--- title: CloudGPU Agent Compute Lab emoji: 🧭 colorFrom: green colorTo: gray sdk: static pinned: false license: mit short_description: Read-only agent compute discovery and billing explorer --- # Agent Compute Lab A static, free-to-host demo: billing illustration, live anonymous readiness link, a minute-source estimate handoff, verified internal smoke-test evidence, and Skill/MCP installation. No API keys, customer data, paid GPU allocation or model execution occurs here. Source: https://github.com/cloudgpu-app/cloudgpu-agent-skills Copy `docs/index.html` to this Space's root as `index.html`, and `docs/assets` as `assets`. The self-check opens the actual CloudGPU endpoint directly, so browser cross-origin restrictions cannot be confused with a platform outage. ## Publishing Install `huggingface_hub`, configure `HF_TOKEN` in a secure environment and `HF_SPACE_ID=your-account/agent-compute-lab`, then run `python publish.py`. The token is read only from the environment and is never printed. `python publish.py --dry-run` validates the static bundle without credentials or network. Anonymous quotes are planning estimates only. On the external Space, validated inputs open the CloudGPU quote form; a cross-origin POST is not attempted. The user explicitly requests the estimate there. Whole-hour offers still require a separate binding quote and owner acceptance before allocation.