Skip to main content

Conversational Search

The widget supports a conversational search mode that maintains context between successive searches, allowing natural refinement of results.

Activation​

Conversational search requires enabling the AI assistant:

NeuroonWidget.init({
  container: '#neuroon-search',
  token: 'YOUR_TOKEN',
  features: {
    aiAssistant: true, // Enables conversational mode
  },
});
Plan required

The AI assistant is available on Growth plans and above.

How it works​

Session context​

The widget maintains a conversation history during the session:

User: "running shoes"
→ Results: athletic shoes

User: "Nike only"
→ Context: shoes + Nike
→ Results: Nike running shoes

User: "under $100"
→ Context: shoes + Nike + price < $100
→ Refined results

Conversation turns​

Each interaction generates a turn with user and assistant information:

interface ConversationTurn {
  id: string;
  role: 'user' | 'assistant';
  content: string;
  timestamp: Date;
  products?: Product[];          // Suggested products
  suggestedFilters?: SearchFilters; // Inferred filters
  confidence?: number;           // 0-1, interpretation confidence
}

Chat Interface​

Open chat panel​

widget.openChat();

Or via the chat icon in the widget interface.

Send messages​

// Programmatically
await widget.sendMessage('I need a lightweight laptop for work');

// Users can also type directly in chat

Conversation history​

// Get full history
const history = widget.conversationHistory;

history.forEach(turn => {
  console.log(`[${turn.role}] ${turn.content}`);
  if (turn.products?.length) {
    console.log(`  → ${turn.products.length} suggested products`);
  }
});

Clear history​

// Reset conversation
widget.clearHistory();

Session Persistence​

History is persisted in sessionStorage during the browser session:

// Automatically saved with key:
// neuroon_chat_{shopId}

// History is restored on page reload
// Cleared when browser closes

Configure persistence​

NeuroonWidget.init({
  features: {
    aiAssistant: true,
  },
  // Disable persistence if preferred
  persistence: {
    chat: false, // Default: true
  },
});

Inferred Filters​

The assistant analyzes queries and extracts filters automatically:

QueryInferred filters
"cheap Nike shoes"brands: ['Nike'], priceMax: inferred
"gaming laptop"categories: ['gaming'], tags: ['gaming']
"red dress size M"tags: ['red'], size: 'M'
"under 50 dollars"priceMax: 50
// Listen for inferred filters
widget.on('chat:message', ({ turn }) => {
  if (turn.suggestedFilters) {
    console.log('Suggested filters:', turn.suggestedFilters);
    // { brands: ['Nike'], priceMax: 100 }
  }
});

Confidence Indicator​

Each response includes a confidence level:

widget.on('chat:message', ({ turn }) => {
  if (turn.confidence !== undefined) {
    if (turn.confidence > 0.8) {
      // High confidence - apply filters automatically
      widget.setFilters(turn.suggestedFilters);
    } else {
      // Low confidence - show options to user
      showFilterOptions(turn.suggestedFilters);
    }
  }
});
RangeMeaning
0.9 - 1.0Very high - Clear interpretation
0.7 - 0.9High - Good inference
0.5 - 0.7Medium - May need confirmation
0.0 - 0.5Low - Ambiguous query

Chat Events​

EventPayloadDescription
chat:open{}Chat panel opened
chat:close{}Chat panel closed
chat:message{ turn }New message (user or assistant)
chat:typing{ isTyping }Typing indicator
chat:error{ error }Chat error
widget.on('chat:typing', ({ isTyping }) => {
  if (isTyping) {
    showTypingIndicator();
  } else {
    hideTypingIndicator();
  }
});

Best Practices​

  1. Limited context: History is limited to the last 10 interactions for relevance
  2. Periodic cleanup: Consider clearing history when changing sections or categories
  3. Fallback: If assistant doesn't understand, offer manual filter options
  4. Visual indicators: Show when assistant is "thinking"
  5. Optional persistence: Disable if context shouldn't persist between reloads

Next steps​