The magic should belong to everyone.

Using a model from OpenAI or another frontier lab can feel extraordinary. We admire the research that makes those experiences possible.

It opens up a larger question: how do we make that possibility available to more people, in more languages, on more kinds of machines? We want a future where someone can be inspired by a breakthrough and then build something of their own.

If you depend on a model for your work, you should have a say in how it runs. You should be able to choose a model, understand its limits, keep a version that works for you, and move to another provider.

We support open weights because they make more of those choices possible. Official, uncensored, and specialized models all have a place. People should have room to set their own runtime policies and take responsibility for how they use their tools.

We oppose regulatory capture: rules shaped to protect established companies by making it harder for everyone else to participate.

Safety deserves careful research, public scrutiny, and evidence about actual harms. Researchers, independent developers, and the communities affected by these systems should have a voice in that work. Access to useful intelligence should not depend on the continuing permission of a small group of companies.

Open models become useful when people can actually run them. That takes practical information: which model fits, which engine works, how fast it runs, and what it costs. It also takes places for independent hosts to offer useful compute and get paid.

We are building OW Terminal to connect those pieces. Our aim is a network where people can choose between their own hardware and community compute, and hosts can compete on price, performance, and reliability.

An impressive number is only useful if you can understand where it came from. We publish model and hardware pairs with their configurations, source links, and measurement status. A reported result and a reproduced result should be easy to tell apart.

The same standard applies to the product: available machines, completed jobs, fees, and the parts still being built should be described plainly. Trust grows when people can check what you say.

People deserve to know who can see their work. Running a model entirely on your own device can keep a request there. Sending it through our pool means OW Terminal and the selected host handle it.

Shielded ZEC protects payment details on chain; it does not hide the prompt from those operators. We want stronger privacy, and we will describe the protection we actually provide as we build it. A private payment and private inference are separate promises.

The point is what someone gets to do with these tools: learn a subject, make art, investigate a problem, build a business, or help their community. We want more people to be able to experiment, contribute, and make a living from useful work.

Our part is to make open models easier to understand, run, and share access to. The best outcome is a world with more capable people and more places for new ideas to come from.