always be available to perform work. However, in real-world systems, disconnections and timeouts may occur causing agents to go down, suspending execution and resulting in system failure. A robust, dynamic MAS should consider the possibility of unavailable agents and have a mechanism in place to remain available and successfully produce an output. We propose a framework that prompts an available agent when a disconnection occurs to decide which other available agent within the system is best suited for the given task. That determined agent is then prompted with this new task, and workflow continues normally. This repo contains all the source code for this implementation using python's langgraph library.
Install the source code using git clone
$ git clone https://github.com/john-bris/MAS_BASE
Enter the directory of the project
$ cd MAS_BASE
Install dependencies
$ pip install -r MAS_CODE/requirements.txt
Run the project
$ python MAS_CODE/qa_main.py
Note: agents are run via ollama running locally on the system. Ollama must be downloaded (following these instructions) and running, along with the corresponding LLM downloaded on ollama.
To be added