Vast.ai CLI and SDK

Vast.ai's own Agent Skills teach your agent its GPU cloud's command line and Python SDK, so it can search the GPU marketplace, launch and manage instances, move files, and run serverless endpoints for you. What its skills teach your agent - vastai (CLI): search offers with filters and sorting (GPU model, count, VRAM, reliability, price, region), create instances from an image, poll until they're running, connect over SSH, copy files, read logs, and stop or destroy instances; plus interruptible (spot) rentals, volumes, serverless endpoints and deployments, templates, SSH keys, API keys, billing, teams, environment variables and machine hosting - vastai-sdk (Python): the VastAI class for the same jobs in code, the low-level sync and async clients, and the Serverless client for sending requests to an inference endpoint How it connects - Install the vastai CLI (curl -fsSL https://vast.ai/install.sh | bash on Linux or macOS, or pip install vastai), then sign in once with vastai set api-key and a key from your Vast.ai console. - Add the skills with npx skills add vast-ai/vast-cli. Vast.ai documents Claude Code, Cursor, Windsurf and Codex. - The skill has your agent use --raw for JSON output it can read reliably. Good to know - Renting GPUs costs money. Storage is charged from creation and GPU time once an instance is running; the skill reminds your agent to destroy instances when done, and to time out polling loops so a stuck instance doesn't keep charging. - Destroying an instance is permanent. Stopping keeps the disk without GPU charges. - The CLI runs on Linux and macOS; on Windows use pip.