Obsidian WebLLMate

中文

Embed AI chat Web Homepages in Obsidian.

Introduction ✨

Select text in your notes or PDFs, automatically send it to the LLM web panel in the sidebar to start a conversation, intelligently save responses to Wiki notes, and create bidirectional reference links in the original text.

Why I built this: While reading papers and notes, I frequently encountered unfamiliar terms. My previous tedious workflow was:

1. Copy the text;
2. Open KIMI or similar sites to ask questions;
3. Manually copy the answer back to my notes.

This repetitive process was exhausting, so I automated the entire workflow into this plugin.

Key Features 🎯

• ⚡ Quick Query: One-click queries on selected text via hotkeys, context menu, or toolbar button

• 🔗 WIKI References: Auto-generates WIKI notes with bidirectional reference links at the highlighted text after receiving answers

• 🔄 Traceable Links: Automatically adds the conversation URL to the WIKI for easy source tracing

• 🔍 History Search: Quickly search through conversation history

• 🌐 Multi-Platform Support: Deeply integrated with KIMI, Qwen, YuanBao, ChatGPT and more

• 📄 PDF++ Enhancement: Highly recommended to use with PDF++ for an enhanced experience

Quick Query ⚡

WIKI References 🔗

History Search 🔍

Traceable Links 🔄

Multi-Platform Support 🌐

PDF++ Enhancement 📄✨

Works with PDF++ to highlight selected text and create bidirectional PDF references

Additional Notes 💡

Plugin Highlights:

• 💰 Zero Cost: All platforms offer free tiers

• 🎨 Beautiful UI: All platforms have well-designed interfaces

• 🌍 Cross-Platform: Chat history syncs across devices

⚠️ Note: This plugin is designed for light LLM usage. For heavy usage, direct API integration is recommended.

RoadMap 🗺️

• Support automatic image/file upload for multimodal conversations (multimodal models excel at formula recognition)

• Build application framework (abstract the workflow as LLM API to implement useful capabilities such as Function Call)

• ......

Adapter Development 🔧

To develop a new adapter, implement the WebLLMAdapter interface from src/types.ts. Inheriting from base classes in src/adapters/bases/ is recommended.

The base class provides an executor: WebExecutor property with a chainable API for convenient DOM manipulation.

Example 🌰:

const html = await this.executor
	.waitFor(selector1)      // Wait for element to appear
	.queryAll(selector2, global=true)  // Query all elements globally
	.at(-1)                     // Get the last element
	.query(selector3)           // Query within the element
	.html()                     // Get innerHTML
	.done();                    // Complete script building and execute

Each chained call doesn't execute immediately but builds a script that runs when done() is called.
See src/utils/webviewer/WebExecutor.ts for details.

Support the Project 💝

If this project helps you, consider sponsoring:

• Afdian: http://afdian.com/a/lumosmoon

• UniFans: https://app.unifans.io/c/lumosmoon

Acknowledgments 🙏

Thanks to KIMI, Tongyi Qianwen, Tencent Yuanbao, ChatGPT and other platforms for their excellent services. This plugin is built upon their web interfaces.