Getting Started

Seek.js is a disaggregated AI search framework that moves indexing to build time, search to the browser, and reasoning to the edge—search that pays for itself by avoiding the recurring cost and operational burden of hosted vector databases for typical documentation and marketing sites.

What you build

The framework is organized around four modules:

1. @seekjs/parser extracts semantic text from rendered HTML.
2. @seekjs/compiler turns chunks into a compact .msp index.
3. @seekjs/client runs hybrid search locally in the browser.
4. @seekjs/ai-edge streams cited answers from the edge.

Install

bun add @seekjs/core

Quick Start

import { extractHtml } from "@seekjs/parser";

const chunks = await extractHtml({
  inputDir: "./dist",
  urlBase: "https://yoursite.com",
});

Next steps

• Architecture — Build-time, browser, and edge responsibilities.
• Modules — Package boundaries and how they connect.
• Roadmap — Experimental milestones and research tracks.
• Comparison — How Seek.js differs from static search, vector DBs, and AI chat SaaS.

Seek.js is designed to eliminate the vector database tax by keeping the search index static and the query path local-first. As APIs solidify, this documentation will track versioned releases and migration notes.