chop intercepts verbose CLI output before it reaches your AI's context window - compressing up to 90% of tokens without losing signal.
Every verbose tool output your AI agent sees - every stack trace, every test line, every docker table - gets tokenized and sent on every single API call. The waste compounds silently.
chop sits between your terminal and your AI agent as a transparent hook. Install once, compress forever.
Run chop init --global to register a PreToolUse hook. Claude Code, Gemini CLI, Codex - all supported. One command per agent.
Every time your agent runs a shell command, chop intercepts the output before it hits the context window. Pattern-matched against 50+ known command signatures.
A dense, semantically faithful summary replaces the verbose output. Unknown commands auto-detected via JSON, CSV, log-pattern, or table heuristics.
git, docker, kubectl, npm, npx, nx, gh, terraform, aws, az, gcloud, cargo, pytest, rspec, tsc, biome, eslint, and more. Run chop list to see all.
Run chop diff <cmd> to see raw output vs. compressed output side-by-side. Great for auditing and writing custom filters.
Unknown commands handled via heuristic detection for JSON, CSV, tabular data, and structured logs. Compresses what it doesn't know too.
Smart fallbackEvery compression is logged locally. Run chop gain to see your savings over time with session history.
Customize rules, disable commands, or add custom filters per repository. Checked in with your project for team-wide consistency.
ConfigurableWorks with Claude Code, Gemini CLI, Codex CLI, and Antigravity IDE via dedicated init commands. One binary, all agents.
UniversalDownload the binary, run one init command for your agent. That's it - chop starts working silently from the next command forward.
Binary lands in ~/.local/bin (macOS/Linux) or %LOCALAPPDATA%\Programs\chop (Windows)
chop installs a transparent hook that intercepts tool output before your agent's context is updated.
chop tracks every compression in a local SQLite database. Run chop gain to inspect cumulative savings across sessions.