Mosvera
A declarative runtime for aesthetic intent in AI-native systems
Aesthetic direction lives in scattered prompt strings; I wanted it to be a versioned, queryable, compilable artifact instead — built as an open standard for strangers, not for me.
Every generative system I run has the same rot at its core: the aesthetic — the palette, the line weights, the “vintage datasheet schematic” look that this very site is built on — lives as a brittle prompt string copy-pasted between scripts. Mosvera is the bet that aesthetic intent should be a first-class declarative artifact: palettes, templates, modifiers, and compositions as typed JSON documents you can version, query, compose, and compile down to a provider-specific payload. It is deliberately built public-first — no references to my own systems, examples a stranger would recognize, open-source conventions throughout. The name itself came out of a full brand-availability process (web search for commercial collisions, domains, GitHub org, npm scope, social handles), because the prior name “Tessera” died on exactly the methodology failure of checking DNS alone and nothing else.
The repo is a monorepo of the spec, the runtime, the providers, the schemas, and an MCP server, with the canonical home at mosvera.io and the other TLDs redirecting there. Schema $ids are rooted at the public domain, decisions are tracked as numbered ADRs, and the registry validates every entry against Zod schemas with cross-reference rules enforced at load time — a stack entry that names a template that doesn’t exist is a loud failure, not a silent no-op. The model has six first-class concepts now, including compositions: a brand-flexible, addressable-by-slug carrier for “this ordered stack of templates with these overrides, applied to whatever brand the call cares about.” That was the natural unit once I noticed the agent re-passing the same three-slug array on every banner call.
The hard part is discipline, not code. Mosvera operates under six doctrines that function as a constitution — public-first, adoption-over-convenience — and the test on every commit is whether a doctrine points one way even when the other path is more convenient. The active track is the Phase 6 public unlock: the repo split, npm publication, schema hosting, and launch essay all have to be verified live before the standard counts as shipped, and the doctrine explicitly forbids claiming any of that is done before it is. It is the one project in the lab built to outlive the lab.