What is overllm?

overllm is a code checking tool that looks through your codebase to find places where you called an AI model, like GPT or Claude, to do something that plain code could already do just as well. If you asked an AI model to sort a list, extract a date from text, or parse JSON, those are things a normal programming function handles instantly and for free, without the delay, cost, or unpredictability of calling an AI service. It works by reading your source code with a real code parser rather than by running any AI model itself. For Python it uses the language's own built-in parser, and for JavaScript and TypeScript it uses a tool called tree-sitter. Because there is no model involved, it runs quickly, needs no internet connection or API key, and always gives the same result for the same code, which makes it fast enough to run automatically before every commit. When you point overllm at a file or folder, it prints out a short list of problems it found, such as an AI call being made inside a loop when it could be batched into one request, or a call whose only job is a mechanical task like sorting or counting. It stays quiet on clean code, meaning a well-written project will often produce no output at all, and it exits with an error code when it finds something, so it can block a commit or fail a CI check if you want that. overllm never sends your code anywhere. All of the analysis happens on your own machine, with no network calls and no telemetry, and the core tool is only a couple thousand lines of Python with no required external dependencies, so it can be read and verified easily. It recognizes AI calls from providers including OpenAI, Anthropic, Google, Mistral, Cohere, Groq, and several others, across both Python and JavaScript codebases. For an existing project with many pre-existing issues, overllm can save a one-time snapshot of current findings so that future checks only flag new problems rather than every old one at once. It can also automatically fix a small number of the safest, most obvious issues on its own when you run it with a fix option. It is installed with a single pip command and configured through a small settings file in your project.