Simple MD

A local-first Markdown editor built with Tauri, React, and Rust. The filesystem is the source of truth — no cloud accounts, no proprietary store, no lock-in.

Features

• Four view modes, with the active document always front and center:
  • Display — read-only rendered view with GFM, syntax highlighting, and KaTeX math
  • Text — raw Markdown source in CodeMirror 6
  • Hybrid — true WYSIWYG editing via Milkdown / ProseMirror, lossless Markdown round-trip
  • Split — source on the left, rendered preview on the right
• Inline charts rendered from fenced ```chart blocks (JSON spec) via Recharts
• Folder workspaces with rename / delete via right-click, plus single-file open from disk
• HTTPS import of remote Markdown (auto-converts github.com/.../blob/... URLs to raw)
• OS file association for .md, .markdown, .mdown, .mkd — opens the app on double-click
• Local AI assistant powered by Gemma 4 running entirely on-device:
  • Pick between Gemma 4 E2B (Q8_0, ~5GB) and E4B (Q4_K_M, ~2.5GB)
  • One-time download with progress, license-gated by Google's Gemma Terms
  • Chat with the active document attached as context
  • Insert / Replace / Append assistant replies straight into the editor

First-run setup

npm install
npm run ai:fetch-sidecar    # downloads llama-server for your platform (~60MB)
npm run icons:rebuild       # rasterizes app icons from public/favicon.svg (optional)
npm run tauri:dev

Required tools on macOS:

• Node 20+
• Rust toolchain (stable)
• librsvg (brew install librsvg) — only needed if you run icons:rebuild

The sidecar fetch script pulls a pinned llama.cpp release from GitHub. If you're on a host the script doesn't recognize, drop a llama-server binary into src-tauri/binaries/llama-server-<rust-target-triple> manually.

Day-to-day commands

npm run tauri:dev                         # full desktop app, hot reload
npm run dev                               # vite-only frontend (some features no-op)
npm run lint                              # eslint flat config
npm run build                             # tsc -b && vite build
npm run tauri:build                       # release bundle
npm run tauri build -- --debug --bundles app    # debug .app bundle
npm run ai:fetch-sidecar                  # refresh the bundled llama-server
npm run icons:rebuild                     # regenerate Tauri icon set from favicon.svg

Project layout

src/
  App.tsx                      shell — modes, tabs, workspaces, AI panel
  app/
    types.ts                   DocumentMode, OpenDocument, ChartSpec, etc.
    sampleDocument.ts          welcome doc shown on first launch
  components/                  Toolbar, TabStrip, StatusBar, ModeToggle, WorkspaceSidebar, Dialog
  features/
    editor/
      EditorPane.tsx           CodeMirror 6 source editor (forward-ref EditorApi)
      HybridEditor.tsx         Milkdown WYSIWYG editor for the Hybrid mode
    preview/MarkdownPreview.tsx  react-markdown + remark-gfm + remark-math + KaTeX
    charts/InlineChart.tsx     Recharts renderer for fenced chart blocks
    reports/                   structured-report descriptors
    ai/                        AICallout, AIPanel, LicenseDialog, useAi hook, runtime IPC
  lib/
    desktop.ts                 thin wrappers around Tauri commands + plugin-dialog
    document.ts                path/string helpers + file:// normalization
    chartSpec.ts               chart JSON validation/series inference
src-tauri/
  src/
    lib.rs                     workspace + file commands, OS file-open routing
    ai.rs                      AI runtime: model download, sidecar lifecycle, chat passthrough
  binaries/                    llama-server + dylibs (NOT committed — see .gitignore)
  examples/probe-runtime.rs    standalone end-to-end runtime test
docs/product-and-technical-spec.md   long-form product+tech spec
scripts/
  fetch-llama-server.sh        downloads the AI sidecar from llama.cpp releases
  rebuild-icons.sh             rasterizes favicon.svg into the Tauri icon set

Architecture notes

• Rust ↔ JS contract: every Tauri command in src-tauri/src/lib.rs and src-tauri/src/ai.rs has a matching wrapper in src/lib/desktop.ts or src/features/ai/runtime.ts. Type changes need to land in both halves.
• AI sidecar runs out-of-process. We spawn llama-server directly from src-tauri/binaries/ via tokio::process::Command (not Tauri's sidecar mechanism, because that wouldn't co-locate the binary with its dylibs). Health check polls /health for up to 120s; if the process dies before that, the user gets the tail of stderr.
• Mode wysiwyg displays as "Hybrid" in the UI. The internal name split used to be hybrid; renamed when the WYSIWYG mode shipped.
• No git repo by default — this project ships without one. Initialize with git init if you want history.

Build output

The macOS debug app bundle lands at:

src-tauri/target/debug/bundle/macos/Simple MD.app

Release builds (npm run tauri:build) target each platform's installer. DMG packaging on macOS still needs polish.

Detailed spec

docs/product-and-technical-spec.md has the long-form product and technical spec.