Provide compute

Console hostNewest

Host from any terminal — a VS Code panel, an SSH session, a headless server. The @piedpiperrh/host CLI serves inference through your local Ollama and earns credits per request, no browser required.

When you need it

Browser and extension hosts need a WebGPU browser session. The console host doesn't: it runs wherever Node ≥ 18 and Ollama run — Linux boxes, Macs, Windows machines, GPU servers in a rack. If your machine can run ollama run llama3.2, it can earn credits.

Start hosting

shell

1# 1. Install Ollama (https://ollama.com) and pull a supported model
2ollama pull llama3.2:3b
3
4# 2. Start hosting with an API key from Settings -> API keys
5npx @piedpiperrh/host --key lg_pk_your_key_here

The CLI detects which of your installed Ollama models exist in the Pied Piper catalog, registers them, and starts serving jobs. Leave it running; stop with Ctrl-C and it deregisters cleanly. Miss heartbeats for ~5 minutes and the session is ended server-side.

Options

flags

1piedpiper-host --key lg_pk_...      # or set PIEDPIPER_API_KEY
2  --price 2                         # credits you charge per request
3  --title "RTX 4090 box"            # marketplace listing (makes you pinnable)
4  --vram 24000                      # advertised VRAM cap in MB
5  --ollama http://127.0.0.1:11434   # non-default Ollama endpoint

How you get paid

Identical to every other tier: each completed request transfers your --price in credits (× the priority multiplier for priority requests) from the borrower to you. Earnings show on the dashboard; withdraw from Credits & wallet.

Supported models

The network identifies models by their catalog id; the CLI serves the equivalent Ollama weights. Currently mapped:

Field	Type	Description
Llama	3.2 1B/3B · 3.1 8B	`llama3.2:3b` is the most-requested model on the network — the best starter.
Qwen 2.5	0.5B–7B + Coder	Including `qwen2.5-coder:7b` for code workloads.
Gemma 2	2B / 9B	`gemma2:2b` fits modest GPUs.
Others	7B–8B class	Phi 3.5, Mistral 7B, DeepSeek-R1 distills, Hermes 3, SmolLM2, TinyLlama.

Ollama models outside the catalog are ignored

The CLI only serves models it can identify on both sides. If nothing you have installed matches, it exits with pull suggestions.

Trade-offs

Power: native inference speed, bigger models than a browser tab, runs headless 24/7.
Setup: needs Ollama installed and a model pulled — one step more than clicking start in a tab.
Best for: anyone with a always-on machine or a spare GPU server who wants passive credit income.