ReCache Reuses Tool Schema KV Cache, Cutting VRAM by 92%

A new arXiv paper proposes ReCache, which builds independently reusable KV cache states per tool schema for LLM agents, saving 92% of VRAM and speeding up first-token generation by 3.7x.

2026-09-08 ~ 2026-09-08 · 2 related posts

• ReCache reuses tool-schema KV states, cutting agent memory 92% with little accuracy loss — techNmak · 2026-09-08
• ReCache cuts agentic KV memory 92.43% and speeds first token 3.655x with minimal accuracy loss — techNmak · 2026-09-08