Hi Product Hunt! 👋 I built ecotokens after noticing that large command outputs — git diffs, cargo build logs, pytest results — were eating Claude Code's context window fast. The result in production (2 weeks of real use): • 4 129 hook executions • 6 714 085 tokens saved (89.6 % overall reduction) • git diff alone: 1 682 049 → 782 tokens on a single command How it works: a PreToolUse / BeforeTool hook intercepts every shell command, applies a family-specific filter (not just truncation!), and returns the compressed output to the model. Savings are recorded to a local JSONL file and displayed in a live ratatui dashboard. Supported AI CLIs: Claude Code, Gemini CLI and Qwen Code. One install command: `ecotokens install` MIT license, built in Rust. Happy to answer questions — feedback very welcome!

EcoTokens

Save 90% of your AI coding tokens — automatically

Save 90% of your AI coding tokens — automatically