Tanuki Context cuts log prompts by 94% by turning text into images

0syna · reddit · 2026-07-27

Tanuki Context is an open-source tool that tries to cut LLM token costs by converting some context into images.

The author says a log dump of 37,111 tokens can be compressed to 2,240 tokens, roughly a 94% reduction.
The idea is that pricing for images is based on pixel size, while text is charged per token.
The system includes a log distiller that collapses repeated lines but preserves errors, a columnar codec for JSON, and a cost model that decides when image encoding is not worth it.
It argues against itself when retrieval or prompt caching is cheaper.
The tool can run as an MCP server (npx -y tanuki-context) or as a proxy that rewrites requests in place.
The author notes limits: it needs a vision-capable model, it does not reduce output tokens, and retrieval can still win for narrow queries.

Code, benchmarks, and npm/GitHub links are provided.

More from coding & agent