tokenpatch helps your existing AI coding app spend fewer frontier-model tokens. The strong model decides; a cheaper executor writes bounded patches and reports cost per applied patch.
Use Codex, Cursor, or Claude Code as your normal coding app. Start requests with tp:; tokenpatch only needs executor credentials.
pip install tokenpatch tokenpatch bootstrap # In Codex, Cursor, or Claude Code: tp: add a status filter. Only modify src/orders.tsx and src/orders.css.
Use this path for terminal workflows, CI, or deterministic support diagnostics.
tp setup tokenpatch init tokenpatch doctor --api tp do "Add a status filter to the order list" --allowed-file src/orders.tsx tp report
Each run checks allowed_files, records usage, and keeps recovery details in the local project.
The report prints a Savings Snapshot with actual cost, baseline cost, savings ratio, applied patches, and cost per applied patch.