gpu.ai open source

what's inside

The parts of GPU.ai that launch machines

• Provisioning engine

  Places each launch on a supplier with real stock, tracks it from creating to running, and cleans up when it ends.

• Supplier framework

  One adapter interface for every GPU source. Ships with a generic adapter for any supplier that speaks the open Supplier API.

• gpu CLI

  Launch, list, SSH into, and terminate instances from the terminal.

• Python and TypeScript SDKs

  Typed clients generated from the same OpenAPI spec the server serves.

• MCP server

  Lets AI agents in Claude and other MCP clients find and launch GPUs on your deployment.

quickstart

Running in a few minutes

You need git, Docker, and Go. The mock supplier gives you machines to launch before you connect real hardware.

git clone https://github.com/gpuai-dev/gpuai-oss.git
cd gpuai-oss
docker compose -f deploy/docker-compose.yml --profile mock up -d --build

KEY=$(docker compose -f deploy/docker-compose.yml exec -T gpuctl gpuctl apikey create --name admin)
make build
echo "$KEY" | ./bin/gpu login --stdin

./bin/gpu pricing -o table
./bin/gpu instances create -t rtx_4090 -o table
./bin/gpu instances list -o table

To connect real suppliers, see the supplier guide.