- Ecosystem
The LMCache Ecosystem
Products and projects that build on LMCache or implement its interfaces.
LMCache is open source KV cache infrastructure for LLM inference. Teams use it directly, run it inside commercial products, and build software that works with its interfaces. This page lists both kinds of work. If your product or project belongs here, refer to “Get listed” below.
- Products
Built on LMCache
Commercial products and services that include LMCache. Choose one of these if you want to run LMCache in production with enterprise support.
Tensormesh Platform
A data management layer for AI inference, from the team that created LMCache. It runs a multi-tier KV cache across GPU memory, host memory, and external storage, and deploys on Kubernetes with your existing accelerators, storage, and inference engines.
- Open source
LMCache-Compatible Projects
Open source projects that implement the LMCache interfaces. A compatible project can replace or extend LMCache without changes to the inference engine or the storage layer. LMCache is the reference implementation and is the first project on this list.
LMCache
- Reference implementation
Open source KV cache layer for LLM inference. It stores, shares, and reuses KV cache across GPU, CPU, disk, and remote storage.
Your project here. Learn how to get listed.
- Listing
Get listed
Listing is open to any product or project that meets the criteria below. You check your own work against the criteria and tell us in a submission. We review the submission and add your card. The criteria will become more formal as the LMCache interface specification matures.
Built on LMCache
Your product or service qualifies if:
- It ships LMCache as a component, or runs LMCache as part of a managed service.
- Your public documentation names LMCache as a component.
- You use the LMCache name and logo correctly, and link back to lmcache.ai.
LMCache-Compatible
Your project qualifies if:
- It is open source under an OSI-approved license, AND
- It implements one or more LMCache interfaces with the same behavior as LMCache:
- Inference engine connector: how an engine such as vLLM or SGLang stores and retrieves KV cache.
- Storage plugin interface: how a local or remote storage backend plugs into the cache.
- Cache control: how a controller finds, moves, and removes cached KV data.
- It passes the LMCache test suite for each interface it claims. Conformance tests will be published with the interface specification.
Why this matters: A shared interface lets users change cache layers, storage backends, or vendors without rewriting their inference stack. It also gives every compatible project a fair, public path into the ecosystem.
