Jina Reader — Local Edition (v7)
Professional-Grade • Local-First • Privacy-Centric
Transform any webpage into high-fidelity, LLM-friendly Markdown entirely within the browser.
Designed for AI Workflows. Engineered for Privacy.
The Philosophy
In an era of cloud-dependency and rate-limited APIs, Jina Reader (Local) represents a pivot back to local-first computing. This project eliminates the middleman, removing the need for r.jina.ai cloud calls by implementing the entire extraction and transformation pipeline directly in the browser’s runtime. Content remains local, speed is absolute, and usage is infinite.
Architectural Excellence
The system is built on a modular, Service-Oriented Architecture (SOA) designed to handle the complexities of the modern web (Shadow DOM, dynamic iframes, and SPA layouts).
1. High-Level System Flow
The following diagram illustrates the lifecycle of a single “Copy” request, from the user’s action to the final clipboard entry.
graph TD
User([User Action: Click/Hotkeys/Menu]) --> Orchestrator[Background Orchestrator]
Orchestrator --> Config[Config Service]
Config --> Service[LocalReaderService v7]
subgraph "Target Tab Context"
Service -- Script Injection --> Engine[v7 Intelligence Engine]
Engine -- DOM Scoring --> Root[Content Root Discovery]
Root -- Cleaning --> Clone[Sanitized Clone]
Clone -- Conversion --> MD[Markdown/JSON/HTML]
end
MD -- Message Passing --> Service
Service -- Clipboard API --> OS[System Clipboard]
Service -- Notification API --> User
2. The v7 Intelligence Engine
At the heart of the project is the Heuristic Scoring Engine. Instead of “guessing” where the content is, it calculates a text-to-link density ratio and penalizes interactive “noise” (dashboards, buttons, navs).
sequenceDiagram
participant P as Web Page
participant E as Intelligence Engine
participant S as Scoring Module
participant C as Converter
E->>P: Traverse DOM Tree
P-->>E: Candidate Elements (div, main, article)
E->>S: Compute (Text Density - Link Density)
S-->>E: Final Scores
E->>P: Flatten Shadow DOM & Iframes
E->>C: Execute Pro-Grade Markdown Conversion
C-->>E: LLM-Friendly Payload
Feature Matrix
| Feature |
Local Engine (v7) |
Cloud Reader |
| Privacy |
100% Local |
Cloud Processed |
| Rate Limits |
Infinite |
Strictly Limited |
| Cost |
$0 |
Token Based |
| Shadow DOM |
✅ Supported |
❌ Limited |
| Table Formatting |
✅ Pro Grade |
✅ Basic |
| Full Pageshots |
✅ Stitched Canvas |
❌ N/A |
Repository Structure
p:/extensions/JinaClip - Copy Page for LLM/
├── scripts/
│ └── build.js # Professional packaging pipeline
├── src/
│ ├── background/
│ │ └── index.js # Lifecycle & Event Orchestration
│ ├── services/
│ │ └── localReader.js # Core Extraction & Transformation Engine (v7)
│ ├── utils/
│ │ └── logger.js # System-wide observability
│ └── config/
│ ├── constants.js # Immutable definitions
│ └── defaultSettings.js
├── icons/ # High-fidelity visual assets
└── manifest.json # Extension Manifest (MV3)
Technical Pipeline: Full Page Capture
The “Pageshot” feature utilizes a multi-stage stitching pipeline to capture full-length articles without loss of detail.
- Dimension Analysis: Calculates
scrollHeight and viewHeight.
- Synchronized Scrolling: Executes discrete jumps with paint-settle delays.
- Canvas Orchestration: Assembles visible segments in a
OffscreenCanvas.
- Data Delivery: Converts the final buffer to a high-quality Markdown-wrapped Base64 string.
Installation & Development
Onboarding for Contributors
The codebase is written in vanilla ES6+ to ensure zero-coupling and maximum longevity.
- Clone & Setup:
git clone <repo-url>
npm install
- Build for Production:
npm run build
- Browser Loading:
- Open
chrome://extensions/
- Load the
dist/ folder as an “Unpacked Extension.”
Professional Usage
- Quick Action: Left-click the extension icon for a standard Markdown copy.
- Precision Extraction: Right-click the icon to select specific formats (JSON, HTML, Text).
- Automation: Use the Global Shortcut
Alt+Shift+J for immediate capture.
Credits & License
This project was architected and developed by Ahmad Hassan (B-Ted) as a part of the movement toward high-performance, local-first AI tools.
Distributed under the MIT License. Contributions that align with the core philosophy of “Clean Code & User Privacy” are welcomed.
"The best way to read the web is locally."
评论