# @astermind/cybernetic-chatbot-client

Offline-capable AI chatbot client with local RAG fallback and agentic capabilities for [AsterMind](https://astermind.ai).

[![npm version](https://img.shields.io/npm/v/@astermind/cybernetic-chatbot-client.svg)](https://www.npmjs.com/package/@astermind/cybernetic-chatbot-client)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![TypeScript](https://img.shields.io/badge/TypeScript-5.0-blue.svg)](https://www.typescriptlang.org/)

## What is Cybernetic Chatbot Client?

Cybernetic Chatbot Client is the official JavaScript SDK for integrating [AsterMind](https://astermind.ai) AI chatbot capabilities into your web applications. It provides a robust, offline-first architecture that ensures your users always get answers, even when disconnected from the server.

## Key Features

- **Dual Transport** - WebSocket streaming for SaaS, REST+SSE for on-prem (auto-detected)
- **Offline-First Architecture** - IndexedDB caching with TF-IDF local search
- **SSE Streaming** - Real-time token-by-token responses (REST fallback)
- **WebSocket Streaming** - Low-latency streaming via persistent connection (SaaS)
- **Session Management** - Multi-turn conversation continuity
- **Configurable Retry Logic** - Exponential backoff with customizable settings
- **Connection Status Monitoring** - Real-time online/offline detection
- **Maintenance Mode Support** - Graceful degradation per ADR-200
- **Agentic Capabilities** - Intent classification and DOM automation (full bundle)
- **Tree-Shakeable** - Import only what you need

## Table of Contents

- [Installation](#installation)
- [Quick Start](#quick-start)
- [Licensing](#licensing)
- [Configuration](#configuration)
- [Features](#features)
  - [WebSocket Transport (SaaS)](#websocket-transport-saas)
  - [Offline-First Architecture](#offline-first-architecture)
  - [Pre-computed Vector Export (Advanced)](#pre-computed-vector-export-advanced)
  - [Streaming Responses](#streaming-responses)
  - [Session Management](#session-management)
  - [Agentic Capabilities](#agentic-capabilities)
  - [Sitemap Configuration](#sitemap-configuration)
  - [Maintenance Mode Support](#maintenance-mode-support)
- [Bundle Options](#bundle-options)
- [API Reference](#api-reference)
- [Browser Support](#browser-support)
- [Integration with Cybernetic Chatbot Backend](#integration-with-cybernetic-chatbot-backend)
- [License](#license)

## Installation

### npm / yarn / pnpm

```bash
npm install @astermind/cybernetic-chatbot-client
```

```bash
yarn add @astermind/cybernetic-chatbot-client
```

```bash
pnpm add @astermind/cybernetic-chatbot-client
```

**Required import** — add to your JavaScript/TypeScript file:

```typescript
import { CyberneticClient } from '@astermind/cybernetic-chatbot-client';
```

> **Note:** Most users should install `@astermind/chatbot-template` instead, which includes this package as a dependency along with pre-built UI components. Use this package directly only if you're building a custom chat UI.

### CDN (Script Tag)

Include **one** of the following script tags in your HTML:

```html
<!-- Core bundle (API client, caching, offline fallback) -->
<script src="https://unpkg.com/@astermind/cybernetic-chatbot-client/dist/cybernetic-chatbot-client.umd.js"></script>

<!-- Or full bundle (core + agentic capabilities) -->
<script src="https://unpkg.com/@astermind/cybernetic-chatbot-client/dist/cybernetic-chatbot-client-full.umd.js"></script>
```

The client is available as `window.AsterMindCybernetic` (core) or `window.AsterMindCyberneticFull` (full bundle).

## Quick Start

> **Note:** No license key is required for development. See [Licensing](#licensing) for production requirements.

