60% Fable cost cut by converting code to images and having the model OCR it
Summary
The repo README introduces pxpipe, a local proxy that reduces input token usage for AI models by rendering bulky context (system prompts, tool docs, and prior history) into images which the model OCRs. It claims substantial end-to-end savings on token-dense content, notes the approach is lossy, and provides benchmarks and usage guidance. It also outlines model compatibility, configuration options, and caveats around exact recall and potential occasional errors from imaged content.