What is llm-token-governor?

Token Governor is a tool for people building software on top of AI chat models like Claude, ChatGPT, or Gemini. Its purpose is to reduce how many tokens, the small chunks of text these models are billed by, a prompt uses while still getting a correct answer, and it works with Anthropic, OpenAI, Gemini, and any other provider that speaks the OpenAI style API. It ships as a small web app with a single screen dashboard. A person pastes in a prompt, sets some constraints like how strict the output shape should be, and the app shows a projected savings estimate along with a verdict on whether shrinking the prompt is actually worth it, since sometimes the extra setup costs more tokens than it saves. There is also a batch mode that scores a whole list of prompts at once, flags the ones that do not really need a model at all, and exports the results as a spreadsheet style file. A live A B mode sends the same prompt both the normal way and the shrunk way through the tool's own server and reports the real token counts each one used, using whichever AI provider the person has configured. Beyond the dashboard, the same server can act as a gateway that a person's existing app points at directly, so every request their app makes gets these token saving rules applied automatically, then streamed back exactly as usual. Support for a given provider mostly comes for free since almost every provider now offers an interface compatible with OpenAI's format, so switching providers is often just changing a web address and a model name in a settings file. Getting started requires Node.js 18 or newer, cloning the project, installing its dependencies, and adding an API key from at least one AI provider to a settings file, needed only for the live comparison feature. The project has 2 GitHub stars and is written in JavaScript.