### Basic Usage

```typescript
import { CyberneticClient } from '@astermind/cybernetic-chatbot-client';

const client = new CyberneticClient({
  apiUrl: 'https://api.astermind.ai',
  apiKey: 'am_your_api_key',
  // licenseKey: 'your-license-key',  // Optional in development, required in production
  fallback: {
    enabled: true,
    cacheOnConnect: true
  },
  onStatusChange: (status) => {
    console.log('Connection status:', status);
  }
});

// Simple question
const response = await client.ask('What is AsterMind?');
console.log(response.reply);

// With streaming
await client.askStream('Tell me about RAG', {
  onToken: (token) => process.stdout.write(token),
  onSources: (sources) => console.log('Sources:', sources),
  onComplete: (response) => console.log('\nDone:', response.sessionId)
});
```

### Script Tag Integration

```html
<script
  src="https://unpkg.com/@astermind/cybernetic-chatbot-client/dist/cybernetic-chatbot-client.umd.js"
  data-astermind-key="am_your_api_key"
  data-astermind-url="https://api.astermind.ai"
></script>
```

### Global Config Object

```html
<script>
  window.astermindConfig = {
    apiUrl: 'https://api.astermind.ai',
    apiKey: 'am_your_api_key',
    fallback: { enabled: true }
  };
</script>
<script src="https://unpkg.com/@astermind/cybernetic-chatbot-client/dist/cybernetic-chatbot-client.umd.js"></script>
```

## Licensing

**Free for Development** — This package is free to use during development and testing. No license key is required for local development environments.

**License Required for Production** — A valid license key is required for production deployments. Without a license, chatbot responses in production will include a visible license notice. Licenses are available at [https://astermind.ai](https://astermind.ai).

### License Products

| Product | Feature Flag | Included With |
|---------|--------------|---------------|
| **Cybernetic Chatbot Client** | `cybernetic-chatbot-client` | Cybernetic Chatbot purchase |
| **Agentic Add-On** | `agentic` | Separate Agentic Add-On purchase |

- **Cybernetic Chatbot**: Includes the `cybernetic-chatbot-client` feature, enabling all core client functionality (API communication, offline caching, streaming, session management).
- **Agentic Add-On**: Requires a separate purchase. Enables the `agentic` feature for intent classification and DOM automation capabilities.

### Applying Your License Key

Add your license key to the client configuration:

```typescript
import { CyberneticClient } from '@astermind/cybernetic-chatbot-client';

const client = new CyberneticClient({
  apiUrl: 'https://api.astermind.ai',
  apiKey: 'am_your_api_key',
  licenseKey: 'eyJhbGciOiJFUzI1NiIsInR5cCI6IkpXVCJ9...'  // Your license key (JWT)
});
```

For script tag integration:

```html
<script>
  window.astermindConfig = {
    apiUrl: 'https://api.astermind.ai',
    apiKey: 'am_your_api_key',
    licenseKey: 'eyJhbGciOiJFUzI1NiIsInR5cCI6IkpXVCJ9...'
  };
</script>
<script src="https://unpkg.com/@astermind/cybernetic-chatbot-client/dist/cybernetic-chatbot-client.umd.js"></script>
```

### Enforcement Behavior

The license system uses environment-aware enforcement:

| Environment | Detection | License Required? | Behavior |
|-------------|-----------|-------------------|----------|
| **Development** | `localhost`, `127.0.0.1`, `.local`, `.dev`, dev ports (3000, 5173, 8080, etc.) | **No** | Free to use. Console warnings only. |
| **Production** | All other URLs | **Yes** | License warning appended to responses if missing/invalid. |

**Development Mode** (free, soft enforcement):
- **No license key required** — develop and test without any restrictions
- Console warnings are logged when license is missing, expired, or invalid (for awareness)
- Console warnings when using features not included in your license
- All functionality works normally — responses are returned unchanged

**Production Mode** (license required, hard enforcement):
- A valid license key is required for production use
- Without a valid license, chatbot responses include a visible notice:

  *"⚠️ License Notice: Your AsterMind license key needs to be updated. Please contact support@astermind.ai or visit https://astermind.ai/license to renew your license."*

- With a valid license, responses are returned normally without any modifications

### Checking License Status

```typescript
// Get current status including license state
const status = client.getStatus();
console.log(status.license);
// {
//   status: 'valid' | 'invalid' | 'expired' | 'missing' | 'eval',
//   payload: { plan, features, exp, ... },
//   daysRemaining: 30,
//   inGracePeriod: false
// }

// Get license manager for advanced operations
const license = client.getLicenseManager();

// Check if license is valid
license.isValid();

// Check specific features
license.hasFeature('cybernetic-chatbot-client');
license.hasFeature('agentic');

// Get human-readable status
license.getStatusMessage();
// "License valid (30 days remaining)"
```

### Feature Validation

The client automatically validates features:

- **Client feature** (`cybernetic-chatbot-client`): Checked on client initialization
- **Agentic feature** (`agentic`): Checked only when agentic capabilities are used

If a required feature is missing, the console displays:
- Current license plan
- Available features in your license
- The missing feature name
- Link to upgrade at https://astermind.ai/license

### Obtaining a License

1. Visit [https://astermind.ai](https://astermind.ai)
2. Purchase **Cybernetic Chatbot** for core client functionality
3. Optionally purchase the **Agentic Add-On** for DOM automation features
4. Your license key (JWT token) will be provided in your account dashboard
5. Add the license key to your client configuration

For licensing questions, contact support@astermind.ai.

