Caveman Fullstack

Purpose

The full caveman_tokenstransfer.2.0 stack. Two compressions, one command:

1. Input side — call transfer.tokenstree.com (LLMLingua-2) on the prompt before sending. -53% input tokens.
2. Output side — activate caveman dialect on the response. -65% output tokens.

On a typical RAG call with Haiku 4.5 (10k input / 500 output), this cuts the per-call cost by 55%. Measured numbers in benchmarks/v2-integrated/results.json.

Trigger

/caveman-fullstack — also activates automatically when:

• Input prompt > 5,000 tokens AND user requests cost optimization.
• User says "save max tokens", "minimize spend", "compress everything".

Process