Built in the UK
Trained and evaluated on UK-owned hardware, on UK public text under the Open Government and Open Parliament Licences.
Solutions · Public sector
Decisions about citizens, made in house.
Public services are putting AI into decisions about people. Myles lets them do it on their own hardware, with models built in the UK, a lineage they can inspect and an error rate a risk owner can sign off.
The problem
Data cannot leave the estate
Casework, benefits and health records cannot be sent to an overseas API, and many networks are disconnected by design.
Black boxes do not pass assurance
A model with an unknown lineage and an unmeasured error rate is hard to take through a security review or an algorithmic transparency record.
Per-call pricing does not fit budgets
Usage-based AI costs grow with demand, which is the opposite of what a fixed annual budget can absorb.
Where Myles sits
Why Myles
Runs on your estate
On servers, laptops and devices you operate, including disconnected and air-gapped networks. Nothing is sent to us.
An error rate you can sign off
A certified error rate on held-out data, with everything below the threshold handed to a person.
The evidence
91.8%correct on 1,400 held-out decisions, ahead of the medium tier of a leading hosted LLM family
2.7%error on the 73% of decisions kept at a 5% target, on unseen test data
£4per million decisions on one CPU thread, against £1,880 for that hosted tier
Myles-400M, our own measurement, September 2026. Method and all results