ATHENA: A customizable LLM-based Code Completion Tool for Visual Studio Code

The integrated use of Large Language Models in IDEs has obtained increasing popularity given their outstanding performance in coding tasks. This leads to AI-powered code completion tools (CCT) that promise to improve developers’ performance, accuracy, and productivity when programming. However, designing these tools is not easy, mainly due to a lack of consideration of developers’ mental models when interacting with AI-based systems. Based on a study conducted with 56 users to elicit the developer’s mental models when using CCTs, we developed ATHENA, a prototype of CCT for Visual Studio Code based on OpenAI’s GPT-4o mini. The prototype implements a high degree of customization in its behavior to accommodate the mental models of developers with diverse expertise and needs.