Co-Founder
Profile summary: Adobe Product Marketing Manager
Tensormesh
Tensormesh is an AI inference platform that stores and reuses KV-cache entries outside GPU memory, so repeated prompt prefixes do not need to be recomputed. Built on the open-source LMCache engine, it aims to lower time-to-first-token, latency, and GPU spend for production LLM workloads. Teams can deploy it on public GPU clouds or on-premises Kubernetes clusters and manage it through REST APIs, SDKs, and a cost observability console.
- Artificial Intelligence
- Developer Tools
- Software Only
Updated 26 days ago
Funding
Raised to date
$24.5MRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
across 2 rounds
C
AV
MFN+1
AVMFN+1$20M · Seed Extension
Announced May 2026
Disclosed cumulative: $24.5M
2/2
RoundDateAmountValuationInvestors
Founders
Co-Founder
Since Jan 2025 · 1 yr 10 mos
Profile summary: Google Research Intern
Chief Scientist & Co-Founder
Since Apr 2025 · 1 yr 7 mos
Profile summary: Microsoft Research Intern
Software Engineer
Since Apr 2025 · 1 yr 7 mos
Profile summary: Tensormesh • Software Engineer · BS Mathematics and CS (2021-2025) at University of Chicago
Chief Product Officer
Since Oct 2025 · 1 yr 1 mo
Profile summary: HP Product Manager
CTO & Co-Founder
Since Apr 2025 · 1 yr 7 mos
Profile summary: Microsoft Research Intern
Product
Problem
Repeated LLM inference requests recompute token-level activations for identical prompt prefixes, consuming GPU cycles and increasing service latency and operating costs. These inefficiencies limit the throughput of production inference workloads, and agentic workflows that resend system prompts and conversation history compound the cost.
Solution
Tensormesh provides an AI-native caching layer built on the LMCache engine that stores KV-cache entries outside GPU memory and reuses them for requests with matching prefixes. Cached context is served directly rather than reprocessed, which reduces repeated computation and is designed to lower time-to-first-token, latency, and GPU utilization. The service can be deployed on public GPU clouds or on-premises Kubernetes clusters. Operators connect it to inference workflows through REST APIs, command-line tools, and SDKs, and track cache hit rates and cost metrics in an observability console. Enterprise security and audit options are also available.
Target Audience
Primary customers are AI platform teams, enterprise SaaS providers, and large-scale LLM application developers running inference workloads on cloud or private GPU clusters.
Features
- Distributed KV-cache storage outside GPU memory, with on-demand rehydration for reuse across requests
- Cache-aware routing that directs prompts toward nodes holding relevant cache entries
- REST API, Python and Go SDKs, and CLI tools for integration into inference pipelines
- Compatibility with inference servers such as vLLM and plugin hooks for custom model stacks
- Metrics dashboard for cache-hit rates, latency, GPU-hour savings, throughput, and per-model cost analytics
- Kubernetes deployment options, including Helm charts, operators, and on-premises isolation
- Enterprise controls including SSO/RBAC, encrypted data transit, and optional SOC 2-ready audit logging