Share your GPU, get any AI model
Leave this tab open and its GPU serves an open model to other people. You earn 80% of what they pay, as credit you can spend on GPT, Claude, Gemini and 200+ other models. No install, no email, no wallet.
Who is serving right now
| Model | Tabs online | Price per 1M tokens (in / out) |
|---|---|---|
| P2P: Qwen3.5 2B | - | $0.02 / $0.06 |
| P2P: Qwen3.5 4B | - | $0.03 / $0.10 |
| P2P: Qwen3.5 9B | - | $0.05 / $0.18 |
Anyone can call these models from the chat or the API by their ids, p2p/qwen3.5-9b and the others above.
How it works
- Pick a model your GPU can hold and press Start sharing. The weights download once (1 to 6 GB) and stay in your browser's cache.
- When someone asks a
p2p/…model a question, Obscurify sends it to an idle tab. Your tab writes the answer with WebGPU and streams it back through Obscurify. - They pay the listed price. 80% of it is credited to your account, and it reaches your balance after 24 hours.
- Spend the balance on any model, in the chat or through the API. It can't be withdrawn as money.
What each side can see
- The people you serve never learn who you are. Obscurify relays every word, so they don't get your address either.
- You never learn who they are. Your tab does process the text of their questions, which is why P2P models are labelled as such: nobody's private chat is sent to a stranger's browser unless they picked a P2P model.
- Obscurify records which account served a job, the model, a token count and the credit, so it can pay you. Nothing about who asked.
Honesty checks
Now and then an idle tab gets a check question with a known answer. A tab that answers it wrong, for example by sending made-up text instead of running the model, loses everything it has earned that hasn't cleared yet and is kept out for a day. That's why earnings wait 24 hours.
What you need
- A browser with WebGPU: Chrome or Edge on Windows, macOS or ChromeOS, and recent Safari. On Linux, Chrome needs WebGPU switched on.
- A GPU with the memory shown next to each model. Integrated graphics can serve the 2B model.
- An account number, so the credit has somewhere to go. It takes no email.
Why this exists
Private AI shouldn't depend on one company's data centre. Open models can already run on the GPUs people own, and research groups like Covenant AI and Pluralis have trained large models across volunteers' machines. Serving them across those machines is the next step, and it needs a place where your compute and your identity stay separate. P2P sharing is that place, starting with models that fit in one browser tab. The state of peer-to-peer inference page tracks the rest.