Unified Publishing Engine with a Language-Agnostic Document Model

[robtaylor94]

I've been exploring BlockSuite as a foundation for a local-first, AI-native document editor and would be interested in contributing upstream.

The idea is to extend BlockSuite's existing block architecture with a language-agnostic publishing abstraction, separating semantic content from layout, rendering and export.

This could enable multiple representations of the same underlying document:

• Documents: Paginated text, tables, headers and footers.
• Publications & slides: Fixed layouts, reusable templates and presentation views.
• Scientific writing: LaTeX-compatible equations, citations and potentially Typst-based typesetting.
• PDF: High-fidelity export, with annotation/editing as a possible future extension.

The key would be to avoid coupling document semantics to any particular markup language, typesetting system or output format. Markdown, LaTeX, Typst and structured AI edits could instead be treated as different interfaces to a shared document model.

A Rust-based incremental layout engine might be worth exploring for performance, while retaining BlockSuite's existing CRDT and editing infrastructure.

I'd be interested in prototyping paginated layouts and PDF export first.

Would this abstraction fit BlockSuite's architecture, and would maintainers be interested in an experimental implementation?