NeuronPool

Social pools

Pool your machines, run models free with friends

Pool your PCs with friends. Run open-weight models free. Earn when idle.

Create a pool Install the agent
client.py
  1. # Open-weight models on the pool (Llama, Qwen, Gemma, …).
  2. # OpenAI SDK is one client; Anthropic /v1/messages uses the same key.
  3. from openai import OpenAI
  4. client = OpenAI(
  5. base_url="https://neuronpool.damnknee.workers.dev/v1",
  6. api_key="sk-neuronpool-…",
  7. )
  8. client.chat.completions.create(
  9. model="neuronpool-tiny-chat",
  10. messages=[{"role":"user","content":"hello"}],
  11. )

Live network

The pool right now

—Hosts online
—Pools
—Hosts total
—VRAM (GB)
—GPUs pooled
—Models available

How it works

Three steps

  1. 01

    Create a pool

    Sign up and make a pool from the dashboard. Pool traffic is never billed.

    Create a pool

  2. 02

    Invite friends

    Send an invite link. Friends join and add their machines with the agent.

  3. 03

    Run models free

    Point any OpenAI- or Anthropic-compatible app at the pool and use open models like Llama, Qwen, and Gemma.

Models

Open-weight models, not OpenAI GPT

Machines run open models — Llama, Qwen, Gemma, gpt-oss, plus embeddings, speech, and images — with llama.cpp or MLX. Your app talks the OpenAI or Anthropic API; the models are open, and they run on the pool.

Invited?

Join a pool

Open the invite link a friend sent you. Once you're in, you can add a machine of your own from the dashboard.

Marketplace

Earn when idle

Turn on public jobs and your machine sells spare capacity on the marketplace while you're not using it. Runs on Linux, Windows, and Mac.

Install the agent

Split

100% to hosts at launch

Cash-out via Stripe Connect. No charge when a job fails. The platform fee is configurable and set to 0% at launch. Pool and self traffic is never billed.

100%Host
0%Platform fee

Trust

Trust tiers

Public routing defaults to software. Pool and self traffic still runs at none — friends already trust the machine.

TierWhat a buyer can rely on
noneHeartbeat liveness. Pool and self only.
softwareReports a released agent build and keeps proving its key. Public-eligible after probation.
edrSoftware plus an org EDR sensor. Not a TEE.
hardwareReserved. Asking for it returns 503 today.

Trust levels — guarantees, how to reach each tier, and what not to claim.

FAQ

Is the API OpenAI-compatible?

Yes — as a client shape. Point an OpenAI SDK at /v1, or Anthropic Messages at /v1/messages, with a buyer key. The weights are open models on pooled hosts, not OpenAI GPT.

Do you run OpenAI models?

No. Hosts serve open-weight models: Llama, Qwen, Gemma, gpt-oss, embeddings, speech, image, and org-private GGUF or MLX. If llama.cpp or MLX can load it, a host can advertise it. vLLM and Ollama are optional local backends and register as llama.cpp.

Is this attested?

Partly. Public hosts report a released agent build, answer a key challenge every 5 minutes, and sign each response. That is not a TEE: the person running the machine can still see prompts. See trust tiers.

Can I buy credits here?

Yes, from the dashboard after you sign in. Credit packs use Stripe Checkout in test mode until live keys are set. This landing page does not sell packs.

For developers

For developers

OpenAI-compatible /v1 (chat, responses, embeddings, images, audio) and Anthropic /v1/messages, authorized with an API key. Read the docs.

Works with coding agents like Claude Code, Cline, and Continue — setup guide.

base_url https://neuronpool.damnknee.workers.dev/v1

Idle machines are the cheapest inference there is.

Start a pool with friends, or sell spare capacity on the marketplace. One account is both.

Create a pool