Moai Team is the studio I build in. We make agentic products — software where part of the work is carried out by language models — and we do not keep the construction to ourselves: the standard those products follow and the platform they run on are both public, under permissive licences. What is claimed here can be read in the code instead of taken on trust.
Six separate tools wired into one working system: the gateway every model call passes through, a memory that can show its sources, measurement of how the agents actually behave, adversarial security checks, fleet operations — and the standard all of it follows. One command brings the whole set up on your own machine; one page shows what it is doing. It is an assembly rather than a rewrite: the six are vendored as submodules and each stays usable on its own.
An engineering specification for building products on language models that can be trusted: boundaries of autonomy, verifiable quality gates, a human above money and irreversible actions. It carries five reference implementations — the components of the platform above. All five are still pre-1.0 and move with the standard.
A course platform built around Telegram: the author assembles a course, the student takes it inside the messenger. Built solo, end to end; now run with a partner.
An operating system for social-impact organizations: projects, partners, donors and reporting in one place, with a public catalogue of projects on top.
The engine this site runs on: writing, meetings, applications and a single queue for every action, built for one operator — me. The page you are reading is served by it.