Semantic Memory of Browsing history Autor: ArnavLifelessCoder
Your AI-powered second brain for search browsing history by meaning, get instant summaries, discover connections. 100% on-device.
Metadane rozszerzenia
Zrzuty ekranu
O tym rozszerzeniu
✨ New
Ask My Memory — end a search with a question mark (e.g. "how does HNSW indexing work?") and get a synthesized answer pulled from across your browsing history, with clickable sources. Fully on-device — no AI servers.
Memory Map — a brand-new tab that plots your entire knowledge space as an interactive 2D map. Related pages cluster together by topic; hover to preview, click to open.
Address-bar search — type mem then your query in the address bar to search your memory without opening the popup.
Light / Dark / Auto theme — pick one in Settings; Auto follows your system.
Backup & restore — export your full index to a file and import it back (great for moving between devices).
🔧 Improved
Smarter search ranking that mixes meaning, keywords, titles, and recency — and hides weak, unrelated matches.
Filter results by time (all / today / week / month) and navigate them with ↑/↓ + Enter.
Cleaner results: search-engine pages, chat apps, and webmail are no longer indexed, and revisiting a page updates it instead of creating duplicates.
A real last-7-days activity chart, per-page delete in History and results, and a "pause indexing" switch.
🔒 Privacy
Removed all third-party favicon requests — everything now renders locally. Still 100% on-device by default.
Semantic Memory is a Chrome extension that makes browser history actually useful. Instead of relying on exact keywords, it indexes the pages you visit using local AI embeddings so you can search by meaning. If you vaguely remember reading "that article explaining transformers" or "the blog comparing PostgreSQL and MongoDB", it'll find it even if those exact words never appeared in the title.
Everything runs locally in your browser by default. Your browsing data never leaves your device unless you explicitly enable encrypted sync. Along with semantic search, the extension can summarize the current page, find similar pages you've visited before, organize your history into a timeline, and show simple analytics about your browsing habits.
Built with Manifest V3, React, TypeScript, IndexedDB, ONNX Runtime, and Transformers.js, with an optional FastAPI backend for encrypted cross-device sync.
Ask My Memory — end a search with a question mark (e.g. "how does HNSW indexing work?") and get a synthesized answer pulled from across your browsing history, with clickable sources. Fully on-device — no AI servers.
Memory Map — a brand-new tab that plots your entire knowledge space as an interactive 2D map. Related pages cluster together by topic; hover to preview, click to open.
Address-bar search — type mem then your query in the address bar to search your memory without opening the popup.
Light / Dark / Auto theme — pick one in Settings; Auto follows your system.
Backup & restore — export your full index to a file and import it back (great for moving between devices).
🔧 Improved
Smarter search ranking that mixes meaning, keywords, titles, and recency — and hides weak, unrelated matches.
Filter results by time (all / today / week / month) and navigate them with ↑/↓ + Enter.
Cleaner results: search-engine pages, chat apps, and webmail are no longer indexed, and revisiting a page updates it instead of creating duplicates.
A real last-7-days activity chart, per-page delete in History and results, and a "pause indexing" switch.
🔒 Privacy
Removed all third-party favicon requests — everything now renders locally. Still 100% on-device by default.
Semantic Memory is a Chrome extension that makes browser history actually useful. Instead of relying on exact keywords, it indexes the pages you visit using local AI embeddings so you can search by meaning. If you vaguely remember reading "that article explaining transformers" or "the blog comparing PostgreSQL and MongoDB", it'll find it even if those exact words never appeared in the title.
Everything runs locally in your browser by default. Your browsing data never leaves your device unless you explicitly enable encrypted sync. Along with semantic search, the extension can summarize the current page, find similar pages you've visited before, organize your history into a timeline, and show simple analytics about your browsing habits.
Built with Manifest V3, React, TypeScript, IndexedDB, ONNX Runtime, and Transformers.js, with an optional FastAPI backend for encrypted cross-device sync.
Ocenione na 5 przez 4 recenzentów
Uprawnienia i dane
Wymagane uprawnienia:
- Mieć dostęp do kart przeglądarki
- Mieć dostęp do danych użytkownika na wszystkich stronach
Opcjonalne uprawnienia:
- Mieć dostęp do danych użytkownika na wszystkich stronach
Zbieranie danych:
- Autorzy tego rozszerzenia twierdzą, że nie wymaga ono zbierania danych.
Więcej informacji
- Odnośniki dodatku
- Wersja
- 2.2.0
- Rozmiar
- 306,38 KB
- Ostatnia aktualizacja
- 1 dzień temu (6 lip 2026)
- Powiązane kategorie
- Licencja
- Licencja MIT
- Historia wersji
- Dodaj do kolekcji
How It Works
You browse the web
↓
Content script extracts clean text (Readability.js)
↓
Text is chunked (512 tokens, 64 overlap)
↓
Chunks are embedded in-browser (MiniLM-L6-v2, ONNX int8, ~23ms/chunk)
↓
Embeddings stored in an in-memory vector index + IndexedDB (persisted)
↓
You search from the popup: "that article about transformers"
↓
Query is embedded → ANN search → ranked results with similarity scores
Everything above runs entirely in your browser. No servers, no API calls, no data leaves your machine.
The optional backend adds:
- Cross-device sync (end-to-end encrypted — server sees only opaque blobs)
- Cross-encoder re-ranking for precision boost
- Topic clustering + knowledge graph (UMAP + k-means)
Features
The popup is organised into four tabs, all powered by on-device ML:
| Feature | What it does |
|---|---|
| Semantic Search | Natural-language search over your history with per-result similarity scores. Matching query terms are highlighted in titles and previews, and results are deduplicated by page (best-scoring chunk wins). |
| Quick Summary | One-click extractive summary + key points of the page you're currently on (centroid-based sentence ranking). |
| Find Similar | Surfaces related pages from your history based on the current page's content. |
| Timeline | Your browsing history grouped by day, filterable by today / week / month / all. |
| Analytics | Total pages, chunks, domains, reading time, and a top-domains breakdown. |
| Settings | Domain blacklist (skip indexing), opt-in encrypted sync config, and data export / clear. |
| Context menu | Right-click selected text → "Search Semantic Memory" (opens the popup pre-filled). |
| Shortcut |
Ctrl+Shift+S(Cmd+Shift+Son macOS) opens the popup. |Recent searches, animated stats, and a live "index active" indicator round out the UI.