I am Zachary, co-founder of Promptropy - the world’s first prompt compression API. Linked is a demo for our first public endpoint for context-agnostic text compression.

Prompt compression is based on the idea that LLMs don’t need many of the quirks of natural language, allowing them to extract equivalent meaning from a prompt 2-20x smaller than the original one. This allows us to intelligently remove tokens from a prompt—making it nearly unreadable to a human, yet still perfectly clear to the LLM (try this by asking a model like GPT-4 to summarize a long passage vs a compressed one).

Please feel free to try out the demo we linked! We would love the HN community’s feedback on what we have built and what can be better.

Thank you!