Modules

Seek.js is shipped as focused packages you can adopt incrementally. Together they implement the architecture pipeline.

@seekjs/parser

Role: Extract semantic text from HTML and bind each chunk to a source URL (for citations and deep links).

Typical inputs: your static output directory (./dist), a public urlBase, optional CSS selectors for main content, and ignore rules for pages you do not want indexed.

@seekjs/compiler

Role: Turn chunk batches into a compact .msp binary (MessagePack-serialized index) using your chosen embedder (for example Cloudflare Workers AI embeddings in CI).

This is where schema validation, Orama-backed lexical + vector fusion, and compression/quantization strategies (to fight index bloat) live.

@seekjs/client

Role: Hydrate the index in the browser: preload, IndexedDB caching with ETag-aware updates, and hybrid search in WASM-backed engines.

Framework integrations (for example React hooks) expose search, results, and lifecycle helpers so you can build widgets that feel instant.

@seekjs/ai-edge

Role: Stream cited answers from the edge given { query, chunks }. Providers are pluggable; Cloudflare Workers AI is the reference path in the README.

This package is optional if you only want deterministic search—but it completes the “Ask AI” story without sending your whole corpus to a third-party chat API on every request.

---

Install surface: Many apps will start from a meta package such as @seekjs/core once published; until then, follow the README’s proposed imports per module. See Getting Started for a minimal example.