An AI-native Markdown reader for Windows and Linux — renders Mermaid diagrams, KaTeX math, Vega-Lite charts, and GFM tables the way modern LLMs produce them.

Data charts (Vega-Lite)	Diagrams (Mermaid)

Math + tables	Release notes view

MarkView is a reader, not an editor. Open it when you want to:

• See LLM output rendered immediately — paste a Claude / ChatGPT / Gemini reply that contains Mermaid diagrams or Vega-Lite charts and see them as visuals, not raw JSON / code blocks. No copy-pasting into a separate site.
• Read LLM-generated technical reports — KaTeX math, GFM tables, syntax-highlighted code, and diagrams render in one consistent view.
• Open a .md file and just read it — release notes, README files, docs from a cloned repo. No editor UI in the way.

If you need editing, sync, plugins, or note-taking, Obsidian, Typora, and MarkText are great. MarkView is intentionally read-only and lightweight — no account, no sync, no telemetry.

Note: Windows SmartScreen may warn about downloaded .exe / .msi files because they are not code-signed. To bypass: click "More info" → "Run anyway". Alternatively, right-click the file → Properties → check "Unblock" → OK. The Microsoft Store version does not have this issue.

Format	Link	Notes
.deb (1.0.5)	MarkView-1.0.5-amd64.deb	Ubuntu 24.04+, Debian 13+ (uses system WebKit)
.AppImage (1.0.5)	MarkView-1.0.5-amd64.AppImage	Any distro with libfuse2, self-contained
Snap Store		sudo snap install markview-reader
Flathub	flatpak install io.github.scos-lab.MarkView	pending review

# .deb install
sudo dpkg -i MarkView-1.0.5-amd64.deb

# .AppImage install
chmod +x MarkView-1.0.5-amd64.AppImage
./MarkView-1.0.5-amd64.AppImage

After install, open the About panel and click "Set MarkView as default .md handler" to bind .md files to MarkView system-wide.

• Mermaid diagrams — Flowchart, sequence, state, class, ERD, gantt, pie, mindmap, timeline, journey. Each type code-splits on demand.
• Vega-Lite charts — Bar, line, scatter, heatmap, candlestick, stacked area. Any data spec an LLM can emit as JSON renders as SVG.
• KaTeX math — Inline and block, matrices, integrals, series.
• GFM — Tables, task lists, footnotes, strikethrough, autolinks.
• Syntax highlighting — 100+ languages via highlight.js.
• Table of Contents — Auto-generated from headings with scroll spy.
• Folder Browser — Browse a directory tree of .md files.
• In-Document Search — Ctrl+F with highlight and navigation.
• Dark / Light Theme — Follows system preference or manual toggle. Charts re-render with theme-aware colors on switch.
• Font Size Control — Adjustable (12–28 px), persisted.
• File Watching — Auto-reloads when the file is modified externally.
• Drag & Drop — Drop .md files directly into the window.
• Default .md Handler — One-click "Set as default" on both platforms (Windows: auto-registers via registry; Linux: xdg-mime default).
• Recent Files — Last 10 opened files on the welcome screen.
• Print — Clean print output (content only).
• Token Estimation — Document stats with word count and estimated token count.

• Tauri 2 (Rust) — Desktop framework (WebView2 on Windows, WebKit2GTK on Linux)
• React 19 + TypeScript — Frontend UI
• Vite 6 — Build tooling
• TailwindCSS 4 — Styling
• Zustand — State management
• Unified/Remark/Rehype — Markdown processing pipeline

MarkView was born from a controlled experiment: does the format of a specification affect the quality of LLM-generated code?

We wrote the same app spec in three formats — Natural Language (NL), STL, and STLC — and gave each to a separate Google Gemini instance. Same 19 features, same tech stack, same prompt. The results:

STL was the only format that compiled on the first attempt with zero human intervention. The NL version didn't even compile (missing build.rs, broken regex). The STLC version compiled but missed practical details like CSS imports — correct logic, broken visuals.

1. Planning (Claude) — Claude wrote the full application specification in STL format (plan_stl.md), encoding every feature, component, and dependency as typed semantic edges with explicit confidence scores.

2. Implementation (Gemini) — The STL plan was passed to Google Gemini, which generated the complete working codebase — Rust backend, React frontend, styling, and configuration — autonomously in a single pass.

STL (Semantic Tension Language) encodes knowledge as typed, weighted semantic edges:

[MarkView] -> [Markdown_Reader] ::mod(
  rule="definitional",
  confidence=0.99,
  intent="Render .md files into beautifully typeset documents for reading (not editing)"
)

Each edge carries:

• rule — relationship type (causal, definitional, empirical, logical)
• confidence — certainty level (0.0–1.0). 0.99 = hard requirement. 0.7 = nice to have.
• intent — what this actually means in context

Natural language buries priorities between the lines. STL makes them explicit — and transferable between any AI model (Claude, Gemini, GPT) with near-zero information loss.

Learn more about STL at stl-lang.org — source and spec on GitHub.

cd markview
npm install
npm run tauri dev

npm run tauri build

MIT — © 2026 Wuko-Syn DEV / scos-lab. Free to use, fork, and modify.

SCOS-LAB