I am a tab hoarder. At any given time I have 100+ tabs open across multiple Chrome windows, and I’ve accepted that about myself. What I haven’t accepted is losing track of what’s where, having duplicates I don’t know about, and not being able to find that article I was reading three days ago.

So I built Tabby — a Chrome tab manager that connects to your browser via the DevTools Protocol and adds AI-powered organization on top.

How it connectsh2

Tabby doesn’t use a Chrome extension. Instead, it connects directly via Chrome’s DevTools Protocol over WebSocket. You enable remote debugging in chrome://inspect/#remote-debugging, and Tabby reads the DevToolsActivePort file to connect.

This gives it full access to tab lifecycle events — opens, closes, navigations, focus changes. It polls every 30 seconds and tracks everything in a local SQLite database via Drizzle ORM.

The AI layerh2

The interesting part is what happens after tab discovery. Tabby connects to a local Ollama instance for:

• Auto-classification — Every tab gets categorized (work, social, dev, news, etc.) with article detection
• Summarization — Long-form articles get automatic summaries on sync
• Smart grouping — AI-suggested tab groups based on content similarity

All inference runs locally. No data leaves your machine.

Features I actually useh2

Sessions are the killer feature for me. Tabby auto-saves a “Latest” session snapshot on every sync. You can also manually save named sessions, export them as JSON, and restore them later. When Chrome crashes (or I finally close those 100 tabs), nothing is lost.

Duplicate detection finds tabs with the same URL across windows. I’m always surprised by how many I have.

Stale tab detection identifies tabs inactive for 24+ hours. It’s a gentle nudge to either read that article or close it.

Command palette (Cmd+K) lets you search across tabs, history, sessions, and groups. It’s the fastest way to find anything.

The stackh2

Next.js 16 with App Router, shadcn/ui, SQLite via bun:sqlite, and Docker for deployment. The whole thing runs as a local web app with an nginx reverse proxy.

Check it out on GitHub.