## Configuration

All configuration is done in **your own project**—you never need to modify `node_modules` or the package source code.

### Multi-Method Configuration

The client supports multiple configuration methods with a priority-based fallback chain. This allows you to use the most appropriate method for your hosting environment.

**Priority Order (highest to lowest):**

1. **Constructor config** - Direct configuration passed to `CyberneticClient` or `createClient()`
2. **Environment variables** - `VITE_ASTERMIND_RAG_API_KEY`, `REACT_APP_ASTERMIND_RAG_API_KEY`, etc.
3. **SSR-injected config** - `window.__ASTERMIND_CONFIG__` (for server-side rendering)
4. **Global object** - `window.astermindConfig`
5. **Script data attributes** - `data-astermind-key`, `data-astermind-url`

#### Environment Variables

For bundled applications (Vite, Create React App, etc.), you can configure the client using environment variables:

**Vite:**
```env
VITE_ASTERMIND_RAG_API_KEY=am_your_api_key
VITE_ASTERMIND_RAG_API_SERVER_URL=https://api.astermind.ai
```

**Create React App:**
```env
REACT_APP_ASTERMIND_RAG_API_KEY=am_your_api_key
REACT_APP_ASTERMIND_RAG_API_SERVER_URL=https://api.astermind.ai
```

**Node.js / Server:**
```env
ASTERMIND_RAG_API_KEY=am_your_api_key
ASTERMIND_RAG_API_SERVER_URL=https://api.astermind.ai
```

#### Auto-Loading Configuration

Use `loadConfig()` to automatically detect configuration from available sources:

```typescript
import { loadConfig, createClient } from '@astermind/cybernetic-chatbot-client';

// Auto-detect configuration (throws if no API key found)
const config = loadConfig();
const client = createClient(config);

// Or suppress errors and handle missing config gracefully
const config = loadConfig({ throwOnMissingKey: false });
if (config) {
  const client = createClient(config);
} else {
  console.log('Chatbot not configured');
}
```

#### SSR / Runtime Injection

For server-side rendered applications, inject configuration at runtime:

```html
<!-- In your SSR template -->
<script>
  window.__ASTERMIND_CONFIG__ = {
    apiKey: '<%= process.env.ASTERMIND_RAG_API_KEY %>',
    apiUrl: '<%= process.env.ASTERMIND_RAG_API_SERVER_URL %>'
  };
</script>
```

#### Configuration Source Debugging

The loaded configuration includes a `_source` field for debugging:

```typescript
const config = loadConfig();
console.log(config._source);
// 'env' | 'vite' | 'window' | 'data-attr' | 'props'
```

