Squash - Browser Memory Layer for AI

Squash adds an invisible memory layer to your browser, compressing every click into portable context for any AI agent.

Now available in the Chrome Web Store

What it does

Squash studies your browsing patterns, learns how you work, and automatically provides relevant context when you use AI tools like ChatGPT or Claude. No more re-explaining yourself to every new AI conversation.

Quick Start

1. Clone and build:

   git clone https://github.com/kstonekuan/squash-browser-memory.git
   cd squash-browser-memory
   pnpm install
   pnpm build

2. Install in Chrome:

   • Open chrome://extensions/
   • Enable "Developer mode"
   • Click "Load unpacked" → select the dist folder

3. Choose your AI provider (in extension settings):

   • Chrome AI (Local): Private, on-device analysis
   • OpenAI-compatible APIs (Recommended): Use local LLMs via Ollama, LM Studio, vLLM, or TGI
   • Claude API: Remote, requires API key
   • Gemini API: Remote, requires API key

Key Features

• Ambient Analysis: Runs in the background, building understanding over time
• Smart Memory: Learns and refines patterns across sessions
• Privacy First: Local analysis by default, explicit consent for remote services
• Context Injection: Adds a distinctive yellow Squash button to ChatGPT and Claude that inserts relevant info from your browsing history

AI Provider Setup

Recommended: Local LLMs via OpenAI-compatible APIs

We strongly recommend using local LLMs for privacy and cost-effectiveness:

• Ollama: Easy local model deployment (e.g., ollama run llama3.2)
• LM Studio: User-friendly GUI for local models
• vLLM: High-performance inference server
• TGI: Production-ready inference

To use: Set the custom base URL in OpenAI provider settings (e.g., http://localhost:11434/v1 for Ollama)

Chrome AI (100% Local)

• Chrome 138+ required
• Enable in chrome://flags → "Prompt API for Gemini Nano"
• Update in chrome://components → "Optimization Guide On Device Model" (~4GB download)

Claude API (Remote)

• Get API key from Anthropic Console
• Enter in extension's Advanced Settings

Gemini API (Remote)

• Get API key from Google AI Studio
• Enter in extension's Advanced Settings

OpenAI API (Local or Remote)

• Run a local model or get API key from OpenAI Platform
• Enter in extension's Advanced Settings
• Supports custom endpoints for Azure OpenAI or other OpenAI-compatible services

Usage

1. Ambient Mode: Enable in Advanced Settings for automatic hourly analysis
2. Manual Analysis: Click "Analyze History" in the side panel
3. Context Injection: Type in ChatGPT/Claude → click the yellow Squash button → select relevant suggestions

Planned Features

• MCP Server Integration: Expose your browsing patterns and profile as an MCP (Model Context Protocol) server for other AI tools
• More AI Providers: Support for additional LLM providers beyond current options
• Enhanced Context Matching: Smarter relevance algorithms for context suggestions
• Workflow Automation: Detect and suggest automation for repetitive tasks

For Developers

Building the Extension

pnpm dev     # Development with hot reload
pnpm build   # Build extension
pnpm check   # Run all quality checks
pnpm test    # Run tests

See ARCHITECTURE.md for technical details.

Privacy

• Local mode: Your data never leaves your device
• Remote mode: Data sent to API provider (requires explicit setup)
• All data stored locally in Chrome storage

Resources

• Simple Chromium AI - Our open source library for Chrome's built-in AI capabilities
• Mem0 Chrome Extension - Inspiration for context injection UX
• Chrome AI Prompt API - Official documentation for Chrome's built-in AI
• Anthropic TypeScript SDK - Official SDK for Claude API integration
• Gemini SDK - Official SDK for Gemini API integration