caveman-compress

To reduce token usage for markdown memory files, you can compress natural language prose into a concise caveman-speak format. This minimizes input token consumption while preserving all code blocks, technical paths, and markdown structural integrity.

Yes, compressing CLAUDE.md files safely preserves code blocks, URLs, file paths, and markdown structure. The compression strictly targets natural language prose, ensuring all technical substance and instructions remain fully intact for your AI assistant.

Caveman-speak compression is a technique that transforms verbose natural language in markdown memory files into a highly concise format. It shrinks documentation and preference files to save input tokens while retaining their original technical meaning and instructions.

Markdown token optimization can be performed utilizing either the Anthropic API or a local CLI. Both processing methods apply automated validation and targeted error correction to compress your memory files into the concise caveman-speak format.

Compressing project memory files reduces input token overhead by approximately 46% on average. This shrinks large project context files to a fraction of their size, ensuring your context fits better within model limits.

The main limitation of compressing markdown into caveman-speak is that it targets only natural language prose. It strictly avoids altering code blocks and structural elements, meaning token reduction is limited to the prose sections of your markdown files.