Documentation
MemorieAI drops a voice-powered AI guide onto any website via a single script tag. The assistant reads the live DOM, navigates pages, highlights elements, and completes tasks on behalf of the user — in real time.
Quickstart
Paste this tag before </body> on any page:
That's it. The MemorieAI bubble appears in the bottom-right corner. Your users can click it, speak or type, and the assistant will guide them through your site.
Don't have an API key yet? Request early access and we'll set you up within 24 hours.
How It Works
When the page loads, embed.js injects a small floating bubble and panel into your DOM. When the user activates it:
- The script builds a semantic map of the current page — every button, link, heading, form field, and section — by reading the live DOM.
- That map is injected into the AI's context so it knows what's on screen right now.
- The user speaks or types. The AI understands intent and selects the right browser tool to act.
- Tools execute on the page: navigate, highlight, click, fill, scroll, or read.
- After any navigation the map is rebuilt for the new page automatically.
Configuration
Script Tag
Data Attributes
| Attribute | Default | Description |
|---|---|---|
| data-key | — | Your MemorieAI API key. Required. |
| data-lang | auto | Force a language (en, es). Default detects from browser. |
| data-position | bottom-right | Bubble position: bottom-right, bottom-left. |
| data-theme | auto | Color theme: light, dark, auto. |
| data-color | #004488 | Primary accent color (hex). Applies to button and highlight. |
| data-label | MemorieAI | Name shown in the panel header. |
Custom Prompt (Pro+)
Pro and Enterprise customers can define a custom system prompt per site from the dashboard. The prompt receives the current page map automatically — you only need to describe your site's structure, workflows, and personality.
Include exact URL paths for your most common tasks. The AI will navigate directly to them when user intent matches — no vague descriptions needed.
Browser Tools
MemorieAI has six tools it can call on any page. Selectors used by the AI are always drawn from the live page map — it never invents IDs.
| Tool | Parameters | What it does |
|---|---|---|
| navigate_to | url | Navigate to any URL (relative or absolute). Page map is rebuilt after load. |
| highlight_element | selector | Places a glowing animated border on the element so the user sees exactly where to look. |
| click_element | selector | Programmatically clicks a button, link, or any interactive element. |
| fill_field | selector, text | Types text into an input, textarea, or contenteditable field. |
| scroll_to | selector | Smoothly scrolls the page so the element is in view. |
| read_content | selector | Reads the text content of an element aloud to the user. |
Multi-Tenant Architecture
Each customer gets their own data-key. The key is validated server-side against the requesting domain — so a key registered for acme.com will not work on evil.com.
The Vapi API key and assistant ID are never exposed to the browser. They live only on the MemorieAI backend and are returned as short-lived session tokens.
Domain Locking
From your dashboard, register the domains each key is allowed on. Requests from unlisted domains are rejected with a 403 domain_not_allowed error before any Vapi session is created.
localhost and 127.0.0.1 are always allowed for development — no need to register them.
Languages
MemorieAI detects the user's language automatically from their browser locale and from the language they speak or type. The assistant responds in the same language. Currently well-tested:
- English (en)
- Spanish (es)
- More coming — request a language