We're building AI that self-improves and collaborates with people and teams to reach what neither could alone: built to 10x agency and ambition, not replace them.

Models are not products.

Today's AI doesn't learn from you. The model is frozen at ship; the harness is static; memory features store facts but the system doesn't get sharper because you used it yesterday. A truly intelligent product does the opposite: invisible underneath, evolves to fit the person using it, gets sharper on its own.

Two research bets.

• Continual learning: systems that keep getting sharper from real production use, without regressing on what they already got right.
• Pluralistic alignment: staying true to the specific person or organization, not a population average.

We own the stack.

We're not a wrapper around someone else's API. The whole stack, from research to product, is ours. That lets us take bets other labs won't, change direction with evidence, and put every interaction to work twice: serving the user now, training what comes next. Not one model for millions, but millions of distinct intelligences, each diverging into something only its user has.

Why this matters.

Intelligence at this scale only works as co-evolution. The system gets sharper the longer you use it. You get more capable the longer you have it. The relationship is the unit of progress.

A lot of capability and creativity is local. One person, one community, one tradition of craft. This is Hayek's knowledge problem: the knowledge that matters is dispersed across individuals, beyond any central authority's reach. The best decisions belong with the people who hold the local knowledge. So does the agency to act on them. Most never compounds into anything visible because the leverage isn't there. Ramanujan almost didn't reach the world. The next decade of AI is the difference between finding the next one and never knowing they existed.

Who's building it.

Mission-driven small team at the intersection of AI research and design. Battle-tested in building products for billions of users and top tech companies.

Two products. One engine.