loginsignup

An ASCII drawing of a dual-fan graphics card that slowly turns. Its fans spin faster when the network serves more tokens.

Rent GPU time by the hour or per token.
Earn with your GPU.

A peer-to-peer market for local LLM inference. Hosts run one command. Buyers call an OpenAI-compatible API.

Rent GPU timeEarn with your GPU
0
GPUs online
0
tok/s now
0
tokens served

Live network

online booked · updates every second
0 online · 0 booked · 0 tok/s
01 empty slot · npx sublet-cli start
02 empty slot
03 empty slot
04 empty slot
0 tokens served

How it works

Host · earn with your GPU

  1. 01 Connect your computer Run login on the GPU machine and approve the code on this site.

    host
    npx sublet-cli login --server https://sublet.duckdns.org
open https://sublet.duckdns.org/link?code=BCDF-GHJK

  2. 02 Take a request The market shows what fits your GPU. Take one, then start it.

    host
    npx sublet-cli start r_7Kq2…
● loading Llama-3.1-8B · Q4_K_M

  3. 03 Earn Booked hours pay out when the timer ends. Per-token calls pay as they stream.

    host
    npx sublet-cli jobs
◆ booked · 06:14:32 left

Buyer · rent GPU time

  1. 01 Get an API key New accounts get free test credits.

    buyer
    open /app → create API key
sk-•••••••• (shown once)

  2. 02 Pay per token Call any online host with the OpenAI API format.

    buyer
    curl https://sublet.duckdns.org/h/<host>/v1/chat/completions

  3. 03 Or book by the hour Paste a HuggingFace .gguf link. A host loads it for you alone.

    buyer
    curl https://sublet.duckdns.org/r/<booking>/v1/chat/completions
$0 per call while booked