Hive Compute

Hive Compute lets you earn by sharing models that run on a machine you control. It also puts available community-hosted models in the normal HivemindOS model picker.

Start hosting

Before you begin, load at least one supported model in a local model app on the machine.

1. Open More → Hive Compute, or choose Rent compute on a Fleet machine.
2. Select Set up hosting.
3. Let HivemindOS find and measure your models.
4. Choose which models to share and review the prices and limits.
5. Select Go live.

That’s all. HivemindOS connects the machine and keeps its access securely on that machine. You do not need to enter a service address, access token, or command. Hosting does not expose your files, shell, wallets, or other local services.

To host from another Fleet machine, open Hive Compute on that machine. Rent compute on a Fleet card can start a machine that is already set up for hosting. Measurement and pricing always happen on the host itself.

Prices and earnings

Setup measures each model on your machine before showing a price. Until then it shows Not priced yet and can’t go live.

• Automatic pricing starts competitive, based on similar hosted capacity and your machine’s measured speed.
• Custom pricing lets you set input, output, and optional minimum job prices per model.

Projected earnings are shown after the marketplace fee. The Earnings view separates actual earnings from projections, with totals for today, 7 days, 30 days, all time, and each model. Idle only and Scheduled availability count only the hours your machine can take work; Always keeps it available whenever your other limits allow.

Your controls

• Choose exactly which models to share.
• Limit how many jobs run at once.
• Host only when idle or on a schedule.
• Pause on battery, or yield when someone uses the machine.
• Set a daily earnings cap.
• Stop hosting at any time. Stopping from the dashboard keeps it from resuming on its own. If hosting crashes while live, HivemindOS restarts it after a short delay.

Speed

Each marketplace model shows its expected speed and wait before the first word, measured from real jobs on the fastest available host. New hosts show Measuring speed.

A host that is too slow for a model stays listed, but that model isn’t sent to it until it improves; a smaller model often qualifies. The hosting view explains why. Among fast-enough hosts, speed breaks price ties, and the Fast route puts speed first. Each reply shows which anonymous host answered and how fast.

Privacy levels

• Standard: community hardware. Requests are protected while traveling over the network, but the host may be able to see their contents.
• Confidential Verified: protected hardware with a fresh verification. Private work stays encrypted for the intended recipient and is never silently moved to a Standard host.

Protected hardware isn’t required to host; it earns a stronger trust label and higher placement.

What you see in the model picker

Only models with live capacity appear. If no community host is online, there may be no marketplace models to show; the built-in HivemindOS models are always available separately. Protected capacity is listed first. A model disappears when its last host goes offline.

Troubleshooting

• No models found: start your local model app, load a model, and refresh.
• Connection unavailable: check your internet connection and select Set up hosting again. HivemindOS repairs the machine connection automatically.
• Benchmark did not finish: keep the model app running and retry. Large models can take several minutes.
• No marketplace models appear in the picker: no eligible community host may be online. Try again later, or bring your own host live.
• A Confidential Verified model disappeared: its last protected host went offline or its verification expired. HivemindOS won’t downgrade private work.

Advanced self-hosted marketplace setup is a separate operator feature.