Open source · MIT License
GPU inference,
one command away
Launch on-demand GPU workers on Vast.ai for llama.cpp inference. Any model, any tool, pennies per session.
$
$
Features
What it does
One command setup
Run vasted setup, answer a few questions, done. No YAML, no Docker, no SSH.
Auto GPU selection
Finds the cheapest GPU that can actually run your model. No scrolling through Vast dashboards.
OpenAI-compatible
Works with OpenCode, Claude Code, Cursor — anything that talks to a /v1 endpoint.
Telegram bot
Manage your workers from your phone. Start, stop, check status — all from Telegram.
Non-interactive mode
Feed it to your agents or CI pipeline. No prompts, just flags and exit codes.
Auto-idle shutdown
Walk away and it kills the server after 30 min idle. No surprise bills at 3am.
Get started
Install in seconds
$ uv tool install vasted
Copied!
Then run vasted setup to configure