AI-Powered Features

Algolia’s AI-powered features bring conversational search and intelligent filter suggestions to your application. Use the Chat and Filter Suggestions connectors to create modern, AI-enhanced search experiences.

Overview

InstantSearch provides two AI-powered connectors:

Chat: Conversational search interface powered by AI agents
Filter Suggestions: AI-generated filter recommendations based on search context

Both features use Algolia’s Agent Studio or custom AI endpoints.

Chat

Create a conversational search experience where users can ask questions in natural language.

Basic Setup

import { connectChat } from 'instantsearch.js/es/connectors';

const renderChat = (renderOptions, isFirstRender) => {
  const {
    messages,
    status,
    sendMessage,
    input,
    setInput,
  } = renderOptions;
  
  if (isFirstRender) {
    const container = document.querySelector('#chat');
    
    container.innerHTML = `
      <div id="messages"></div>
      <div class="input-area">
        <input id="chat-input" type="text" placeholder="Ask a question..." />
        <button id="send-btn">Send</button>
      </div>
    `;
    
    document.querySelector('#send-btn').addEventListener('click', () => {
      const input = document.querySelector('#chat-input');
      sendMessage(input.value);
      input.value = '';
    });
  }
  
  // Render messages
  const messagesContainer = document.querySelector('#messages');
  messagesContainer.innerHTML = messages.map(msg => `
    <div class="message ${msg.role}">
      ${msg.parts.map(part => 
        part.type === 'text' ? `<p>${part.text}</p>` : ''
      ).join('')}
    </div>
  `).join('');
  
  // Show loading state
  if (status === 'loading') {
    messagesContainer.innerHTML += '<div class="loading">Thinking...</div>';
  }
};

const chatWidget = connectChat(renderChat);

search.addWidgets([
  chatWidget({
    agentId: 'your-agent-id', // From Algolia Agent Studio
  }),
]);

Chat with Custom Transport

Use a custom AI endpoint:

const chatWidget = connectChat(renderChat);

search.addWidgets([
  chatWidget({
    transport: {
      api: 'https://your-ai-api.com/chat',
      headers: {
        'Authorization': 'Bearer YOUR_TOKEN',
      },
      prepareSendMessagesRequest({ messages, trigger }) {
        return {
          body: {
            messages,
            stream: true,
          },
        };
      },
    },
  }),
]);

Chat State Management

From src/connectors/chat/connectChat.ts, the chat connector provides:

type ChatRenderState = {
  // Current chat messages
  messages: UIMessage[];
  
  // Chat status: 'idle' | 'loading' | 'streaming' | 'error'
  status: string;
  
  // Send a message
  sendMessage: (message: string) => void;
  
  // Regenerate last response
  regenerate: () => void;
  
  // Stop ongoing generation
  stop: () => void;
  
  // Clear error state
  clearError: () => void;
  
  // Error if any
  error: Error | null;
  
  // Current input value
  input: string;
  
  // Set input value
  setInput: (input: string) => void;
  
  // Current search state
  indexUiState: IndexUiState;
  
  // Set search state
  setIndexUiState: (state: IndexUiState) => void;
};

Tool Integration

Implement custom tools for the AI to use:

const chatWidget = connectChat(renderChat);

search.addWidgets([
  chatWidget({
    agentId: 'your-agent-id',
    tools: {
      search_index: {
        onToolCall({ toolCall, addToolResult }) {
          // AI wants to search the index
          const { query, facetFilters } = toolCall.args;
          
          // Perform search
          performSearch(query, facetFilters).then(results => {
            addToolResult({
              output: JSON.stringify(results.hits.slice(0, 5)),
            });
          });
        },
      },
      
      get_product_details: {
        onToolCall({ toolCall, addToolResult }) {
          const { productId } = toolCall.args;
          
          fetch(`/api/products/${productId}`).then(res => res.json()).then(product => {
            addToolResult({
              output: JSON.stringify(product),
            });
          });
        },
      },
    },
  }),
]);

Apply Filters from Chat

The connector provides methods to apply filters from AI suggestions:

