There is no language but Lisp, and AI is its application.
A personal, persistent, self-evolving Lisp you drive in natural language — and, when you want precision, in Lisp directly.
$ oml oml 0.1 · image restored — 42 objects, 17 skills talk to me. (drop to Lisp with a leading form) you> summarize project X and save it as a skill "weekly-x" oml> done — saved skill weekly-x; next time just say "weekly X". you> (skills) => (:recall :note :weekly-x …)
A body that remembers
A bare model forgets. oml is a persistent image — its objects, state, and learned skills survive restarts. Yesterday's work is still here today.
It grows
Every request is fulfilled by running Lisp in the image. When a workflow works, it is kept as a named skill — so the longer you live with an image, the better it understands you.
Natural language first
You talk to it. Lisp is the precision layer underneath, there when you want it. This is the difference between a toolkit and a companion.
Safe to inhabit
The agent acts through a deny-by-default grant — a bounded vocabulary, not ambient power. Safe to hand a capable but untrusted mind the keys.
The loop
natural language
→ the agent composes and runs Lisp in the image
→ if it works, it is saved as a skill
→ next time that intent is a known capability,
and your language is understood more richly
A plain assistant cannot do this — it has no persistent body in which a vocabulary can accumulate. oml does.
Documentation
Read the vision, architecture, and the accepted ACP client design.