Skip to content
tokz

Guide

What is prompt compression?

Prompt compression reduces the context sent to a language model by selecting useful source material. Tokz's structural path returns offsets into the original payload; its prose path returns extractive chunks with provenance.

Compression is not summarization

A summary generates a new representation. Extractive compression selects source spans, so the retained bytes can be checked against the input.

Where it helps

Tool results, JSON, logs, diffs, and retrieved documents often contain more context than one model call needs.

What to measure

Compare answer coverage, achieved ratio, latency, and provider cache economics on your own corpus.