AI BriefingClaude Code Token Usage Cut by 50%
Key point
By reducing repeated reads and verbose output in Claude Code, measured savings of 50%+ input and 20%+ output tokens were achieved.
Details
Built a token optimizer for Claude Code to reduce the cost of repeatedly reading the same files and the cost of prose that bloats responses.
It consists of three parts.
- caveman: Injects output compression rules via the
UserPromptSubmithook, keeping code, errors, commits, and PRs while removing only chit-chat and euphemisms. - OpenWolf: Wraps the
Readhook to replace repeated file reads with cached descriptions, and persists project memory and buglog across sessions. - Dashboard: At
http://127.0.0.1:3847, records per-session token usage as local JSON, separating the prose axis fromtool_usetokens.
A lower-bound estimate from n=17 sessions showed 50%+ savings in input tokens, 20%+ reduction in output, and 35%+ savings when isolating just the prose axis.
Installation is done by pasting a one-line command into a Claude Code session; it backs up ~/.claude/settings.json, verifies with npm run doctor, and proceeds without root privileges, according to the announcement.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Atlas: Open-Source Source Control Tool for Multi-Agent Coding Workflows
Foremerge v0.5.0 Released: Rust-Based Open-Source Tool Detects Intent Conflicts Between Parallel AI Coding AgentsToken-Space Font Compiler Generates Fonts with Fixed Token WidthsWhiteboard Launches Open-Source VS Code Fork for Human-AI Collaborative Software Design