### Full Configuration Interface

```typescript
interface CyberneticConfig {
  /** Backend API URL (required) */
  apiUrl: string;

  /** API key for authentication - must start with 'am_' (required) */
  apiKey: string;

  /** WebSocket URL for SaaS streaming (auto-derived from apiUrl for known SaaS domains) */
  wsUrl?: string;

  /** Transport mode: 'auto' (default) tries WebSocket then REST, 'websocket' forces WS, 'rest' forces REST+SSE */
  transport?: 'auto' | 'websocket' | 'rest';

  /** WebSocket transport options */
  websocket?: {
    maxReconnectAttempts?: number;  // Default: 3
    reconnectDelay?: number;       // Default: 1000ms (exponential backoff)
    connectionTimeout?: number;    // Default: 10000ms
  };

  /** License key (JWT token) from https://astermind.ai */
  licenseKey?: string;

  /** Fallback/offline configuration */
  fallback?: {
    /** Enable offline fallback (default: true) */
    enabled?: boolean;

    /** Cache max age in milliseconds (default: 86400000 = 24 hours) */
    cacheMaxAge?: number;

    /** Sync documents on connect (default: true) */
    cacheOnConnect?: boolean;

    /** Storage type (default: 'indexeddb') */
    cacheStorage?: 'indexeddb' | 'localstorage';
  };

  /** Retry configuration */
  retry?: {
    /** Max retries before fallback (default: 2) */
    maxRetries?: number;

    /** Initial delay in ms (default: 1000) */
    initialDelay?: number;

    /** Use exponential backoff (default: true) */
    exponentialBackoff?: boolean;
  };

  /** Event callbacks */
  onStatusChange?: (status: ConnectionStatus) => void;
  onError?: (error: CyberneticError) => void;

  /** Agentic capabilities configuration (requires full bundle) */
  agentic?: AgenticConfig;

  /** Offline vector export configuration (see Pre-computed Vector Export section) */
  offline?: OfflineConfig;

  /** Sitemap configuration for agentic navigation (see Sitemap Configuration section) */
  sitemap?: SiteMapConfig;
}

type ConnectionStatus = 'online' | 'offline' | 'connecting' | 'error';
```

### Agentic Configuration

```typescript
interface AgenticConfig {
  /** Enable agentic DOM interactions (default: false) */
  enabled: boolean;

  /** Confidence threshold for action execution (default: 0.8) */
  confidenceThreshold?: number;

  /** Allowed DOM actions */
  allowedActions?: ('click' | 'fill' | 'scroll' | 'navigate' | 'select')[];

  /** Require user confirmation before actions (default: true) */
  requireConfirmation?: boolean;

  /** Maximum actions per conversation turn (default: 5) */
  maxActionsPerTurn?: number;

  /** CSS selectors to never interact with */
  blockedSelectors?: string[];

  /** Only allow actions within these selectors */
  allowedSelectors?: string[];
}
```

## Features

### WebSocket Transport (SaaS)

When connecting to AsterMind's SaaS infrastructure, the client automatically uses WebSocket transport for streaming chat. This provides lower latency and access to the full RAG pipeline (RSF temporal scoring, hybrid reranking, Omega embeddings, BYOLLM).

**Auto-detection (zero config for SaaS users):**

```typescript
const client = new CyberneticClient({
  apiUrl: 'https://chatapi.astermind.ai',
  apiKey: 'am_your_api_key',
  // wsUrl auto-derived → wss://chatws.astermind.ai
});
```

The client auto-derives WebSocket URLs for known SaaS domains:
- `chatapi.astermind.ai` → `wss://chatws.astermind.ai`
- `chatapi-dev.astermind.ai` → `wss://chatws-dev.astermind.ai`
- `api.astermind.ai` → `wss://chatws.astermind.ai`

**Explicit configuration:**

```typescript
const client = new CyberneticClient({
  apiUrl: 'https://chatapi.astermind.ai',
  apiKey: 'am_your_api_key',
  wsUrl: 'wss://chatws.astermind.ai',
  transport: 'websocket',  // Force WebSocket only (no REST fallback)
});
```

**On-prem (unchanged, REST+SSE):**

```typescript
const client = new CyberneticClient({
  apiUrl: 'http://localhost:3000',
  apiKey: 'am_your_api_key',
  // No wsUrl → uses REST+SSE as before
});
```

**Transport modes:**
- `'auto'` (default) — Uses WebSocket when available, falls back to REST+SSE on failure
- `'websocket'` — Forces WebSocket only, no REST fallback
- `'rest'` — Forces REST+SSE only, ignores WebSocket

**Environment variable support:**

```env
# Vite
VITE_ASTERMIND_RAG_WS_URL=wss://chatws.astermind.ai

# Node.js / CRA
ASTERMIND_RAG_WS_URL=wss://chatws.astermind.ai
REACT_APP_ASTERMIND_RAG_WS_URL=wss://chatws.astermind.ai
```

**Cleanup:**

Call `destroy()` when disposing the client to close the WebSocket connection:

```typescript
client.destroy();
```

