A GPU for Claude Code
Claude Code works in a shell, and so does Lium. With the Lium agent skill and the lium CLI, Claude Code can find a GPU, rent it, run your training or inference job on it and remove it when the job is done.
Set it up once, then ask Claude Code to use a GPU
npx skills add Datura-ai/lium-skill --skill lium # teaches the agent the commands curl -fsSL https://lium.io/install.sh | bash # the lium CLI export LIUM_API_KEY=sk_... # then, in Claude Code: "rent the cheapest RTX 4090 for one hour and run train.py on it"
What the agent runs
lium ls --format json lists nodes and prices, lium up <node-id> --yes --no-ssh --ttl 1h rents one and returns when it is ready, lium exec runs a command without an interactive terminal, and lium rm removes the pod.
Every command answers in JSON where it can and exits with a code the agent can branch on. docs.lium.io/developers/agent-scripting lists the contract flag by flag.
Keep the agent on a budget
Give the agent its own API key instead of yours. A key can be limited to reading and renting, can be made to see only the pods it creates, and can carry a daily, monthly or lifetime spend limit. lium keys create sets all of these.
Renting with --ttl removes the pod when the time is up, even if the agent forgets.
Docs inside the conversation
docs.lium.io/mcp is an MCP endpoint for searching and reading the Lium docs, and lium.io/llms-full.txt is the whole CLI and SDK reference in one file.
Frequently asked questions
Does this work with Codex or other agents?
Yes. Anything that can run shell commands can drive the lium CLI. The skill installs the same way for other agents that support skills.
Can an AI agent do this without a browser?
Yes. Signup, API keys, renting, running commands and removal all work from the CLI or the REST API with no browser and no e-mail confirmation. docs.lium.io/developers/agents is the one-page guide.
How is a pod billed?
Every pod has an hourly price and is billed per second, with no minimum rental period. Billing stops when the pod is removed. A pod rented with --ttl removes itself when the time is up.
For AI agents — CLI install, agent skill, headless signup: llms-full.txt