// From src/connectors/chat/connectChat.ts
function updateStateFromSearchToolInput(
  params: { query?: string; facetFilters?: string[][] },
  helper: AlgoliaSearchHelper
) {
  // Clear existing filters
  const attributesToClear = getAttributesToClear({ results: helper.lastResults!, helper });
  helper.setState(clearRefinements({ helper, attributesToClear }));

  // Apply new facet filters
  if (params.facetFilters) {
    const attributes = flat(params.facetFilters).map((filter) => {
      const [attribute, value] = filter.split(':');
      return { attribute, value };
    });

    attributes.forEach(({ attribute, value }) => {
      if (!helper.state.isDisjunctiveFacet(attribute)) {
        const s = helper.state.addDisjunctiveFacet(attribute);
        helper.setState(s);
      }
      helper.toggleFacetRefinement(attribute, value);
    });
  }

  // Set query
  if (params.query) {
    helper.setQuery(params.query);
  }

  helper.search();
}

Chat with Suggestions

Extract and display suggestions from AI responses:

const renderChat = (renderOptions, isFirstRender) => {
  const { messages, suggestions, sendMessage } = renderOptions;
  
  // Render messages...
  
  // Render suggestions
  if (suggestions && suggestions.length > 0) {
    const suggestionsHtml = `
      <div class="suggestions">
        <p>You might also ask:</p>
        ${suggestions.map(suggestion => `
          <button onClick="${() => sendMessage(suggestion)}">
            ${suggestion}
          </button>
        `).join('')}
      </div>
    `;
    
    container.insertAdjacentHTML('beforeend', suggestionsHtml);
  }
};

Filter Suggestions

AI-powered filter recommendations that help users refine their search.

Basic Setup

import { connectFilterSuggestions } from 'instantsearch.js/es/connectors';

const renderFilterSuggestions = (renderOptions, isFirstRender) => {
  const { suggestions, isLoading, refine } = renderOptions;
  const container = document.querySelector('#filter-suggestions');
  
  if (isLoading) {
    container.innerHTML = '<div class="loading">Loading suggestions...</div>';
    return;
  }
  
  if (!suggestions.length) {
    container.innerHTML = '';
    return;
  }
  
  container.innerHTML = `
    <div class="filter-suggestions">
      <h4>Suggested Filters</h4>
      <div class="suggestions-list">
        ${suggestions.map(({ attribute, value, label, count }) => `
          <button
            class="suggestion-chip"
            onClick="${() => refine(attribute, value)}"
          >
            ${label}
            <span class="count">${count}</span>
          </button>
        `).join('')}
      </div>
    </div>
  `;
};

const filterSuggestionsWidget = connectFilterSuggestions(renderFilterSuggestions);

search.addWidgets([
  filterSuggestionsWidget({
    agentId: 'your-agent-id',
    maxSuggestions: 3,
    attributes: ['brand', 'category', 'color'], // Optional: limit to specific attributes
  }),
]);

How Filter Suggestions Work

From src/connectors/filter-suggestions/connectFilterSuggestions.ts:

Debounced requests - Waits for search state to stabilize
Context-aware - Sends query, hits sample, and current refinements
Minimum skeleton duration - Shows loading for at least 300ms to avoid flashing

const fetchSuggestions = (results: SearchResults, renderOptions: RenderOptions) => {
  if (!results?.hits?.length) {
    suggestions = [];
    isLoading = false;
    return;
  }

  const loadingStartTime = Date.now();
  isLoading = true;
  
