UpdAgent: AI Agent Version Control Framework for Real-Time Updation of Tools

Due to the growing usage of agents in artificial intelligence (AI), the versioning of the tools has become essential. New tools or their versions are introduced regularly. Older versions eventually become obsolete. Servers often face maintenance downtime, which obstructs APIs from being used. Version management is common in software development. Similarly, this paper introduces UpdAgent, a version control system to manage the versions of tools used by AI-driven applications that are based on agents and Large Language Models (LLMs). This centralized management system allows the tool providers to deliver real-time updates to add or improve functionalities, resolve issues in existing tools, and immediately revert updates that generate new bugs or errors. The system is designed to streamline real-time updates to tool functions and APIs, ensuring that the LLMs utilize new or updated tool functionalities and avoid APIs that are under maintenance or obsolete. Automated testing is performed to automate the avoidance of tools that produce errors. The management results in automated adaptation to updated tools and reduced delays in the AI-based applications to enhance system reliability. The experiment was successful in setting up the tables and updating the tools using new data. The code is available at github.com/Pro-GenAI/UpdAgent.