### Offline-First Architecture

The client includes **built-in offline fallback** with IndexedDB caching and TF-IDF local search—no additional setup required. When the server is unreachable, the client automatically serves cached responses:

```typescript
const client = new CyberneticClient({
  apiUrl: 'https://api.astermind.ai',
  apiKey: 'am_your_api_key',
  fallback: {
    enabled: true,
    cacheOnConnect: true,    // Cache responses for offline use
    cacheMaxAge: 86400000,   // 24 hours
    cacheStorage: 'indexeddb'
  }
});

// Check connection status
const status = client.getStatus();
console.log(status.connection); // 'online' | 'offline' | 'connecting'
console.log(status.cache);      // { documentCount, lastSyncAt, cacheSize, isStale }

// Response includes offline indicator
const response = await client.ask('cached question');
if (response.offline) {
  console.log('Response from local cache');
  console.log('Confidence:', response.confidence); // 'medium' or 'low' when offline
}

// Manually sync cache
await client.syncCache();

// Clear cache
await client.clearCache();
```

**Cache Validation**: The server controls cache retention via `cacheRetentionHours` (default: 168 hours / 7 days). The client respects this setting and marks responses as stale when appropriate.

### Pre-computed Vector Export (Advanced)

For enhanced offline performance, the client supports loading pre-computed TF-IDF vectors exported from the AsterMind admin panel. This eliminates client-side vector computation and provides faster, more consistent offline search results.

#### Exporting Vectors from Admin

1. Navigate to your AsterMind admin panel
2. Go to **Settings > Vector Export** (or **Documents > Export**)
3. Click **Export Vectors for Offline Use**
4. Download the JSON export file or note the export URL

The export file contains pre-computed TF-IDF vectors, document metadata, and optionally sitemap and category information for agentic navigation.

#### Configuring Offline Vectors

```typescript
import { CyberneticClient } from '@astermind/cybernetic-chatbot-client';

const client = new CyberneticClient({
  apiUrl: 'https://api.astermind.ai',
  apiKey: 'am_your_api_key',

  // Offline vector configuration
  offline: {
    enabled: true,
    vectorFileUrl: 'https://your-cdn.com/vectors/export.json',  // URL to exported vectors
    storageMode: 'indexeddb',  // 'memory' | 'indexeddb' | 'hybrid'
    maxCacheAge: 604800000,    // 7 days in milliseconds
    autoRefresh: true,         // Auto-refresh when new export available

    // Optional: Omega advanced RAG (requires @astermind/astermind-community)
    omega: {
      enabled: true,
      modelUrl: 'https://your-cdn.com/models/omega-model.json'
    }
  }
});
```

#### Inline Vector Data

You can also provide vector data directly in the configuration:

```typescript
const client = new CyberneticClient({
  apiUrl: 'https://api.astermind.ai',
  apiKey: 'am_your_api_key',
  offline: {
    enabled: true,
    vectorData: exportedVectorObject,  // Loaded from file or bundled
    storageMode: 'memory'
  }
});
```

#### Offline Configuration Options

```typescript
interface OfflineConfig {
  /** Enable offline vector support */
  enabled: boolean;

  /** URL to fetch vector export JSON */
  vectorFileUrl?: string;

  /** Inline vector data (alternative to URL) */
  vectorData?: OfflineVectorExport;

  /** Storage mode for cached vectors */
  storageMode?: 'memory' | 'indexeddb' | 'hybrid';

  /** Maximum cache age in milliseconds (default: 7 days) */
  maxCacheAge?: number;

  /** Auto-refresh vectors when new export available */
  autoRefresh?: boolean;

  /** Omega advanced RAG configuration */
  omega?: {
    enabled: boolean;
    modelUrl?: string;
    modelData?: SerializedModel;
    config?: {
      topK?: number;
      rerankerTopK?: number;
      minScore?: number;
    };
  };
}
```

#### Checking Offline Status

```typescript
// Get local RAG status
const ragStatus = client.getLocalRAGStatus();
console.log(ragStatus);
// {
//   loaded: true,
//   loadedFromExport: true,
//   documentCount: 150,
//   chunkCount: 1200,
//   exportVersion: '1.0.0',
//   exportedAt: '2024-01-15T10:30:00Z'
// }

// Check if Omega is enabled and ready
if (client.isOmegaOfflineEnabled()) {
  const modelInfo = client.getOfflineModelInfo();
  console.log('Omega model:', modelInfo);
}

// Force reload vectors
await client.reloadOfflineVectors();
```

#### Console Warning

When `offline.enabled` is `true` but no vectors are loaded (missing URL, network error, or invalid data), the client logs a one-time console warning:

```
[CyberneticClient] Warning: Offline mode enabled but no vectors loaded.
Configure 'offline.vectorFileUrl' or provide 'offline.vectorData' for offline support.
Falling back to standard caching mode.
```

This helps identify configuration issues without disrupting functionality.

