Skip to content
1kunaPublic

About

Hardware-aware inference control plane

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Mycelium

Turn the computers you already own into one big computer.

Your machines (a desktop, a mini PC, a laptop and a server) join one Mycelium. Each job you send goes to a free machine with the right hardware.

Mycelium links your computers together. You hand it a job, and it runs that job on whichever machine is free and has the right hardware. You never have to pick a machine yourself.

Like the fungus it's named after, it's one network with many branches, and each of your computers is a branch.

What it's for

  • Run any command or script on whichever machine is free and has the hardware it needs.
  • Put idle computers to work instead of leaving their power unused.
  • Share files between machines through your network drive, without copying them by hand.
  • Keep the computers you use responsive. Mark a machine as in daily use, and it only takes work while you're away.
  • Grow easily. Your agent adds each new machine with one command.
  • Use what you have: Nvidia, Intel or Apple GPUs, or just a CPU, on Linux and Apple Silicon Macs.

For example: ML training, simulations, rendering, data processing or builds.

Coming next: Windows, and splitting one huge job across several GPUs.

Set it up with your AI agent

Open Claude Code, Codex or another coding agent on the computer that should be the hub (one that stays on), and paste this:

Set up Mycelium on my computers. Clone https://github.com/1kuna/mycelium,
then read docs/agent-setup.md and follow it step by step. Ask me which
computer is the hub, which others to add, and where my shared network drive
is mounted on each. Show me every install plan before running it, and never
print or commit the secret token.

The agent installs the hub, adds your other machines, checks that each one shows up, and runs a test job. Before you start, you need:

  • Linux or Apple Silicon Mac computers on the same local network
  • A shared network drive (NAS) mounted on each one
  • Working GPU drivers, if a machine has a GPU

Try it without any setup

On one laptop with Go installed, from a clone of this repo, start a pretend cluster with fake GPUs. The jobs really run.

go build -o myce ./cmd/myce && ./myce dev
# leave this running; in a second terminal:
MYCE_QUEUE=http://127.0.0.1:7331 MYCE_TOKEN=dev-token ./myce run -- echo hello

If you see "queued waiting for a hypha", the job is waiting for a worker machine.

More

Guides · Agent setup · Laptop walkthrough · Running jobs · Operations · Design

About

Hardware-aware inference control plane

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages