Tensormesh

San Francisco-based KV-caching SaaS inference platform built on the open-source LMCache project. Claims up to 10x reduction in LLM latency and GPU cost. Founded 2024 by University of Chicago, UC Berkeley, and CMU researchers.