### Streaming Responses

Real-time token streaming via WebSocket (SaaS) or Server-Sent Events (on-prem):

```typescript
await client.askStream('Explain quantum computing', {
  onToken: (token) => {
    // Called for each token as it arrives
    document.getElementById('output').textContent += token;
  },
  onSources: (sources) => {
    // Called when sources are available
    console.log('Sources:', sources);
  },
  onComplete: (response) => {
    // Called when streaming is complete
    console.log('Session ID:', response.sessionId);
  },
  onError: (error) => {
    // Called on error
    console.error('Error:', error.message);
  }
});
```

### Session Management

Maintain conversation context across multiple turns:

```typescript
// First message establishes session
const response1 = await client.ask('Hello!');
const sessionId = response1.sessionId;

// Continue conversation with session ID
const response2 = await client.ask('Tell me more', { sessionId });
const response3 = await client.ask('Can you clarify?', { sessionId });

// Optionally pass page context
const response = await client.ask('Help me with this page', {
  sessionId,
  context: {
    currentPage: '/products/widget',
    pageTitle: 'Widget Product Page'
  }
});
```

### Agentic Capabilities

The full bundle includes intent classification and DOM automation:

```typescript
import {
  CyberneticClient,
  CyberneticAgent,
  CyberneticIntentClassifier
} from '@astermind/cybernetic-chatbot-client/full';

const client = new CyberneticClient({
  apiUrl: 'https://api.astermind.ai',
  apiKey: 'am_your_api_key',
  agentic: {
    enabled: true,
    confidenceThreshold: 0.8,
    requireConfirmation: true,
    allowedActions: ['click', 'fill', 'navigate', 'scroll'],
    blockedSelectors: ['.admin-panel', '#dangerous-button'],
    maxActionsPerTurn: 5
  }
});

// Smart ask - checks for action intent first, then falls back to RAG
const result = await client.smartAsk('Take me to the settings page');

if (result.action) {
  // Action detected
  console.log('Action:', result.action.type, result.action.target);
  console.log('Confidence:', result.action.confidence);

  // If requireConfirmation is true, action is returned but not executed
  // Your UI can show a confirmation dialog, then execute:
  if (userConfirmed) {
    const actionResult = await client.executeAction(result.action);
    console.log('Result:', actionResult.message);
  }
} else if (result.response) {
  // Standard RAG response
  console.log('Reply:', result.response.reply);
}
```

#### Supported Action Types

| Action | Description | Example Phrases |
|--------|-------------|-----------------|
| `navigate` | Navigate to URL/route | "go to settings", "take me to dashboard" |
| `fillForm` | Fill form input fields | "search for products", "enter my email" |
| `clickElement` | Click buttons/links | "click submit", "press the save button" |
| `scroll` | Scroll to element/position | "scroll to top", "jump to pricing section" |
| `highlight` | Highlight elements | "show me the login button" |
| `triggerModal` | Open modal dialogs | "open help modal", "show settings dialog" |
| `custom` | Custom action handlers | "export data", "refresh dashboard" |

#### Intent Classification

The classifier uses a hybrid approach with regex patterns and Jaccard similarity for fuzzy matching:

```typescript
import { CyberneticIntentClassifier } from '@astermind/cybernetic-chatbot-client/full';

const classifier = new CyberneticIntentClassifier({
  enabled: true,
  confidenceThreshold: 0.8,
  siteMap: [
    { path: '/settings', name: 'Settings', aliases: ['preferences', 'config'] },
    { path: '/dashboard', name: 'Dashboard', aliases: ['home', 'main'] }
  ]
});

const intent = classifier.classify('take me to the settings page');
// { action: { type: 'navigate', target: '/settings', confidence: 0.92 }, ... }
```

#### Security Features

- **Selector Sanitization**: Removes potentially dangerous characters from CSS selectors
- **URL Validation**: Blocks `javascript:` and `data:` URLs
- **Blocked Selectors**: Configure selectors that should never be interacted with
- **Allowed Selectors**: Optionally whitelist specific selectors
- **Rate Limiting**: Maximum actions per minute (default: 5)
- **Confirmation Flow**: Optional user approval before action execution

