DEV Community

Arjun Shah
Arjun Shah

Posted on

SuperCompress is now on PyPI! pip install supercompress in 1 line

#llm #ai #python #opensource

I just published SuperCompress to PyPI! 🎉

pip install supercompress — that's all it takes.

What is it?

A tiny ~5K parameter CPU policy that scores every line of context for relevance before sending to the LLM. It keeps only what matters for the answer.

The Numbers

  • 65% fewer tokens → same answers
  • 100% oracle recall → never drops the answer line
  • ~60ms CPU latency → no GPU needed
  • Open source → MIT with non-commercial clause

Quick Start 

pip install supercompress

from supercompress import compress
result = compress(context, question)
print(f"Saved {result['kv_savings_pct']}% tokens")
Enter fullscreen mode Exit fullscreen mode

Live Demo

Try the interactive comparison tool: https://supercompress.vercel.app/compare

Or read the technical deep-dive: https://dev.to/arjunkshah/how-i-built-a-prompt-compressor-that-saves-65-on-llm-costs-3m80

GitHub: https://github.com/arjunkshah/supercompress
PyPI: https://pypi.org/project/supercompress/

Top comments (1)

Collapse
 
alexshev profile image
Alex Shev •

Line-level compression is a good direction because context waste is usually structural, not just verbose. The important test is whether the compressor preserves the one boring line that actually answers the question, not just whether the prompt gets shorter.