VitePress DepthIndex Guide

VitePress DepthIndex is a production-grade, offline-first search and reasoning engine designed to turn standard documentation sites into AI-native experiences.

Unlike traditional doc-search options (which require heavy cloud scraping, proprietary APIs, or paywalls), DepthIndex works entirely on-device, processing indices in your browser without sacrificing user privacy or incurring API costs.

Key Features

1. On-Device Hybrid Search: Combines BM25 keyword retrieval (exact matching) with dense TF-IDF Cosine similarity (semantic vector similarity) in a single unified client-side search engine.
2. AI-Powered Chat Assistant: Offers a conversational UI directly on top of your docs. Answers questions using local document synthesis or hybrid cloud models.
3. PWA & Offline-First Support: Integrates service workers to cache search indexes and HTML pages, allowing fully functioning documentation search and chat while completely disconnected.
4. LLMs.txt Auto-Generation: Builds optimized documentation summaries at build time (llms.txt, llms-full.txt, and llms.jsonl) for search agents or external LLM scraping.
5. No Telemetry / Privacy-First: Zero external network requests by default. Personalization settings (such as query history and topic affinity tracking) are stored locally in the browser.

Architecture

The plugin is split into two parts:

• Build-Time Indexer: Crawls all markdown files, chunks sections logically, calculates TF-IDF document frequencies, builds inverted index posting lists, and emits compressed index assets.
• Client Runtime: Unpacks compressed index lazily, handles NLP stemming/expansion, and powers the Vue-based Search Modal and Conversational Chat overlay.