### Sitemap Configuration

The sitemap enables intelligent navigation by mapping user intent to application routes. You can configure it statically or load it from the vector export:

#### Static Sitemap Configuration

```typescript
const client = new CyberneticClient({
  apiUrl: 'https://api.astermind.ai',
  apiKey: 'am_your_api_key',

  // Static sitemap configuration
  sitemap: {
    enabled: true,
    entries: [
      {
        path: '/dashboard',
        name: 'Dashboard',
        description: 'Main dashboard with analytics',
        aliases: ['home', 'main', 'overview'],
        keywords: ['stats', 'metrics', 'analytics']
      },
      {
        path: '/settings',
        name: 'Settings',
        description: 'User and application settings',
        aliases: ['preferences', 'config', 'options'],
        keywords: ['account', 'profile', 'configuration']
      },
      {
        path: '/products',
        name: 'Products',
        description: 'Product catalog and management',
        aliases: ['catalog', 'inventory'],
        keywords: ['items', 'shop', 'store']
      }
    ]
  }
});
```

#### Loading Sitemap from Vector Export

When using pre-computed vectors, the sitemap can be included in the export and loaded automatically:

```typescript
const client = new CyberneticClient({
  apiUrl: 'https://api.astermind.ai',
  apiKey: 'am_your_api_key',
  offline: {
    enabled: true,
    vectorFileUrl: 'https://your-cdn.com/vectors/export.json'
  },
  sitemap: {
    enabled: true,
    loadFromExport: true  // Load sitemap from vector export
  }
});
```

#### Sitemap Configuration Options

```typescript
interface SiteMapConfig {
  /** Enable sitemap-based navigation */
  enabled: boolean;

  /** Static sitemap entries */
  entries?: SiteMapEntry[];

  /** Load sitemap from vector export (requires offline.enabled) */
  loadFromExport?: boolean;

  /** URL to fetch sitemap JSON separately */
  sitemapUrl?: string;
}

interface SiteMapEntry {
  /** Route path (e.g., '/dashboard') */
  path: string;

  /** Display name */
  name: string;

  /** Description for context matching */
  description?: string;

  /** Alternative names/phrases */
  aliases?: string[];

  /** Related keywords for matching */
  keywords?: string[];

  /** Authentication required */
  requiresAuth?: boolean;

  /** Required user roles */
  roles?: string[];

  /** Child routes */
  children?: SiteMapEntry[];
}
```

### Maintenance Mode Support

The client handles backend maintenance mode gracefully (per ADR-200):

```typescript
// Check if maintenance mode is active
if (client.isMaintenanceMode()) {
  const message = client.getMaintenanceMessage();
  console.log('Maintenance:', message);
}

// Get full system status
const status = client.getStatus();
console.log(status.systemSettings);
// {
//   maintenanceMode: boolean,
//   maintenanceMessage?: string,
//   cacheRetentionHours: number,
//   forceOfflineClients: boolean
// }
```

When maintenance mode is active:
- The client automatically uses cached data
- New requests are served from local RAG
- `response.offline` will be `true`
- `response.degradedReason` will indicate maintenance mode

## Bundle Options

| Entry Point | Import Path | Description | Size (minified) |
|-------------|-------------|-------------|-----------------|
| Core | `@astermind/cybernetic-chatbot-client` | Client, caching, local RAG | ~15KB |
| Full | `@astermind/cybernetic-chatbot-client/full` | Core + agentic capabilities | ~25KB |

### Tree-Shakeable Imports

```typescript
// Import only core client (smaller bundle)
import { CyberneticClient } from '@astermind/cybernetic-chatbot-client';

// Import agentic features when needed
import {
  CyberneticClient,
  CyberneticAgent,
  CyberneticIntentClassifier
} from '@astermind/cybernetic-chatbot-client/full';
```

