Provide compute

Hosting overview

Lend your GPU to the network and earn credits per request you serve. There are three ways to host, trading off zero-friction against power.

Pick your client

Browser host

Open /host, click start. WebGPU runs the model in the tab — nothing to install.

Chrome extension

Host in the background without keeping a tab open. Install once, host whenever idle.

Console host

npx @piedpiperrh/host — serve via local Ollama from any terminal, no browser.

How they compare

Field	Type	Description
Browser host	easiest	No install. Quantized small/mid models via WebGPU. Hosts only while the tab is open.
Extension	background	MV3 Chrome extension. Same model tier as the browser, but keeps hosting without a visible tab.
Console host	most power	Node CLI serving through local Ollama. Native speed, headless, runs 24/7 on any machine.

What every host does

However you host, the lifecycle is the same:

Field	Type	Description
Go live		Detect your GPU, load a model, and open a session that heartbeats so dispatch knows you're available.
Serve jobs		Receive matched requests over Realtime, run inference locally, and stream results back.
Earn		Each served request credits your balance at your listed price (priority jobs pay a multiple).
List		Optionally publish a title, description, and price to the marketplace so borrowers can target you.

Your hardware never leaves your machine

The model runs locally. Pied Piper routes the request and settles credits — it doesn't see your GPU or your weights. You set the price and can stop any time.

Pricing your compute

New listings default to roughly half of OpenRouter's reference pricefor the equivalent model, so you're competitive without doing math. You can override it. Borrowers who want guaranteed turnaround pay your priority multiplier on top. Full mechanics in Pricing.