Over the past few years, the number of available Large Language Models and of providers offering them has grown steadily, with new models appearing regularly. This growth makes it increasingly difficult to keep track of the models deployed on a platform and to select a model that suits a given task. The Logos framework integrates locally running and cloud-based models, but its model lifecycle management remains mostly manual, which makes the process tedious and time-consuming.
This thesis aims to rethink model lifecycle management in Logos: it introduces an overview page that lists the available models together with their key parameters, a structured workflow for requesting and deploying new models, and functionality for deprecating models that see little use. These improvements reduce the effort required to include new models, help users choose a model that fits their task, and align the set of deployed models with actual usage.