Community-owned · Open-weight · Non-profit by design
Flywheel is a community-owned platform of niche AI employees, one model per trade. The weights are open and yours to keep. Every dollar of profit goes straight back into training the next version. And because they’re hosted offshore and run on open weights, no government and no company can quietly switch them off — no matter how good they get.
Why we exist
In 2015, the most famous lab in the world launched as a non-profit with a promise: build AI in the open, for everyone. The moment it got valuable, the weights went closed and the mission went for-profit. Flywheel is the opposite, on purpose — owned by the people who use it, funded in the open, and impossible to quietly take back.
No VCs to answer to, no exit to engineer. The people who use the models own and steer them. We run it like a non-profit: the platform serves its members, not shareholders.
100% of profit is reinvested into the next training run. No dividends pulled out, no margin skimmed. The community funds the models; the models get sharper; everyone benefits.
Open weights, hosted offshore. Download a model and it’s yours forever — no court, no government, no company can reach in and turn it off. The better it gets, the more that matters.
Where the money goes
Each model is funded by the community and retrained on real, consented usage. The more a niche invests, the more retrain cycles it gets — so the leaders here are simply the models the community has bet on hardest. Every figure below is reinvested profit and member funding, not a number we pulled out.
Approximate community-reported training totals, updated as new runs complete. Generation (e.g. v4.2) counts community-funded retrain cycles and is separate from the Apache-2.0 open-weights release tag.
Own it together
This isn’t a product you rent from a company. It’s infrastructure we build together. Members fund the next training runs and help steer which niches we train next — and in return, you own a piece of the commons you helped build. The more the community puts in, the better every model gets, for all of us.
Community membership and governance — a way to fund and steer the commons. It is not an offer of securities, shares, or an investment contract, and nothing here is a promise of financial return.
Member funding pays for the compute and data behind the next generation of every model.
Members vote on which niches we train next and how profit gets reinvested.
Every release stays Apache-2.0 and public — your access can never be revoked or rented away.
Built to outlast us
As these models get genuinely capable, the pressure to control them goes up. We designed that pressure out from day one.
Hosting and governance sit outside any single country's reach, so no regulator can order the models dark.
Every model is public on Hugging Face. Download it once and you can run it forever, fully offline.
There's no central license server to revoke. A lapsed plan, a takedown, a shutdown — none of it bricks a model you already hold.
How you run it
Point any OpenAI SDK at our hosted API or run the open weights with llama.cpp or vLLM. Change one line — the base URL — and ship.
Apache-2.0, public on Hugging Face. Run them on your box or ours. Nothing leaves your network unless you turn it on.
Opt in and your consented usage trains the next version — the flywheel. Flip ghost mode any time and the pipe goes dark.
10 niche models · Qwen3.6-35B-A3B base · Apache-2.0 open weights · OpenAI-compatible
Pick a model and ship today, or join the community building the alternative to closed, for-profit AI.