caveman-compress

You can compress natural language memory files into caveman format to reduce Claude session input tokens. The process creates a .original.md backup and validates output with retries to ensure content fidelity.

Caveman compression shrinks markdown memory files into a caveman format to minimize context length and per-session costs. It performs type detection and safe compression while preserving URLs, headings, and file structure.

Yes, compressing markdown memory files requires Python and the anthropic and tiktoken dependencies. You can operate the compression locally via a CLI workflow or through a programmatic API fallback.

To compress local project notes, run the CLI workflow by navigating to the caveman-compress directory and executing python3 -m scripts. This produces a compressed version of your target memory file alongside a backup.

Compressing markdown memory files applies targeted fixes that preserve URLs, headings, and file structure. It validates results with retryable quality checks to ensure content fidelity is maintained.

The best way to back up markdown files during token reduction is automatic .original.md backup creation. The compression process safely generates this backup before applying any caveman format changes to your memory files.