SF Compute
SF Compute (San Francisco Compute Company) runs a spot-priced market for very large scale GPU clusters. The api.sfcompute.com REST API lets buyers and sellers place market orders for blocks of H100 GPU-hours, manage tradable cluster contracts, query live market prices, check balances, and provision managed Kubernetes clusters, nodes, and VMs - all driven by the `sf` CLI and language SDKs.
APIs
SF Compute Orders / Market API
Place, list, fetch, and cancel buy/sell market orders for blocks of GPU compute. Buy orders specify instance type, quantity, price, duration, and start; sell orders list unused ...
SF Compute Contracts API
List and retrieve tradable cluster contracts - the multi-node H100 InfiniBand reservations produced when buy orders fill. Contracts carry an instance type, an interval/quantity ...
SF Compute Prices / Quote API
Request a live market quote for a desired instance type, quantity, duration, and start window. Returns an indicative per-GPU-hour and total price derived from current orderbook ...
SF Compute Clusters / Nodes API
Manage managed-Kubernetes clusters, nodes, VM instances, procurements (auto-scaling fleets), and the kubeconfig credentials used to reach them. Covers listing instances and zone...
SF Compute Balance API
Read account balance (available and current cents), credit balance, transactions, and account identity. Used to confirm spend headroom before placing market orders and to reconc...
SF Compute Inference API
OpenAI-compatible inference served on SF Compute capacity. Models are listed via the models endpoint and called through a standard chat completions interface, letting bought GPU...