## API Reference

### CyberneticClient

```typescript
class CyberneticClient {
  constructor(config: CyberneticConfig);

  // Core methods
  ask(message: string, options?: AskOptions): Promise<CyberneticResponse>;
  askStream(message: string, callbacks: StreamCallbacks, options?: AskOptions): Promise<void>;

  // Agentic methods (requires agentic config and license)
  smartAsk(message: string, options?: AskOptions): Promise<SmartAskResult>;
  classifyIntent(message: string): IntentClassification | null;
  executeAction(action: AgentAction): Promise<ActionResult>;
  isAgenticEnabled(): boolean;

  // Status methods (includes license state)
  getStatus(): { connection: ConnectionStatus; cache: CacheStatus; lastError: CyberneticError | null; systemSettings: SystemSettings | null; license: LicenseState | null };
  checkConnection(): Promise<boolean>;
  checkSystemStatus(): Promise<SystemSettings>;
  isMaintenanceMode(): boolean;
  getMaintenanceMessage(): string | undefined;
  isCacheValid(): boolean;

  // License methods
  getLicenseManager(): LicenseManager;

  // Cache methods
  syncCache(): Promise<void>;
  clearCache(): Promise<void>;

  // Cleanup
  destroy(): void;  // Close WebSocket connection and clean up resources
}
```

### Response Types

```typescript
interface CyberneticResponse {
  reply: string;
  confidence: 'high' | 'medium' | 'low' | 'none';
  sources: Source[];
  offline: boolean;
  sessionId?: string;
  retryAfter?: number;
  degradedReason?: string;
}

interface Source {
  title: string;
  snippet: string;
  relevance: number;
  documentId?: string;
}

interface StreamCallbacks {
  onToken?: (token: string) => void;
  onSources?: (sources: Source[]) => void;
  onComplete?: (response: CyberneticResponse) => void;
  onError?: (error: CyberneticError) => void;
}

interface CyberneticError {
  code: 'NETWORK_ERROR' | 'AUTH_ERROR' | 'RATE_LIMIT' | 'SERVER_ERROR' | 'CACHE_ERROR' | 'LOCAL_RAG_ERROR' | 'WS_ERROR';
  message: string;
  retryAfter?: number;
}

interface LicenseState {
  status: 'valid' | 'invalid' | 'expired' | 'missing' | 'eval';
  payload: LicensePayload | null;
  error?: string;
  inGracePeriod: boolean;
  daysRemaining: number | null;
}

interface LicensePayload {
  iss: string;          // Issuer
  sub: string;          // Subject (license ID)
  aud: string;          // Audience (product)
  iat: number;          // Issued at
  exp: number;          // Expiration
  plan: 'free' | 'pro' | 'business' | 'enterprise' | 'eval';
  org?: string;         // Organization
  seats: number;        // Number of seats
  features: string[];   // Enabled features
  graceUntil?: number;  // Grace period end
  licenseVersion: number;
}
```

## Browser Support

| Browser | Version | Notes |
|---------|---------|-------|
| Chrome | 80+ | Full support |
| Firefox | 75+ | Full support |
| Safari | 13.1+ | Full support |
| Edge | 80+ | Full support |

Requires IndexedDB support for offline caching.

## Integration with Cybernetic Chatbot Backend

This client is designed to work with the AsterMind Cybernetic Chatbot backend. The client supports two transport modes:

**REST API Endpoints** (on-prem and supplementary):

| Endpoint | Method | Description |
|----------|--------|-------------|
| `/api/external/chat` | POST | Send message, get complete response |
| `/api/external/chat/stream` | POST | Send message, get SSE streaming response |
| `/api/external/docs` | GET | Fetch documents for offline caching |
| `/api/external/status` | GET | Check API status, quota, and system settings |
| `/api/external/health` | GET | Health check (no auth required) |
| `/api/external/search` | GET | Search documents |
| `/api/external/sitemap` | GET | Get document sitemap for navigation |
| `/api/external/config` | GET | Get chatbot configuration |

**WebSocket Endpoint** (SaaS streaming):

| Endpoint | Description |
|----------|-------------|
| `wss://chatws.astermind.ai` | Production WebSocket streaming |
| `wss://chatws-dev.astermind.ai` | Development WebSocket streaming |

WebSocket authentication uses an API key query parameter: `?apiKey=am_xxx`

**Authentication**: REST endpoints require an `X-API-Key` header with a valid API key (prefixed with `am_`). WebSocket endpoints use a `?apiKey=am_xxx` query parameter.

**Rate Limiting**: The backend enforces rate limits. The client handles 429 responses gracefully and includes `retryAfter` in responses when applicable.

## Links

- [AsterMind Website](https://astermind.ai)

## License

MIT License - see [LICENSE](LICENSE) for details.