AI Documentation

Overview

ZenTree Tabs features a local AI-powered tab grouping engine that intelligently organizes browser tabs using semantic understanding, domain clustering, and proximity analysis.

Core Logic

1. Domain-First Phase

Before any AI computation, ZenTree identifies high-confidence domain clusters. If you have 3 or more adjacent tabs from the same root domain (e.g., wikipedia.org), they are instantly grouped with 95% confidence score.

2. Semantic Clustering

For unassigned tabs, the engine converts titles into 384-dimensional vectors (embeddings). It then calculates the Cosine Similarity between these vectors to determine if they share a conceptual intent—even if they come from different websites.

3. Proximity Bias

To respect your browsing flow, we apply an exponential decay penalty to semantic scores based on tab distance. Tabs at opposite ends of your window are much less likely to be grouped than those sitting side-by-side.

Group Naming

Instead of generic labels, ZenTree uses a multi-tier strategy to pick the most descriptive title for a group:

• Reference Detection: Extracts core subjects from documentation-style titles.
• Phrase Frequency: Identifies common 2-3 word segments across the group.
• Semantic Centroid: Finds the representative label that best covers the cluster's intent.

Privacy Architecture

Unlike traditional AI tools, ZenTree's intelligence is localized. The MiniLM model is downloaded once to your local storage and executed exclusively within your browser's memory via WebAssembly.

• No Server Latency: Inference happens at CPU speed on your hardware.
• Air-Gapped: The AI engine has no network permissions; it literally cannot send data anywhere.
• Zero Retention: Your tab data is processed for clustering and immediately flushed.

Troubleshooting AI

If groupings feel inaccurate, consider the following:

• Initialization: The engine requires a one-time download (~23MB) when first used.
• Title Context: The AI relies on page titles. Websites with generic titles (e.g., "Home") provide less semantic signal.
• Complexity: Processing 100+ tabs simultaneously may increase CPU usage for a few seconds.