Every time you paste a file into Claude, ChatGPT or Gemini, you're spending tokens on verbose headers, repeated column names, XML namespaces and whitespace that the AI doesn't need.

TokenPinch converts your files to a .pinch format — a token-optimized structure inspired by TOON (Token-Oriented Object Notation) — and generates a ready-to-paste prompt with the compressed content already included.

How it works: - CSV/Excel: converts to TOON tabular notation with header aliasing ("transaction_date" becomes "tdat", etc.) - JSON arrays: flattened to table format - XML: strips namespaces and empty tags - Includes a self-describing [AIX-FORMAT] header so any AI knows how to decode it

Built this over a weekend to scratch my own itch. Curious what the HN crowd thinks — especially whether the .pinch format is worth standardizing or if there's a better approach I'm missing.

Future solutions include on demand API conversions services that could be use by dev community in order to improve tokens for dev purposes.

This is focus in optimization for that we call Tokens Economy.