Private grids for frontier AI
Reuse your existing hardware to run LLMs privately and securely.
wget trellis.unfoldml.com/cliYour compute, your data.
Your prompts and results never travel to a third-party API.
Trellis pools your existing hardware into a private compute grid. Each node is allocated a fraction of the model weights, so a Trellis grid can serve a large model no single machine could fit.
Adapts to your resources.
Laptops and GPU workstations in the same grid.
Nodes can run different operating systems (Linux and Mac OS for now) and span CPU-only and GPU hardware (CUDA and Metal supported).
Encrypted end-to-end.
Every link in a Trellis grid is secured with mutual TLS, thanks to SPIFFE/SPIRE.
Security is the top priority for Trellis: all communication in a Trellis grid is encrypted end-to-end. The Managed plan shares some physical infrastructure with other customers, but the data plane of each grid is always isolated. The Self-Hosted plan is fully air-gapped.