Community-powered local AI p2p network

Let's run AI.
On our own machines.
Not Big Tech's cloud.

Let's build our own decentralized AI p2p network.

Get the app → How it works → The mission →

Share your idle GPU and offer tokens to the network; use the community's when you need more — prompts go straight, peer to peer. Hard rate limits are built in, so you can offer a text model and tap someone else's image model. Every share earns credits; using the network spends them. Offer some, get some.

See it in action

Share your AI

Already running Ollama or LM Studio? Open the app and your idle GPU joins the commons. You set what to share; close the app and you're instantly out.

Earn credits for every request you serve.

$ lapsus [provider] engine=lmstudio · sharing gemma-4-e2b [provider] online — ready to serve ✓ served gemma-4-e2b · +1177 CC

$ lps ask 1 "what is p2p?" asking the network… Peer-to-peer is a decentralized network where machines share resources directly, no central server. ― 29 in / 287 out · 1177 CC

Use it — chat

Pick a community model, send a prompt, and get the answer back from someone else's machine over a direct encrypted channel. Multi-turn chat, Markdown, attach a file — right in the app or the terminal. Spend the credits you earned by sharing.

Remember: offer some then get some.

Use it — as an API

Run lps gateway and you get a local, OpenAI-compatible endpoint. Point any OpenAI SDK or tool at it — just change the base_url — and it runs on the community network instead of Big Tech's cloud. The prompt still travels peer to peer.

LangChain, LlamaIndex, Aider, Open WebUI, your own scripts — anything that speaks OpenAI.

The whole network, behind one API.

$ lps gateway & # → http://localhost:4040/v1 $ curl .../v1/chat/completions -d @req.json { "object": "chat.completion", "model": "google/gemma-4-e2b", "choices": [{ "message": { "role": "assistant", "content": "Peer-to-peer is a decentralized network where machines talk directly, no central server." }, "finish_reason": "stop" }], "usage": { "prompt_tokens": 29, "completion_tokens": 287, "total_tokens": 316 } }

But why?

Because the compute should be ours.

A cooperative for compute. Everyone chips in resources; the ones doing the work get paid out — owned by the people who run it.

Let's build a decentralized, global, peer-to-peer data center — especially for local AI, run from the community, for the community.
Let's own the infrastructure we use, instead of renting it from Big Tech's cloud.
Reach beyond your own machine — use models others offer that your hardware can't run. Someone out there can.
Offer some, get some — earn credits when you share, and make local AI usable for people who couldn't run it otherwise.
Coming: rewards for the AI you offer — and a way to back the commons even without a GPU, so support flows to the people providing the compute.

Open source

Building something open?

Ready to join the commons?

Run a local model, flip the switch, and you're part of the network.

Get the app →