I own GPUs
Set up a private, branded, multi-tenant GPU cloud: choose a deployment
model, enroll your fleet, create tenants, and run it day to day.
I rent GPUs
Get your account, rent whole nodes from your provider’s pool, and watch
every GPU you pay for with live telemetry and logs.
What the platform actually does today
Fleet observability
GPU utilization, VRAM, thermals, power, ECC and Xid faults from every node
at 15-second resolution — plus per-node log search with fault highlighting.
Hard multi-tenancy
Customer tenants isolated server-side, with two automated leak tests running
in CI on every change.
Self-service rental
Customers rent whole nodes by GPU model from your live availability and
release them when they are done.
Your brand, your metal
Your name and logo on your console, deployed managed, into your own cloud
account, or on your own hardware.
Three consoles, one platform
Neither of the first two can reach past its own line. That boundary is enforced
on the server from the session, never from anything the browser sends — see
Isolation.
Everything in these docs is labeled for what it is. Features marked LIVE are
running in production today. Features marked SOON or ROADMAP appear in
the product as visible, disabled previews — never a silent gap. See
How we label things.
Where to go next
How it works
Operators, tenants, nodes, and agents — the four nouns everything else is built from.
Deployment models
Managed, sovereign, or on-prem — and what changes between them.
Data residency
Component by component, where every byte lives.