Parallel Reader

---

Inspired by this reading workflow demo.

Features

• Adaptive segmentation — the LLM decides natural topic boundaries. Short sections merge, long ones split. No dependency on markdown headings.
• Scroll-sync — scrolling the editor auto-highlights the matching card on the right.
• Streaming — real-time SSE streaming for OpenAI Chat and Anthropic APIs, so you see progress as it generates.
• 20+ providers — Anthropic, OpenAI, Gemini, OpenRouter, Groq, DeepSeek, Moonshot, Ollama, LM Studio, and more. Plus Claude Code CLI and Codex CLI backends.
• Persistent cache — summaries are cached by content SHA-1 + settings fingerprint. Reopen a note and cards appear instantly. Edits or config changes show a stale banner.
• Rich rendering — cards render through Obsidian's MarkdownRenderer, so tables, bold, code, and wikilinks all work natively.
• Card editing — right-click any card to copy, edit, delete, or jump to source.
• Export — save cards as a Markdown note in your vault, or copy to clipboard.
• Bilingual UI — full Chinese and English support for commands, settings, and notices.

Quick Start

Step 1: Install the Plugin

1. Go to the Releases page and download three files from the latest release: main.js, manifest.json, styles.css
2. Open your vault folder, navigate to .obsidian/plugins/ (create it if it doesn't exist), and create a new folder called parallel-reader
3. Put the three downloaded files into that folder
4. Open Obsidian → Settings → Community plugins → find Parallel Reader → toggle it on

Tip: Can't see the .obsidian folder? On macOS press Cmd+Shift+. in Finder; on Windows enable "Show hidden files" in File Explorer.

Step 2: Set Up Your AI Provider

1. In Obsidian, go to Settings → Parallel Reader
2. Choose a Provider preset (e.g. Anthropic, OpenAI, DeepSeek, etc.)
3. Paste your API Key
4. (Optional) Change the Model if you prefer a different one
5. Click Test to verify the connection

That's it! Open any note and run the command "Parallel Reader: Generate" from the command palette (Cmd/Ctrl+P).

If you have Claude Code or Codex installed locally, you can use them as backends instead of API keys. Just switch the backend in settings — the plugin automatically detects common install locations. If auto-detection fails, you can manually enter the path in settings.

Usage

How It Works

The LLM returns structured JSON:

{
  "cards": [
    {
      "title": "Short heading",
      "anchor": "Verbatim quote from source for positioning",
      "gist": "One-sentence lead-in",
      "bullets": ["Supporting detail 1", "Supporting detail 2"]
    }
  ]
}

Anchor is the key mechanism — a verbatim quote that the plugin locates via indexOf with multi-level fallbacks, keeping scroll-sync working without relying on headings.

Gist + bullets gives both overview and scannable detail — pure prose felt like a wall of text, pure bullets felt fragmented.

Development

npm install
npm run dev       # watch mode
npm run build     # production build
npm run typecheck # TypeScript strict mode
npm run lint      # Biome
npm test          # build + typecheck + tests
npm run test:unit
npm run test:component
npm run test:contract
npm run test:e2e  # packaged plugin + disposable Vault smoke

For CI / release evidence, run the contract gate:

bash .e2e/gate.sh --json    # writes .e2e/artifact.json (gitignored)

Add TEST_LIVE=1 to opt into real local Vault / provider checks.

Star History

License

MIT