TabLab

Lab for bookmark hoarders

A Chrome extension that turns your tabs and bookmarks into an analytics workspace. Local-first — all AI runs in the browser or on your machine, nothing leaves your device.

---

Screenshots

AI categorization — tabs and bookmarks classified by topic with intent labels

Semantic Map — 2D embedding projection clusters related pages automatically

Topic River — how your interests shift over time

Personal Radar — browsing pattern overview across categories

Tag Co-occurrence — discover connections between topics in your collection

---

What it does

TabLab enriches your browser data with AI and visualizes it in 18+ views:

Category	Views
Organization	List, Kanban, Triage, Shelf
Exploration	Semantic Map, Tag Constellation, Domain Graph, Treemap
Time	Timeline, Activity Heatmap, Topic River, Session Story
Discovery	Magazine, Reading Queue, Focus Rings, Personal Radar
Analysis	Overlap Explorer, Shadow Map, Domain Drilldown, Tag Co-occurrence

AI enrichment (local, no cloud):

• Category — classifies each tab/bookmark into a topic (Development, Finance, AI & ML, …)
• Tags — 3–5 topical tags per page for filtering
• Intent — how you use the page: article, reference, tool, service, transactional, video, social, repository
• Embeddings — semantic vectors for similarity search and 2D projection (Semantic Map)

---

AI providers

Everything runs locally:

Provider	Used for	Notes
Gemini Nano	Chat (classify/tag/intent)	Built into Chrome, no setup ** NOT WORKING YET **
LM Studio	Chat + Embeddings	Any OpenAI-compatible local server
WebLLM	Chat	WebGPU in-browser, no server needed
Transformers.js	Embeddings + Zero-shot NLI	ONNX/WASM, runs in offscreen document

---

Getting started

Prerequisites

• Chrome 120+ (or any Chromium-based browser with extensions)
• Node.js 18+, pnpm

Install & build

git clone https://github.com/aicrafted/tab-lab
cd tab-lab
pnpm install
pnpm build

Load in Chrome

1. Open chrome://extensions
2. Enable Developer mode
3. Click Load unpacked → select the dist/ folder

Configure AI

Click the settings icon in the extension header. Options:

• Gemini Nano — enable in chrome://flags/#optimization-guide-on-device-model (no extra setup)
• LM Studio — run LM Studio with local server on http://localhost:1234/v1, pick a chat and embedding model
• WebLLM — select a model (e.g. Qwen3-0.6B), it downloads once and runs on WebGPU

---

Development

pnpm dev        # watch mode — rebuilds on save
pnpm typecheck  # TypeScript check without emitting

After each build, reload the extension in chrome://extensions (click the refresh icon).

Project structure

src/
  background/     Chrome service worker
  offscreen/      Transformers.js host (WASM inference, embeddings, NLI)
  lib/
    classifier.ts  Category classification
    intent.ts      Intent classification
    tagger.ts      Tag generation
    embedder.ts    Embedding fetch + 2D projection (UMAP)
    llm.ts         chatComplete() — routes to Gemini Nano / LM Studio / WebLLM
    webgpu-provider.ts  Offscreen messaging client
  components/
    views/         All 18+ view components
  pages/main/      Extension popup (React app)

---

License

MIT — github.com/aicrafted/tab-lab

© 2025 AICrafted