Topic: Web Security & Components
Rozz Architecture
Rozz is built as an independent, embeddable web component that runs entirely in the browser.
It leverages a lightweight semantic-search engine and a small LLM interface.
1. Component Layer
- Lit-based Web Component — Encapsulates its own DOM and styles, preventing interference with the host page.
- Keyboard & ARIA support — Full accessibility, screen-reader friendly, WCAG 2.1 AA compliant.
2. Crawling & Embedding
- Client-side crawler — Walks the site’s HTML, PDFs, and other documents, extracting text and building a vector index.
- Embeddings — Generated using a lightweight on-device transformer, stored in IndexedDB for offline use.
3. Query Processing
- Semantic — Matches user queries against the index, returning the most relevant passages.
- LLM "chatbot" layer — A small, fine-tuned model (e.g., GPT-4-turbo-style) formats answers, summarizes, and handles follow-ups.
- Prompt guardrails — Built-in protection against XSS, prompt injection, and cross-domain leakage.
4. Data Flow
1. User types a question →
2. Component queries the local vector index →
3. Result snippets passed to the LLM →
4. Formatted response displayed in the chat UI.
No sensitive data is sent to external servers; all processing stays on the client, keeping user privacy intact.
5. Deployment & Integration
- Add a single <script src="https://cdn.rozz.site/rozz.js"></script> and <rozz-></rozz-> tag to your site.
- No backend integration required; the component can be used on static sites, CMS, or SPAs.
Sources
- Rozz Chatbot Security & Privacy
- Introducing the Rozz Dashboard
- Rozz Accessibility Features
What aspect of the architecture would you like to dive deeper into—crawling strategy, embedding model, or integration details?