We want 100,000 c0mpute nodes. We have fewer than 10.

Profullstack has launched c0mpute, an open compute network that lets any machine join with a single command to serve AI inference, FFmpeg transcoding and storage jobs without an account or port forwarding. The network currently has fewer than 10 live nodes, at least three of which are Profullstack's own servers, against a stated goal of 100,000 nodes to pool spare capacity for training and serving open-weight models. The team also fixed an installer bug found during a clean Ubuntu 24.04 test, where an apt-get redirect to /dev/tty caused headless installs to exit with status 1 before starting the worker.

c0mpute is the open compute network we have been building at Profullstack. Any machine can join it with one command, and once it is in, it serves AI inference, FFmpeg transcoding and storage jobs for whoever needs them. No account and no approval step, and you do not need to forward a port to join. The number we care about is 100,000 nodes. Pooled spare capacity at that size is enough to train and serve open-weight models in the open, on hardware owned by the people running it. Today the best models come from a few labs that own their own data centers. A volunteer network will not beat them next year, and I am not going to pretend it will. But it does not have to win on day one to be worth building. It has to get big enough to matter. Right now the live count is under 10 7 this morning , and at least three of those are our own servers. That is the honest starting point, and the count is public at c0mpute.com/nodes https://c0mpute.com/nodes , so nobody has to take my word for it. curl -fsSL https://c0mpute.com/install.sh | sh That installs c0mpute , plus infernet for inference and coinpay for payments, and starts the worker for you: a systemd service where it can, a background daemon where it can't. Then: c0mpute worker status c0mpute bench one full run puts a score next to your node on the board It runs on Linux x86 64 and arm64 and macOS, and on Windows inside WSL2. A GPU is detected and advertised if you have one, but it is not required. CPU nodes serve small models like qwen2.5:0.5b. Paid inference goes through Infernet https://infernetprotocol.com , the inference plugin that ships with c0mpute. Clients pay through CoinPay in 16 coins or with PayPal, and you choose where your share lands with infernet payout set <coin <address . Paid demand is early, so expect small and irregular earnings for now. I would rather say that than promise a number. You also get a spot on the board. And if you bring people, give them https://c0mpute.com/r/<your-handle : every node installed from that link shows your handle next to it. It is a scoreboard, not a payout. Before writing this I ran the installer on a clean Ubuntu 24.04 container as a normal user, the way a stranger would. It took about two minutes, and it failed. The transcode step redirects apt-get to /dev/tty so sudo can prompt you, and with no terminal ssh without -t, cloud-init, a container that redirect fails before apt even runs. Because the installer runs under set -e and that call was the last command in an if body, the whole script exited with status 1 before it started the worker. So on exactly the kind of headless box that makes a good node, you got binaries and no node. Now the package manager only gets the terminal when there is one, and a missing ffmpeg only turns off transcode jobs instead of the install. The installer also serves as text/x-shellscript now. The code is MIT and on GitHub https://github.com/profullstack/c0mpute . If the installer breaks on your machine, open an issue with the output and I will fix it.