What is entroly?

Entroly is an open-source tool that sits between your code and AI assistants like Claude, Cursor, or Codex, doing two things: trimming down how much code gets sent to the AI (cutting token costs by 70-95% according to its benchmarks), and checking AI responses for unsupported claims before they reach you. The first part, called PRISM, intelligently selects the most relevant files and snippets from a large codebase instead of sending everything, and learns over time which files actually matter for your workflow. The second part, called WITNESS, compares the AI's output against the evidence you provided and flags any claims the evidence doesn't support, all locally, in about two milliseconds, without making an additional AI call. You install it with a single pip command, run one command in your project folder, and a dashboard opens in your browser showing what context was selected, why, and how many tokens were saved. It works as a proxy that wraps 37 different AI tools and providers, meaning you don't have to change your existing workflow, just route calls through Entroly. The hallucination-detection component attaches proof certificates to responses so you can audit exactly what evidence supported each claim. The engine itself is built on Rust and WebAssembly for speed, with a Python interface for easy integration.