🐇 Hop down the research rabbit hole • Never lose a tab

Search open tabs by meaning.
Find pages before you leave the keyboard.

A quiet, Spotlight-like search bar that ships a 23MB quantized vector model inside your browser. No cloud, zero telemetry, and sub-millisecond tab switching.

Add to Chrome — Free
Toggle search: ⌘+⇧+Space or Ctrl+Shift+Space

Interactive Live Demo • Try typing below

On-Device • MiniLM
Quick hops:
↑ ↓ navigate · ↵ switch tab
Tabbit 100% On-Device

Built for tab hoarders and deep researchers.

You remember what a page was about, not what title the copywriter gave it.

🧠

100% On-Device Intelligence

Runs a quantized MiniLM-L6 vector model directly inside a local WebAssembly offscreen document. Zero remote code, zero external API keys, zero cloud latency.

⚡

Sub-Millisecond Vector Search

Top-5 cosine similarity ranking computed directly across typed Float32 arrays in IndexedDB. Instant results in under 2ms even with 50+ tabs open.

🥕

Forever Mode

Hop down rabbit holes without fear. Pages you close stay searchable on your machine. Press Enter to resurrect the exact tab in one stroke.

🔒

Guaranteed Zero Telemetry

No analytics, no ad trackers, no server endpoints. Your tabs, browsing history, and vector embeddings remain strictly inside your machine's sandboxed storage.

⌨️

Spotlight-Style Keyboard Bar

Press ⌘⇧Space from any web page. The frosted glass panel pops up in a shadow DOM root, completely immune to host page CSS interference.

🪶

Lightweight Footprint

Consumes ~2 KB per tab. Standard IndexedDB storage takes under 25 MB for thousands of pages. Zero background battery or CPU drain while idle.

Under the hood.

How Tabbit matches tab meaning in milliseconds without cloud dependencies.

STEP 01
Tab Ingestion
Extracts title, domain, and introductory text upon page load via isolated scripting API.
STEP 02
WASM Embedding
Transformers.js runs MiniLM-L6 in an offscreen worker to compute 384-dimensional dense vectors.
STEP 03
IndexedDB Storage
Vectors and metadata are persisted locally in IndexedDB keyed by normalized URL.
STEP 04
Cosine Similarity
Every keystroke runs brute-force dot products across all tabs in ~1ms to return the top 5 matches.

Simple, transparent pricing.

Start free forever, or upgrade to Pro for power-user superpowers.

Tabbit Free

$0 / forever

Perfect for daily tab switching and finding open tabs with speed.

  • Semantic search on up to 50 active tabs
  • 100% on-device MiniLM-L6 vector AI
  • Global shortcut overlay (⌘⇧Space)
  • Forever Mode for last 24 hours
Install Free
Power User

Tabbit Pro

$29 lifetime access

For researchers, developers, and tab collectors with 100+ tabs.

  • Unlimited tabs across all browser windows
  • Unlimited Forever Mode historical recall
  • Deep full-page search & PDF tab indexing
  • Semantic Tab Clustering ("Close tabs about X")
  • Lifetime license key & priority updates
Get Tabbit Pro

Frequently asked questions.

How is Tabbit different from standard browser search?
Chrome's built-in tab switcher only does exact substring matching on titles and URLs. If a tab is titled "Pricing & Coverage" and you search "car insurance", Chrome finds nothing. Tabbit understands the semantic meaning of the page content.
Does running an AI model slow down my machine?
No. The MiniLM-L6 model is quantized to only 22MB and runs strictly in an isolated WebAssembly offscreen worker. It only embeds tabs once upon page load (taking ~100ms in background) and stays completely idle until you search.
Does any of my browsing data leave my computer?
Zero bytes. Tabbit has no backend servers, no analytics, no tracking scripts, and requires no external API keys. All vectors live in your browser's private local IndexedDB.
What is Forever Mode?
By default, closing a tab purges its vector from memory. Turning on Forever Mode keeps closed tabs searchable (taking ≈ 2 KB per page). Pressing Enter on a closed tab result reopens it instantly.