Runcrate gives AI teams and their agents GPUs and inference from one account.Instances: rent a GPU by the hour (H100, H200, A100, L40S, RTX 4090, RTX PRO 6000 and more, 20+ models) with SSH, Jupyter and VS Code in the browser. A launch reaches running in about four minutes, booting is never billed, and most GPUs bill by the second.Inference: 200+ open models (DeepSeek, Qwen, Llama, Kimi, FLUX, Whisper) behind one OpenAI-compatible API, so the client you already use works. Priced per token, image or second.Storage volumes keep datasets and checkpoints between instances, and reserved clusters cover 8+ GPUs or month-long terms.Built for agents: connect Claude Code, Codex or Cursor to the Runcrate MCP server with one command and sign in through your browser, no key to paste. Your agent can then launch a GPU, run commands on it and shut it down. Also a CLI (brew install aeonmindai/tap/runcrate) and Python and TypeScript SDKs.Who it is for: developers and small AI teams who want a GPU without a sales call, and AI agents that need compute on demand. Typical uses are fine-tuning an open model with LoRA, serving a model with vLLM on your own endpoint, running ComfyUI or FLUX on an RTX 4090, batch transcription with Whisper, and RAG over your own documents.How billing works: there is no subscription and no monthly fee. You top up one prepaid balance and pay only for what runs, with GPU time metered per second on most GPUs and inference billed per token, image or second. Docs, the CLI and the SDKs are public at runcrate.ai.