Potluck
Local AI. A little more together.
Your laptop. The gaming PC upstairs.
That little server with a second life.
Run open models on the hardware you own. Potluck connects your machines, so you choose where the work happens.
Free to run locally. Sign-in required.
Mac & Linux 0.1.7 · Windows 0.1.5
A little world of your own
Pick a machine above. See what happens.
Write an invitation to a neighborhood potluck.
02 / YOUR CHOICE
Your desktop
03 / BACK TO YOU
Pick a machine. Send a little work.
Laptop → your desktop → laptop
A tiny illustrated demo, with example answers. In Potluck, paired machines must be reachable and have a model ready.
Start with one machine.Bring the others along.Keep the choice yours.
The Potluck film · 30 seconds
From the computers at home to friends across the country, a glimpse of the network we’re building. More people bringing more ideas to life, together.
Concept screens show the vision. Trusted circles, the open pool, and model splitting are still in development.
It starts with an idea. Connect the computers around you. At home. At work. With friends. Across the country. Together, power larger AI models. Bring your ideas to life. Potluck. Bring what you’ve got. Build something bigger.
The hardware is already here
One app for the machines you have. Room to grow when you need it.
Draft, think, ask, and code with a model running locally. Choose a model that fits your hardware, download it, and get to work. Once it is installed, local inference does not need an internet connection.
Your model. Your CPU or GPU.Stay on your laptop while a paired computer runs a model it has loaded. Potluck brings the WireGuard connection into the app, including the setup and readiness checks.
A private mesh. No separate VPN account.Give compatible coding agents a local memory for project facts, decisions, and unfinished tasks through MCP. The context you save stays available when the session ends or you switch tools.
Useful on day one, even with one machine.Where this is going
Potluck is building toward a member-operated compute network. People bring the hardware. Every request has a boundary you choose.
Local inference and your own paired computers. The foundation you can use today.
An invited group sharing capacity, with explicit access rules. The next boundary to open.
A wider network of contributors, with opt-in routing and clear rules for the work they serve.
The company builds the software. The ambition is a network powered and governed by its members. We publish the current limits alongside the vision.
The details matter
Model speed depends on your hardware. A private connection is different from private inference. A roadmap is different from a release. Here is the detail behind the page.
A few fair questions
No. Start with local chat and a model that fits your machine. Connecting another computer is optional. You can add it later, without changing how you work.
In local mode, inference stays on that machine. When you select My Machines, your prompt goes to a paired computer over the encrypted mesh so that computer can answer it. You choose the routing scope.
The everyday path sends a request to a machine that can run the model. Splitting a single model across machines is experimental; it is not a promise of pooled GPU memory or datacenter performance.
Trusted-circle routing and the open contributor pool are still in development. The current desktop release is for local inference and your own paired machines. You can follow the next steps on the roadmap. Read the roadmap →
The desktop app is free to download and run locally. Your hardware, electricity, and connectivity are yours to provide. Shared-network pricing and contribution credits are not launched.
Potluck runs models with downloadable weights; their licenses vary by model. The Potluck application source is not publicly released yet. We will publish source and licensing details when that changes.
There is a seat for you
Start with something useful on your own computer.
Help us build the network from there.
Mac/Linux 0.1.7 · Windows 0.1.5 · free to download and run locally.