dimavedenyapin/openvast

openvast is a terminal dashboard that helps you rent a GPU from vast.ai, a marketplace for renting graphics cards by the hour, and use it to run large AI language models through llama.cpp. Instead of manually finding a suitable GPU, setting it up, and configuring your coding tools to use it, openvast automates the whole process from a text-based interface you run in your terminal. You pick a model you want to run, and openvast searches vast.ai for the cheapest GPU offer that has enough video memory to fit it, then launches an instance on that machine. It shows you live progress as the instance starts up, from provisioning through downloading the model weights to becoming healthy and ready to use, along with real-time statistics like generation speed and how much the instance is costing you per hour. Once a rented GPU is healthy, openvast automatically connects it as a usable model inside opencode, a coding assistant tool, and removes that connection again if the instance is paused or deleted. The dashboard lets you pause, resume, or delete instances directly, and comes pre-loaded with a curated list of about nineteen AI models across different memory size tiers, covering families like Qwen, Gemma, Mistral, and Llama, though you can add your own models by editing a YAML configuration file without changing any code. Installation is a single command that sets up Python, the vast.ai command-line tool, and openvast itself, plus generates an SSH key if you do not already have one. You then need to create a vast.ai account and set your API key before openvast can rent GPUs on your behalf. Without a vast.ai key or opencode installed, the tool still runs in a read-only monitoring mode.