Potluck AI

We think the people who use AI should have a real stake in it.

Right now, the most powerful technology of our lifetimes is being built by a small group of companies. Much of the time, the rest of us meet it through a login page and a monthly bill. We want running AI on your own hardware to be a practical choice, too.

That is not the only way this can go.

Potluck AI is building a member-operated AI network, with a company that answers for the software and governed, over time, by the people who power it. Or, well, a potluck: everyone brings what they can, everyone eats, and nobody leaves hungry because they could not afford to host alone.

Your AI runs on your device. We ship an app that runs models with downloadable weights locally. Your chat history is stored on your device. We do not use it to train models. No ads. The conversations you have with your AI stay on your computer unless you explicitly choose otherwise.

When you need something bigger, the community has your back. A college student on a Chromebook should not be locked out of powerful AI because they cannot afford $200 a month. We are building a way for people with idle gaming PCs, workstations, and Apple Silicon Macs to contribute spare compute when they choose. Over time, that shared capacity helps deliver stronger models and better access to the people who need it.

Here is the honest version of how the network protects you. The planned open pool uses blinded routing: the job sent to a contributor omits account identity. The serving machine still processes the prompt itself, and the content could identify you. Software rules about keeping requests in memory cannot prove a modified machine never records them. Cryptographic private inference is not shipped. The public contributor network is still in development; local inference and your own paired machines are the available starting point.

Contributors are recognized for the work they do. The person whose 4090 sat idle all night helping other members deserves something for it. The proposed credit ledger would recognize useful compute and offset usage. Public earning and redemption are not launched. Credits are intended to be separate from governance. A governance token is planned for a later phase, subject to legal review, and we are not promising it to anyone. What stays constant: compute should not buy control of the network, and the richest member does not get the loudest voice.

The people who power it should help govern it. Our aim is for the network to be governed over time by the people who keep it alive, across a bounded set of decisions: which models the network runs, routing and privacy standards, and how the shared treasury is spent. That voice is scoped so compute cannot buy control of the network. The rules should be public. The books should be transparent. Wherever possible, the technology should stay open enough that if we ever stop being worth it, you can leave with your setup intact.

How this is different

There are token networks, GPU marketplaces, open-source communities, and venture-backed AI labs. Some are interesting. Some are useful. None of them are the thing we are trying to build.

We are building a founder-led company whose whole purpose is to stand up a network with no single owner: member-operated compute, governed over time by the people who power it. The company builds and answers for the software. No single owner is the ambition for that network, not a claim about today's company-operated infrastructure.

Concretely, that means four things under one roof, accountable to the same members: AI that runs on your own device, a memory layer that travels with you across the tools you already use, a contributor network that puts community compute behind larger models, and a governance structure that decides where any of it goes next. Plenty of organizations do one of those. None of them do all four, and none of them hand the network to the people who run it.

Why this, why now

Three things frighten people about AI right now, and all three are real.

Jobs get automated away and the value flows up. Potluck cannot stop automation, but a network owned by no single company can make sure that when AI becomes part of daily work, some of the value belongs to the people who actually power and use it.

Surveillance gets normalized. Every prompt you type into a hosted chatbot can become a data point. Local-first AI is the clearest answer to that, and the models are finally good enough to make it practical.

AI gets pointed at things we would never sign off on like ad targeting, military targeting, and political targeting, with no real say from the people it affects. A network with tens of thousands of members is a voice. A million is a constituency. That is how ordinary people stop being spectators.

We do not think we are going to beat frontier models on a benchmark. We are not trying to. We are trying to build a dignified, durable piece of the AI landscape that the people who use it help power and govern.

What we are not

We are not a pump. A governance token is a possibility for the network, planned for a later phase, but you never need it to use Potluck on your own machines, and the credit you earn for work cannot be traded or speculated on. We are not promising you will get rich. We are not claiming we will train a frontier model next year on volunteer laptops; the networking physics do not work yet, and we would rather be honest than hype.

We are not a charity, either. Members pay dues, contributors earn rewards, and the whole thing should sustain itself on real value or it does not deserve to exist.

We are not anti-AI. We love this technology. We just think more than five companies should get to decide what it becomes.

What we believe

1. AI should work for the people who use it, not the other way around.
2. Privacy is a feature, not a premium tier.
3. Access should not be gated by a $200 monthly subscription.
4. The network should belong to the people who power it, not to any single owner.
5. Contributions deserve rewards, but rewards should not buy power.
6. Honesty about what we can and cannot do is worth more than hype.

What we need

We need the first hundred people.

People who will pay a founding-member fee because they want this to exist, not because the product is finished. People with idle GPUs willing to help test the earliest version of a shared network. Lawyers who care about getting a member-governed network right. Designers who would rather ship a beautiful app for a small cause than a mediocre one for a big company. Moderators, writers, skeptics, critics.