planned
Dedicated GPU
Isolation: vm
Distributed NVIDIA compute
Powerful GPUs.
One control plane.
Rent complete GPUs or hardware-isolated MIG slices across a curated network of Linux hosts. Built for predictable isolation, transparent hourly pricing, and workloads that can move.
01Centralized control
Catalog, scheduling, identity, and billing in one place.
02Distributed execution
GPU hosts connect outbound from any supported location.
03Explicit isolation
VMs for full GPUs, restricted containers for MIG slices.
Product path
Start focused. Expand cleanly.
The same workload contract will later support managed inference and isolated agent runs without coupling them to the core scheduler.
planned
Fixed MIG slice
Isolation: restricted container
future
Managed inference
Isolation: managed
future
Agent runtime
Isolation: sandboxed
MVP contract
Ephemeral by design.
Workload disks exist only while a rental is running. Stop or delete the workload and its local filesystem is permanently destroyed. Customers export results before shutdown.
control_plane ready
catalog 0 active offers
storage ephemeral-only
gpu_runtime validation
version 0.1.0