GPU compute,
on demand.
Train models, run fine-tuning jobs, and execute GPU workloads on a distributed network of partner hardware — without managing infrastructure.
From script to GPU in three commands.
A workflow shaped for engineers — no dashboards, no consoles, just the terminal.
Point the CLI at any Python script. The wizard detects the partner's GPU backend and selects the right Docker image automatically.
Your balance is verified, a signed job token is issued, and your script is dispatched directly to an approved partner node over an encrypted connection.
Your script runs in a secure, isolated Docker container on the partner's GPU. Stdout and stderr stream back to your terminal in real time.
Built like infrastructure, not a product.
A small set of primitives, executed sharply. No bloat, no surprises.
Jobs route straight from your machine to the partner's. No middleman proxying your data.
Secure containers for every job
Billing is calculated on actual runtime seconds, not reserved time. Idle time costs nothing.
No web UI required to submit jobs. Install Python, run three commands, get results.
The scheduler matches your job to nodes with the right backend: NVIDIA CUDA, AMD ROCm, or Apple Metal.
Jobs require a pre-authorized token. You're never charged more than your balance allows.
Turn idle GPUs into income.
Register your machine as a partner node. Our setup wizard handles Docker hardening and job dispatch automatically. Earn a share of every job that runs on your hardware.
Payouts settle weekly.
- Minimum100 Mbps up/down
- Baseline1 Gbps symmetric fiber
- Premium2.5–10 Gbps symmetric
- RequiredEthernet (no Wi-Fi)