About givemeanode
givemeanode is on-demand GPU compute whose primary user is an agent, not a person at a console. It is a product of San Francisco Compute, the supercomputing company.
The premise is that an agent doing machine-learning work needs the same things an engineer needs - a machine, a queue, a disk that survives a stop, a bill it can read - and needs them through an interface it can call rather than a dashboard it has to click. So every capability the product has is a tool call first. The web pages are for the human watching.
What you can run
A node is a container on a GPU machine with a persistent, encrypted disk, billed by the minute. If you stop a node it parks. The files stay, the billing stops, and the next command wakes it. Node types are named for the chip you ask for - h100-1 is one H100, h100-8 is a full eight-GPU machine - and CPU-only nodes (cpu-2, cpu-8) exist for downloads, ingest, and preprocessing.
Beside nodes there are batch jobs and sweeps for work that should queue rather than hold a machine, rollout sandboxes that fork in milliseconds for reinforcement learning, object storage, private networks between your own nodes, and metrics, logs, and traces that arrive labeled by the mission they came from.
How you drive it
Three doors onto one set of capabilities, so nothing is reachable from only one of them:
- An MCP server at
https://mcp.givemeanode.com- streamable HTTP with OAuth 2.1, which MCP clients discover and a human approves once. - An HTTP API for scripts and CLIs, with a public OpenAPI document at
/openapi.json. - The gman CLI, installed with
curl -fsSL https://givemeanode.com/cli.sh | bash.
Sign-up is self-serve and the first token is too: gman login mints one from a browser or a device code, and an agent authorizing the MCP server gets one through the consent screen. Nobody has to talk to a salesperson to start. The docs are the full reference, and /llms.txt is the same contract written for an agent to read directly.
What we commit to
We do not use customer content to train, fine-tune, evaluate, or improve machine-learning models - not ours and not anyone else's. That is the only mode the service has, it applies to every account, and it needs no opt-out. The full text is the Data Use Policy, which is part of the Terms of Service; the Acceptable Use Policy says what the machines may not be used for.
Current service state and incident history are public at status.givemeanode.com, what shipped and when is on the changelog, and contact lists every way to reach a human.