dannyeldridge54/aegis

AEGIS is a TypeScript tool that helps you find the best settings for something you are trying to improve, without needing to know much about optimization math. You give it a function that scores how good a set of parameters is, along with the range each parameter can take, and AEGIS searches for the combination that gets the best score. A basic example in the README shows tuning a machine learning model's learning rate, dropout, and number of layers in just a few lines of code. What sets AEGIS apart from similar tools like Optuna, Ray Tune, or Hyperopt is that it tries to pick its own search strategy instead of making you choose one. It includes six built in strategies, including evolutionary search, gradient based search, simulated annealing, and random search, and it uses a method called a multi armed bandit to automatically switch between them based on what is working for your specific problem. As it searches, it moves through phases such as exploring, exploiting, being curious about unexplored areas, and finally converging on a result. It also has a discovery engine that flags notable moments, like a big jump in performance or hitting a boundary in the search space. Other features include support for output messages in eight languages, an event system so you can react to progress in real time, and a daemon mode that lets AEGIS run continuously in the background managing multiple optimization jobs, complete with a small web dashboard and an HTTP API for checking status. The README lists possible use cases such as tuning machine learning models, optimizing server settings, adjusting trading strategy parameters, and tuning game AI behavior, essentially any situation where you have adjustable numbers and a way to score the outcome. Installation is a simple npm install, and the project offers a free tier for students with a .edu email, alongside paid individual, team, and enterprise pricing plans for continued use.