Manifesto
The decisions matter more than the chat.
Every AI system is full of small decisions. Is this message safe? Which team should handle it? Can this agent issue that refund? Is this answer supported by the source? Each one deserves an answer that is fast, measured and made close to the data. That is how you build something people can trust.
A decision deserves its own model.
A model built only to decide can be small, fast and excellent at it. Myles reads the situation, weighs the options you give it and picks one, in a single pass.
Every claim is measured.
We test every model on frozen sets it has never seen and publish what we find. A confidence score should mean something, so we calibrate it and certify the error rate before a decision is kept, and route the hardest cases to a person or a larger model.
Intelligence should run where the data lives.
A decision about a patient, a claim or a citizen belongs inside the organisation that makes it. Myles runs on the laptop, phone or server you already own. Small models also cost less and use far less energy, which means you can check every step, every time.
You should know where your model came from.
Every new Myles model ships with a record of what it was built from: the base model, the data, who made them and under what licence. Our standard for every new Myles model is no Chinese-origin models or data anywhere in the lineage, and no outputs of hosted models as training data. Sovereignty is something you can check.
Build it here.
We train in the UK, on UK public text where it matters, for the public sector and the companies that serve it. A country that depends on AI for its public services should be able to own the decisions those services make.
Small models, measured honestly, owned by the people who use them. That is Myles.