# GPU for Agents

> Your agent rents the GPU it needs, on its own. You pay only while it works: we watch every machine and shut down the idle ones. Open source (MIT), by Jambu.ai.

GPU for Agents (`gpu4agents`, the `gpu` command) lets a coding agent rent a cloud GPU (Vast.ai, RunPod and Lambda built in, any other provider through an adapter), serve an open-weights model with vLLM behind an OpenAI-compatible API, run experiments, and monitor cost. A watchdog on the machine shuts it down after `idle_timeout` (15 min by default).

## Install (for agents)

- [install.md](https://gpu4agents.jambu.ai/install.md): step-by-step setup an agent can follow: install, connect the account, learn the rules
- [install.sh](https://gpu4agents.jambu.ai/install.sh): `curl -fsSL https://gpu4agents.jambu.ai/install.sh | sh -s -- --agent claude|codex|gemini|cursor|auto`
- [jobs.md](https://gpu4agents.jambu.ai/jobs.md): the contract for running your own code on the GPU as a job: training, image, audio or video models, anything that needs CUDA; outputs come back
- [serving.md](https://gpu4agents.jambu.ai/serving.md): serve any Hugging Face model behind OpenAI routes (images, voice, transcription, embeddings, any pipeline), or a server the agent writes
- [providers.md](https://gpu4agents.jambu.ai/providers.md): add a GPU provider gpu does not ship, by writing an adapter against the provider contract

## Use

- [Documentation](https://gpu4agents.jambu.ai/docs/): architecture, setup flow, lifecycle states, the watchdog and its shutdown rule, data flow, event log, and every contract (CLI JSON, exit codes, MCP tools, watchdog API, provider adapters)
- [The skill (SKILL.md)](https://raw.githubusercontent.com/jambuai/gpu4agents/main/skills/gpu4agents/SKILL.md): rules, commands, monitoring, exit codes
- [Agents guide](https://github.com/jambuai/gpu4agents/blob/main/docs/AGENTS.md)
- [Provider contract](https://github.com/jambuai/gpu4agents/blob/main/docs/ADDING_A_PROVIDER.md)
- [Integration contract for apps](https://github.com/jambuai/gpu4agents/blob/main/docs/INTEGRATION.md)
- [Source](https://github.com/jambuai/gpu4agents)