AI that answers to its owner.

What we believe

The most personal software in your family's life — the tutor that hears your child think out loud, stumble, and try again — should not be a rented service running in someone else's data center. It should be something you own, running on hardware you hold.

Modern phones are extraordinary computers. An iPhone can run a genuinely large language model — billions of parameters — with no connection to anything. We build products that take that seriously, instead of using the phone as a thin window onto a server.

How we build

• Local-first, always. The intelligence lives on the device. If a feature can't work without the cloud, we'd rather not ship it.
• Private by architecture. Not private by promise, policy, or toggle — private because the data has nowhere to go.
• Honest by design. Software that grades a child's answer has zero room for flattery or bluffing. We engineer for correctness first, and for admitting uncertainty rather than hiding it.
• Native, not wrapped. We build with Apple's own frameworks, top to bottom — no third-party analytics, no embedded trackers, no borrowed stacks.

What we're building

Our first product is OUMA, the offline AI tutor: spoken lessons, conversational quizzes, honest grading, and parent-readable progress — all from an 8-billion-parameter model running entirely on an iPhone. It is currently in active development, with a private TestFlight beta for families next.

Questions, ideas, or just want to follow along? Write to waitlist [at] ouma [dot] app.