The company keeps the lights on. The Foundation is where the parts that should belong to everyone actually do: the data, the benchmarks, the models, and the community. Nobody has to buy anything to use them.
A shared record of how good reasoning is done
Every practice produces something valuable: calibrated reasoning with a full trail of what was known, assumed, and checked. Pooled together with consent, that becomes a commons: a consent-clean corpus of careful thinking that does not belong to any one company. It is meant to belong to the people who build it, and it is the raw material for everything below.
Does the discipline actually help? We measure it.
It is easy to claim AI is more grounded. It is harder to prove it. So we benchmark the stack in the open: take a task where a model can confidently bluff, and check whether the discipline makes it push back instead. We compare against the plain model, and as the work matures, against the other tools people use. The point is that the answer is public, and reproducible, not a line on a sales page.
Grounded AI that belongs to everyone
The commons is training data for a specific kind of model: one that pushes back when it is not sure, instead of confabulating confidence. We train open-weights models on it and keep them open, for the work where being right matters more than sounding right. This is research-stage. It improves grounding, it does not magically eliminate it, and we will not pretend otherwise. But it is the direction worth pulling toward.
Keeping it alive and learnable
A framework is only as useful as it is understandable. The Foundation keeps the open documentation current and runs the community. The Discord is live, and workshops for people who want to learn how to build AI that is honest about what it knows are in planning. The free framework, the docs, and the help to use it are all part of the same commitment.
Want to be part of it?
Contribute reasoning to the commons, help with the benchmarks, or just come learn. The Foundation is open by design.