🚀 About Split Soul
Split Soul is a Chrome Extension that introduces a new concept — AI-powered digital companions called “souls” that work quietly in the background to enhance your browsing experience. Each soul represents a unique AI persona with specific roles — from summarizing your browsing activity to providing intelligent insights and proactive suggestions — all running locally with Chrome’s built-in AI APIs.
💡 Inspiration
Managing productivity and focus while browsing the web has always been a challenge. With the rise of client-side AI, I wanted to explore how we could use on-device intelligence to create digital “companions” that understand what we do online and help us act smarter, without sending data to the cloud.
The idea for Split Soul came from imagining multiple AI assistants — or souls — each handling a different aspect of digital life, coordinated by a main “soul” that observes, summarizes, and learns.
🧠 What It Does
- AI Companions (Souls): Create multiple AI-driven agents with unique personalities and roles.
- Activity Monitoring: Souls observe browsing patterns and analyze time spent on different sites.
- Intelligent Summaries: The Summarizer API condenses content from visited pages into quick, clear takeaways.
- Action Suggestions: The Prompt, Writer, and Rewriter APIs help generate helpful ideas or recommendations based on your browsing context.
- Privacy-First AI: All analysis happens locally, using Chrome’s built-in Gemini Nano model — ensuring no data ever leaves your device.
- Vector Search: Uses transformer embeddings and Dexie (IndexedDB) for fast, local semantic similarity search.
🧩 How I Built It
- Frontend: React 18 + TypeScript + Material-UI (MUI)
- Build Tool: Vite
- AI Layer: Chrome Built-in AI (Prompt, Summarizer, Rewriter, Writer, Translator, Proofreader APIs)
- Database: IndexedDB (via Dexie) for local storage
- AI Models: Xenova Transformers for local embedding and vector search
- Extension: Chrome Extension (Manifest V3)
- Runtime: Node.js v22.17.0 + npm v9.8.1
The extension runs entirely on the client, leveraging Gemini Nano for low-latency inference and seamless offline functionality.
🧰 Challenges
- Understanding how to combine multiple built-in AI APIs efficiently inside a Chrome Extension.
- Managing persistent local data with privacy while keeping vector search performant.
- Designing a clean and responsive UI using Material-UI within Chrome’s sandboxed environment.
- Handling structured output from AI APIs and aligning multiple AI agents (“souls”) for coordinated responses.
🎓 What I Learned
- The true potential of on-device AI with Chrome’s built-in APIs — especially how Prompt and Summarizer can power contextual awareness.
- Best practices for Manifest V3 background scripts and handling AI context flow efficiently.
- Implementing AI privacy patterns and hybrid inference using local embeddings for faster insights.
- The value of modular architecture when building multi-agent systems within browser constraints.
🌈 What’s Next
I plan to extend Split Soul by:
- Adding multimodal support (voice and image understanding).
- Introducing main–sub soul communication for coordinated multi-agent behavior.
- Incresing the scale from Agentic to MCP based flow.
Split Soul reimagines your browser as a space where intelligent AI companions quietly help you think, focus, and act — all while keeping your data where it belongs: with you.
Built With
- ai-agents
- chrome
- chrome-built-in-ai-apis
- client-side-ai
- dexie
- gemini-nano
- html5
- indexeddb
- javascript
- local-storage
- manifest-v3
- material-ui
- node.js
- npm
- on-device-ai
- privacy-first-ai
- prompt-api
- proofreader-api
- react
- rewriter-api
- summarizer-api
- translator-api
- typescript
- vector-search
- vite
- web-ai
- writer-api
- xenova-transformers
Log in or sign up for Devpost to join the conversation.