Obsidian Parallel Reader

Split-view reading for Obsidian — your original note on the left, AI-generated summary cards on the right, with scroll-sync highlighting.

English · 中文

---

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).

Supported providers

Provider	Notes
Anthropic	Default, recommended
OpenAI	GPT models
Google Gemini	Gemini models
OpenRouter / Groq / DeepSeek / Moonshot / ...	OpenAI-compatible
Ollama / LM Studio	Local models, no API key needed
Custom endpoint	Any OpenAI or Anthropic compatible API

CLI backends (advanced)

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

Action	Effect
Click a card	Jump editor to that section
Right-click a card	Context menu: copy, edit, delete, jump to source
Scroll the editor	Right-side card auto-highlights
Alt+↑ / Alt+↓	Navigate between cards
Enter in summary pane	Jump to active card's source line
Ribbon icon	Open the parallel reader pane
File context menu	Generate / regenerate / clear cache

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

The project also has a host-neutral e2e contract gate for product-shell smoke evidence:

bash .e2e/gate.sh --json

If e2e_contract_validator is not installed, set E2E_CONTRACT_VALIDATOR_PYTHONPATH to the claude-code-addons/scripts directory before running the gate. Runtime evidence such as .e2e/artifact.json and .e2e/results/ is generated locally and ignored by git.

Test classification lives in tests/catalog.json. verify owns unit, component, contract, and local integration checks. Mocked Obsidian/runtime tests are classified as component or contract tests, not product e2e. The .e2e gate intentionally includes only the default product-shell smoke plus optional live local Vault/provider checks with TEST_LIVE=1.

Star History

License

MIT