AI power to the people.
The next AI data centre is all of us. SilentSwarm joins computers that already exist into one swarm that runs open language models too big for any of them alone. Bring a machine or just a question: either way, you are part of it.
Running a lab, a company or a city? SilentSwarm for labs and companies
Who holds the compute decides who gets to use AI
The largest language models run where the largest GPUs are: in data centres few companies can afford to build. Everyone else rents access, at prices and on terms someone else sets.
- Compute sits with a few. A card that can hold a large model alone is out of reach for most people, clubs and schools. So they ask whoever owns one, and accept the answer.
- Capable GPUs are already out there. In homes, offices and classrooms, GPUs stand idle for hours every day. Each is too small for a large model on its own, so those hours go unused.
- The swarm joins them. SilentSwarm splits a model across those machines, so together they run what none of them could alone, and the power stays with the people who own them.
An evolution, not another data centre
No new hall and no new cooling tower. The swarm grows one existing computer at a time, on the cards and connections people already have.
- Older cards count. Our benchmark model, Qwen3-14B, runs across four older GPUs of 4 and 6 GB. None of them could hold it alone.
- Every machine gets a share sized to it. Mismatched cards are welcome: each holds as much of the model as its memory allows.
- Home internet is enough. Machines pass compressed intermediate results, not whole models, and they connect out, so no port has to be opened on your router.
- A pool can be yours. The coordinator is one small always-on machine without a GPU. A club, a school or a group of friends can run its own.
What two home PCs already do
Qwen3-14B in 4-bit, split across two ordinary PCs on one home internet connection, with four older GPUs between them.
A rented data-centre GPU will be faster, and these are small samples on old cards. The benchmark page has the method, the raw results and the limits, including a minute of waiting before the first token of a 1,500-token prompt.
Three ways to take part
You do not need to own a GPU, and you do not need to ask anyone for a data centre.
Bring a machine
Linux, an NVIDIA GPU and one command, without administrator rights. It connects out, works while you let it and stops when you stop it. Every GPU-hour it contributes is on your profile.
Just use it
No GPU and no install. Chat with the pool's models in your browser, or call them through an OpenAI-compatible API. Free while the pool is in beta.
Start a pool of your own
For a club, a school or a group of friends: your coordinator, your machines, and prompts that stay on them. Free for noncommercial use.
Where the swarm stands today
Small so far, and this is where it starts. Read from the coordinator that serves this page, not typed into it. Machine numbers are withheld while the pool is small, so they never describe one person's computer.
Does contributing pay?
Not yet, and we would rather say so than imply otherwise. Today the trade is in kind: your machine helps the pool, and the pool answers you. This is the order we are building the rest in.
- Today: see what you contributed. GPU-hours, tokens served and requests answered are recorded against the member who added a machine, and shown on your profile.
- Next: credits. Contributing will earn credits and using the pool will spend them. They will follow the GPU time a machine was measurably busy for others, never the hours it was merely switched on.
- Later: a published rate. What an hour of a given card will earn, and what a thousand tokens will cost.
- Last: payouts. Real money through a licensed payment provider, once the merchant of record, VAT, reporting and how the licence treats paid hosting are settled.
What exists, and what does not yet
SilentSwarm is an open beta. "For everyone" is the direction, and this list says how far it reaches today.
Available today
- Using the pool without a GPU, in the browser and through an API, free during the beta.
- Joining with one command, without administrator rights and without opening a port.
- Your contribution on record: GPU-hours, tokens served and requests answered, on your profile.
- Pools of your own, free for noncommercial use.
Not yet
- Instant sign-up. During the beta an administrator reviews and approves each new account.
- Windows and other GPUs. A machine needs Linux and an NVIDIA GPU today; Windows is untested.
- Getting paid for a machine. Credits first, payouts after that.
- A sandbox. The worker is an ordinary process with your permissions, so run it on a machine where that is acceptable.
- An open market among strangers. It needs sandboxing, fault tolerance and verification, and we will not claim it before they exist.
- Fast long prompts on old cards. On the cards above, a long prompt waits about a minute before the answer starts.
Questions people ask
- Do I need a GPU to take part?
- No. You can chat with the pool's models in your browser, or call them through an API, without a GPU and without installing anything. The pool is free to use during the beta, and an administrator approves new accounts first.
- What does a machine need to join?
- Linux and an NVIDIA GPU. Joining takes one command in a terminal, installs into one directory in your home and needs no administrator rights. Windows is untested so far.
- Who sees my prompts?
- The machines in your pool that hold the model you picked. A pool is the boundary: its models are served only on its own machines. In a shared pool those machines belong to other members, so send nothing there that they should not see. In a pool of your own, prompts stay on your machines.
- Can I get my machine back?
- Any time. Stop the process and it stops taking work; purge it and the directory is gone. Weekly availability windows say when a machine is open to others.
- Will I earn money with my GPU?
- Not yet. Today you see what your machine contributed. Credits for the GPU time it was busy for others come next, then a published rate, and payouts in real money last.
- How fast is it?
- On two home PCs with four older GPUs, 3.3 tokens per second for one user and 11.6 in total for eight at once. A rented data-centre GPU will be faster. The swarm's offer is a model none of your machines could hold, on hardware that already exists.
- Is it free, and who may use it?
- SilentSwarm is source-available under the PolyForm Noncommercial licence. Any noncommercial purpose is permitted, such as personal study, hobby projects and private entertainment, and so is use by charitable organisations, educational institutions, public research organisations and government institutions. Commercial use needs a licence from NexPatch AI.
Starting a pool with your group?
A club, a school, a makerspace or a few friends: tell us which machines you have and what you want to run. Groups that start early get our help setting up their first pool. We answer every message ourselves.
We use your details only to answer you - see the privacy policy. SilentSwarm is source-available under the PolyForm Noncommercial licence; commercial use needs a licence from NexPatch AI, which you can ask about in the same message.