Your agent reads the instruction
The root AGENTS.md tells it what this world is, who owns it, what it may do and what always needs your approval.
Describe a company context or a personal space. Vercy will package its identity, authority, registries and operating rules into portable files and give your AI agent a safe instruction for growing the world with meta-models.
You download thirteen small files. From then on your AI agent grows the world for you: it reads the instruction in AGENTS.md, notices what your world cannot yet describe, pulls the matching data structures from the Vercy catalogue and fills them as the work demands. You approve each step. What accumulates is your own universe of data.
The root AGENTS.md tells it what this world is, who owns it, what it may do and what always needs your approval.
You mention an entity, a decision or a document that no installed model can hold, or an installed specification points at one you do not have yet. Both are demand signals.
It finds the matching meta-model in the catalogue, proposes it with its version, and installs it only once you agree. The version is pinned by digest.
It interviews you, records where every fact came from, marks what is still unknown and turns that into the next question. Nothing is invented.