GitHub Issue

Caveman Memory, Semantic Compression for Persistent Context Files

Discovered On Apr 5, 2026
Primary Metric open
**What you want** Add Caveman Memory, a lossless semantic compression feature for persistent context files (CLAUDE.md, .claude.md, skills). Provide a CLI like caveman compress that reduces token usage while preserving meaning. **Before/after example** ``` Before: This project uses React with TypeScript for the frontend. Please always use functional components with hooks. After: React + TypeScript frontend. Functional components + hooks only. ``` **Why good** Productive, Non-gimmick technique to reduce repeated input tokens on every request, saving large amounts of context space across sessions, lowering cost, and improving efficiency without losing information.
View Raw Thread