MemexNG-Semantic History Search 作者： Durga Gairhe

MemexNG — Semantic History Search

Find anything you've ever read. MemexNG brings AI-powered semantic search to your browsing history — completely private and fully local. Instead of searching with exact keywords, simply describe what you remember.

🔍 Semantic Search

Find pages by meaning, not exact words:
- "that article about AI privacy risks"
- "React authentication tutorial"
- "video about local LLMs"
- "CSS animation performance guide"
Uses on-device AI embeddings to understand the meaning behind your searches.



🧠 Ask Questions About What You've Read

Use natural language to search and summarize information from your browsing history:
- Retrieve relevant pages
- Surface matching excerpts
- Provide cited answers with clickable source references
- Rediscover forgotten research
Every answer includes clickable source references.



⚡ Fast & Lightweight

• Instant popup search
• Fast HNSW vector search
• Optimized for large histories
• Dark / light / system theme

🔒 100% Private — No Cloud

• Data never leaves your device
• No servers, no telemetry, no accounts
• No tracking, no analytics
• Works fully offline after model download

🏷️ Features

• Semantic history search
• AI natural language queries
• Local vector search index
• Reading list & annotations
• Omnibox search
• Domain filtering & relevance controls
• Indexed page statistics
• Configurable privacy controls

🛡 Permissions

• <all_urls> — page content extraction for indexing
• storage / unlimitedStorage — stores your local search index
• activeTab — reads current page when indexing
• history — optional browsing history indexing
• offscreen — runs AI embedding generation efficiently
  All processing happens locally on your device.

🧩 Open Source

Built with Transformers.js, ONNX Runtime Web, Mozilla Readability, Voy vector search, React + TypeScript.
MemexNG — Your History, Augmented.