Workstation PC with a GPU serving inference at a desk
Hosting

Host inference on your GPU

Inference-only catalog jobs. You supply hardware and uptime. Scalattice routes work and pays on tokens served.

Providers on Scalattice do not run arbitrary customer containers. You host the shared model catalog with the open-source scalattice-agent. That keeps jobs predictable and payouts comparable across the network.

What you need

  • A machine with a supported GPU (or CPU for tiny experiments; real earnings want VRAM).
  • Disk space for GGUF weights. Multi-model setups add up fast; watch the Storage meter in Cloud.
  • A stable connection. Brief blips are fine; long offline stretches mean no jobs.
  • A provider account and machine token from Scalattice Cloud.

Install path

Linux: one install script from Cloud, with your machine token. Windows: setup wizard from the same dashboard. After install, the agent should show connected on the machine tile.

Full steps live in provider docs and on the providers page. Do not invent alternate install URLs.

Models and disk

Enable only models that fit. Weights download from Scalattice mirrors; incomplete downloads keep you out of ready state. Disable models you are not hosting. Remove leftover weights you no longer need. Disk is yours, but clutter hides real capacity.

See choosing a model for which IDs make sense on mid-range cards vs flagship GPUs.

Schedules and “in the pool”

You control when the machine accepts work: always, paused, or hour windows. “In the pool” means connected, schedule open, compute enabled, and at least one catalog model ready. Earnings only happen while that is true and jobs arrive.

Money

There is no connection fee. You earn a majority share of developer token spend on completed inference for models you served. Rates and share show in Cloud. Connect Stripe for payouts when you are ready.

Keep the agent current

When Cloud shows an update on the Agent square, install it. Releases fix routing edge cases, download bugs, and Windows/Linux install quirks. Stale agents are a common reason machines look “connected” but never get useful work.

Checklist

  1. Register as a provider and add a GPU machine.
  2. Install the agent with that machine’s token.
  3. Enable GPUs and one or more catalog models that fit.
  4. Wait until downloads finish and status shows ready / in pool.
  5. Set a schedule you can actually keep.

If something sticks, email support@scalattice.com with the machine name and what the tile shows.