Application Foundation · Any
The interface isn't code anymore. It's data.
A single JSON document draws your web, phone, TV, and car screens from one definition. Schema validation and a deterministic repair layer aim to keep a malformed document from breaking silently at runtime.
- Status
- Prototype
- Last verified
- 2026-07-18
- Access
- TODO_CONTENT_CONFIRMATION
The problem it solves
Hand-coding the same screen separately for each platform (web, mobile, TV, car) causes visual and behavioral drift between platforms.
Any reduces the screen to a single, platform-independent JSON definition; the adaptation decision is made by the component itself, not by platform-specific code.
How it works
Define the screen as a JSON document — by hand, in the Designer, or through an AI agent.
Schema validation and the deterministic repair layer check the document.
buildManifest() resolves bindings and adaptation rules.
The renderer draws the resolved tree on the target platform.
Verified capabilities
A document you can validate against a schema
The document is validated against a JSON Schema; an invalid field returns a field pointer and a suggestion instead of a silent runtime crash.
Adaptation by capability class
Instead of pixel breakpoints, components adapt themselves to screen size class, input modality (touch/pointer/remote/driving), and viewing distance.
One editing language
Edits made in the Designer and agent commands sent over MCP use the same JSON Patch language — humans and agents change the document through the same path.
Rendering from a compiled manifest
Renderers never read the document directly; buildManifest() resolves bindings up front, and drift between platforms is meant to be caught by snapshot tests.
Architecture and integration
Any (also referred to as "BaseUI" in some parts of the ecosystem — the renaming decision is still in progress) validates a single canonical JSON DTO (component tree + theme + adaptivity + bindings + actions) against a JSON Schema and distributes it to multiple renderers (Web React, Flutter).
AI agents produce and edit the document through an MCP server; the same contract is used across the designer, playground, and host-demo dogfooding environments.
Trust and control model
Because human edits and agent edits both go through the same JSON Patch language, there is no separate, unaudited second editing path by design.
Exact figures for the component catalog, golden test documents, and profile count are intentionally left out of this page; the internal test set keeps growing and a comprehensive independent verification hasn't been completed yet.
Known limits / areas for improvement
Naming not yet settled
The product is referred to as both "Any" and "BaseUI" within the ecosystem; the naming decision hasn't been finalized by the product owner yet.
Access model not yet settled
The access model hasn't been confirmed by the product owner yet.
No comprehensive independent test report
Component, profile, and golden-document counts can change over time internally; an up-to-date, independently verified inventory hasn't been published yet.
Evidence
Snapshot tests and golden test documents are part of the internal development process; a publicly available, up-to-date test report or numeric inventory doesn't exist yet. This section will be updated once one is published.