MarkView

A clean, fast, lightweight Markdown reader for Windows.

Download

Version	Link
Microsoft Store	Get it from Microsoft Store
Portable (.exe)	MarkView_v1.0.3_portable.exe
Installer (MSI)	MarkView_v1.0.3_x64.msi

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.

Features

• Markdown Rendering — GFM tables, task lists, math/KaTeX, code syntax highlighting
• Table of Contents — Auto-generated from headings with scroll spy
• Folder Browser — Open a folder and browse all .md files in a tree view
• In-Document Search — Ctrl+F with highlight and navigation
• Dark / Light Theme — Follows system preference or manual toggle
• Font Size Control — Adjustable (12–28px), persisted across sessions
• File Watching — Auto-reloads when the file is modified externally
• Drag & Drop — Drop .md files directly into the window
• Print — Clean print output (content only)
• Token Estimation — Document stats with word count and estimated token count

Tech Stack

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

How This Project Was Built

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:

Metric	Natural Language	STL	STLC
Feature Completion	8.5/19 (45%)	13/19 (68%)	10/19 (53%)
Compiles?	NO	YES	YES
Fully Autonomous?	NO	YES	NO
Human Interventions	1+	0	1

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.

The Process

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.

What Is STL?

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.

Development

cd markview
npm install
npm run tauri dev

Build

npm run tauri build

License

MIT

Developed by

SCOS-LAB