AgenWork

Stop wasting time reading lengthy articles or switching between browsing tools. AgenWork processes everything locally. Write, summarize, translate, and chat instantly, privately.

Comment

🌟 Inspiration

The inspiration for AgenWork came from observing two critical pain points in modern web browsing:

  • Privacy Concerns: Users are increasingly worried about sending their browsing data to external AI services, yet they still need intelligent assistance while browsing.
  • Fragmented AI Tools: Different AI tasks (summarization, translation, content analysis) typically require switching between multiple services, breaking workflow and user focus.

We asked ourselves: "What if AI could work directly in your browser, processing everything locally, while seamlessly adapting to your needs?" This question led us to explore Chrome's Built-in AI APIs, which promised on-device processing with zero external API calls. We envisioned a single, intelligent agent that could understand user intent and automatically route requests to specialized AI capabilities, all while keeping data private and offline-capable.

🚀 What it does

AgenWork is an intelligent Chrome extension that brings powerful AI capabilities directly into your browser through a sophisticated multi-agent architecture. Here's what makes it special:

🤖 Multi-Agent System

  • Summarizer Agent: Extracts key points, creates TL;DRs, generates headlines, or crafts engaging teasers from any webpage
  • Translator Agent: Real-time language detection and translation supporting 30+ languages
  • Prompter Agent: General-purpose conversational AI for answering questions and understanding context
  • Writer Agent: Writing assistance

🎯 Intelligent Intent Detection

AgenWork doesn't require commands or syntax. Just talk naturally:

  • "Give me the key points from this article" → Automatically detects summarization with key-points format
  • "Quick TL;DR please" → Understands you want a brief, concise summary
  • "Translate this to Spanish" → Routes to translator with proper language settings
  • "What is quantum computing?" → Engages conversational AI for explanations

How we built it

AgenWork System Architecture

🔒 Privacy-First Design

  • All AI processing happens on-device using Chrome's Built-in AI (Gemini Nano)
  • No external API keys required
  • No data transmission to external servers
  • Works offline once models are downloaded
  • Zero tracking or data collection

💾 Offline Capabilities

  • Local conversation storage using DexieJS (IndexedDB wrapper)
  • Persistent chat history across sessions
  • Full functionality without an internet connection

🎨 User-Friendly Interface

  • Clean, intuitive chat interface
  • Draggable floating icon for quick access on any webpage
  • Real-time progress tracking for model downloads
  • Markdown-formatted responses for better readability

Technical Features:

  • Modular Architecture: Clean separation of concerns with 4 specialized agents
  • Smart Context Handling: Automatically extracts and provides relevant page content to AI agents
  • Progressive Enhancement: Gracefully degrades when AI APIs are unavailable
  • Comprehensive Error Handling: User-friendly error messages with actionable recommendations

💪 Challenges we ran into

Creating an AI system that accurately detects user intent from natural language was challenging. We needed to:

  • Design sophisticated prompts for the Language Model
  • Implement fallback pattern-matching for when AI isn't available
  • Handle ambiguous requests intelligently
  • Detect subtle preferences (e.g., "brief summary" vs "detailed overview")

🏆 Accomplishments that we're proud of

True Privacy-First AI: We successfully created an AI agent assistant that genuinely respects user privacy. No data ever leaves the user's device, no API keys are needed, and it works completely offline. This is rare in today's AI landscape. Thanks to Google Chrome's Built-in AI APIs.

Intelligent Intent Detection: Our multi-layered intent detection system (AI + fallback patterns) achieves accuracy in understanding what users want. It automatically detects:

  • Summary types (key-points, tldr, teaser, headline)
  • Length preferences (short, medium, long)
  • Language preferences for translation
  • Task complexity and routing needs

What we learned

Privacy-First Design Is Achievable

  • On-device AI is feasible for real-world applications
  • Local storage (IndexedDB) provides excellent offline capabilities
  • Users appreciate privacy-respecting AI tools

Intent Detection Requires Multi-Layered Approach

  • AI-powered detection is excellent but unstable
  • Fallback pattern-matching provides reliability
  • Combining both approaches creates robust systems

What's next for AgenWork

Cross-Browser Support

  • Firefox extension (when AI APIs are available)
  • Edge extension with specific optimizations
  • Safari extension (contingent on API support)

Multi-Modal AI: Integrating image and voice capabilities Proactive Assistance: AI that suggests actions before being asked Learning User Preferences: Personalized AI that adapts to individual usage patterns

AgenWork represents our vision of AI that respects privacy while providing powerful, intelligent assistance. We're excited to continue building features that make browsing smarter, more efficient, and more private.

Read Full Technical Documentation

Built With

Share this project:

Updates