LLM-driven Intelligent Schedule Extraction: Chrome Extension Automatically Identifies Web Events and Syncs to Google Calendar

This Chrome extension is developed by ryoshumei (open-source project, GitHub link: https://github.com/ryoshumei/add-to-calendar). It uses the natural language understanding capabilities of large language models (LLMs) to automatically identify event information (such as title, time, location, and description) on web pages and sync it to Google Calendar with one click, solving the tedious problem of manually copying and pasting schedules. The extension supports cloud API calls (e.g., OpenAI GPT, Anthropic Claude) and local model inference (ONNX Runtime/Transformers.js), balancing accuracy and privacy protection, and is currently in the active development stage.

When browsing the web daily, users often encounter information about activities, meetings, performances, etc. The traditional method of manually extracting key information and filling in the calendar is tedious and error-prone, especially when the information is scattered or the format is inconsistent. With the improvement of LLM capabilities, it has become possible for AI to automatically understand web content and extract structured events. This project combines LLMs with the convenience of browser extensions to create an intelligent schedule extraction tool.

Intelligent Event Recognition: LLMs process unstructured text to recognize natural language time (e.g., "next Wednesday at 2 PM"), relative time, complex locations, and multi-paragraph details, which is superior to traditional rule/regex methods. Chrome Extension Architecture: Adopts Manifest V3 configuration (permission definition), Content Script (page content scanning and triggering), Background Service Worker (API communication and authentication), and Popup interface (user interaction). LLM Integration Strategy: Supports cloud API calls (high accuracy) and local model inference (privacy/offline), and outputs structured JSON (including fields like title, startTime) through prompt engineering.

• Meetings and Event Invitations: Parse meeting topics, times, and participant links, add to the calendar with one click, and reduce the risk of missing events.
• Performances and Ticketing: Extract show times, venues, and seat information to create complete events.
• Course Plans: Batch extract online education schedules, sync to the calendar, and set reminders.
• Travel Itineraries: Extract flight, hotel, and attraction information to build travel schedules.

• Minimal Permission Principle: Only requests necessary permissions and does not collect irrelevant data.
• Local Processing Priority: Supports local model inference, so sensitive content does not need to be transmitted to the cloud.
• Transparent Data Processing: Clearly displays analyzed content, and users have control over their data.
• Google OAuth Authentication: Uses OAuth 2.0 to access calendars, does not store passwords, and tokens are refreshed regularly.

Event Structuring: Convert LLM-extracted information into Google Calendar format, handling time zone conversion, all-day/timed events, recurrence rules, and location map link formatting. Interaction Optimization: Context-aware triggering (only displays the button on event pages/selected areas), progressive confirmation (form check and modification), and shortcut operations (hotkeys/right-click menu).

Planned features include: supporting more calendar services (Outlook, Apple, Notion), intelligent conflict detection, batch event extraction, custom templates (optimization for specific websites), and team collaboration features (shared rules and calendars).

As an open-source project, it not only provides a practical tool but also demonstrates an example of combining LLMs with traditional applications. The modular design facilitates community contributions; developers can refer to the architecture to integrate LLMs into extensions or web applications, with clear contribution paths (improving prompts, adding calendar support, optimizing UI, etc.).