Configure a job
Pick a workload, set requirements and a price cap — the estimate updates live from the network's real capacity as you edit.
Workload
Start from a template with a real public image, then make it yours.
Writes /output/result.txt = hex of h(iterations), h0 = sha256 of every input (sorted by mount name).
Shown on your dashboard and receipts.
Container
A public image and its argv. The sandbox has no host access, no privileged mode and no network unless you allow it below.
Pin a tag (or a sha256 digest). Unpinned :latest images may be held for policy review.
One argument per line, run as the container's full argv: the first line is the executable and replaces the image's ENTRYPOINT (and CMD); no shell, so use sh / -c / … for pipes or variables. Leave empty to run the image's own ENTRYPOINT/CMD.
GPU requirements
V1 runs every GPU of a job on one physical machine. Counts are live, schedulable GPUs.
Runtime & price
The job is stopped at the max runtime. You pay billable GPU-seconds at the matched host's rate, never above your cap.
No compatible capacity priced right now — your cap decides what the job can match later.
Inputs
Uploaded straight to private storage with a signed URL scoped to you. Hosts get short-lived read links for this job's inputs only.
Network & outputs
Default: no outbound network. Workloads can never reach the host's LAN.
The host uploads these paths as one tar bundle; the server re-hashes it.
Payment
The escrow on Robinhood Chain holds the max charge; unused funds are refunded.
Coarse regions only — a ranking hint for the scheduler, not a hard requirement.