dfkhelper/token-goat

Token-Goat is a Python tool that reduces the number of tokens consumed during long AI coding sessions in tools like Claude Code and Codex CLI. Tokens are the unit of text that language models process, longer conversations accumulate more of them, which increases cost and can degrade the model's ability to retain older context. Token-Goat hooks into the AI coding tool at the operating system level, automatically intercepting common sources of waste before they reach the model. It targets three main problems. First, when a screenshot or image is passed to the model at full resolution, it can consume a large amount of context space, Token-Goat automatically compresses images by up to 97.4% before they are sent. Second, when an AI agent re-reads a file it already loaded earlier in the same session, Token-Goat intercepts the request and reminds the agent it already has that content, suggesting a narrower slice instead. Third, when a long session compacts its history into a summary, Token-Goat injects a structured manifest beforehand so the summary preserves information about which files were edited and which symbols mattered. A surgical read command lets the agent pull a single function or Markdown section from a file rather than loading the whole thing, reducing read size by around 85%. Installation requires two commands and runs silently in the background as a service afterwards. It supports Windows, Linux, WSL, and macOS. Everything it writes is documented and reversible with a single uninstall command. The full README is longer than what was provided.