  // Prepare request payload
  const messageText = JSON.stringify({
    query: results.query,
    facets: facetsToSend,
    hitsSample: results.hits.slice(0, hitsToSample),
    currentRefinements,
    maxSuggestions,
  });

  fetch(endpoint, {
    method: 'POST',
    headers: { ...headers, 'Content-Type': 'application/json' },
    body: JSON.stringify({
      messages: [{
        id: `sr-${Date.now()}`,
        createdAt: new Date().toISOString(),
        role: 'user',
        parts: [{ type: 'text', text: messageText }],
      }],
    }),
  })
  .then(response => response.json())
  .then(data => {
    const parsedSuggestions = JSON.parse(data.parts[1].text);
    suggestions = parsedSuggestions.slice(0, maxSuggestions);
  })
  .finally(() => {
    // Ensure minimum skeleton duration
    const elapsed = Date.now() - loadingStartTime;
    const remainingDelay = Math.max(0, MIN_SKELETON_DURATION_MS - elapsed);
    
    setTimeout(() => {
      isLoading = false;
      renderFn(getWidgetRenderState(renderOptions), false);
    }, remainingDelay);
  });
};

Custom Transport

Use your own AI endpoint:

const filterSuggestionsWidget = connectFilterSuggestions(renderFilterSuggestions);

search.addWidgets([
  filterSuggestionsWidget({
    transport: {
      api: 'https://your-ai-api.com/suggestions',
      headers: {
        'Authorization': 'Bearer YOUR_TOKEN',
      },
      prepareSendMessagesRequest(body) {
        return {
          body: {
            ...body,
            model: 'gpt-4',
          },
        };
      },
    },
    maxSuggestions: 5,
    debounceMs: 500,
  }),
]);

Debouncing

Control how quickly suggestions are fetched:

filterSuggestionsWidget({
  agentId: 'your-agent-id',
  debounceMs: 500, // Wait 500ms after search state changes
  hitsToSample: 10, // Send 10 hits for context
})

Transform Suggestions

Filter or modify suggestions before rendering:

filterSuggestionsWidget({
  agentId: 'your-agent-id',
  transformItems(suggestions) {
    return suggestions
      .filter(s => s.count > 0) // Only show suggestions with results
      .map(s => ({
        ...s,
        label: s.label.toUpperCase(), // Transform label
      }));
  },
})

Complete AI Search Example

Combining chat and filter suggestions:

import instantsearch from 'instantsearch.js';
import {
  searchBox,
  hits,
  refinementList,
} from 'instantsearch.js/es/widgets';
import {
  connectChat,
  connectFilterSuggestions,
} from 'instantsearch.js/es/connectors';

const search = instantsearch({
  indexName: 'products',
  searchClient,
  routing: true,
});

// Chat interface
const renderChat = (renderOptions, isFirstRender) => {
  const {
    messages,
    status,
    error,
    sendMessage,
    regenerate,
    stop,
    clearError,
    suggestions,
  } = renderOptions;
  
  const container = document.querySelector('#chat');
  
  if (isFirstRender) {
    container.innerHTML = `
      <div id="chat-messages"></div>
      <div id="chat-suggestions"></div>
      <div class="chat-input">
        <input id="chat-input" placeholder="Ask about products..." />
        <button id="send-btn">Send</button>
      </div>
    `;
    
    const input = document.querySelector('#chat-input');
    const sendBtn = document.querySelector('#send-btn');
    
    sendBtn.addEventListener('click', () => {
      if (input.value.trim()) {
        sendMessage(input.value);
        input.value = '';
      }
    });
    
    input.addEventListener('keypress', (e) => {
      if (e.key === 'Enter' && input.value.trim()) {
        sendMessage(input.value);
        input.value = '';
      }
    });
  }
  
  // Render messages
  const messagesContainer = document.querySelector('#chat-messages');
  messagesContainer.innerHTML = messages.map((msg, index) => {
    const isLast = index === messages.length - 1;
    const parts = msg.parts || [];
    
    return `
      <div class="message ${msg.role}">
        ${parts.map(part => {
          if (part.type === 'text') {
            return `<p>${part.text}</p>`;
          }
          return '';
        }).join('')}
        
        ${isLast && msg.role === 'assistant' ? `
          <div class="message-actions">
            <button onClick="${regenerate}">Regenerate</button>
          </div>
        ` : ''}
      </div>
    `;
  }).join('');
  
  // Show loading/error states
  if (status === 'loading' || status === 'streaming') {
    messagesContainer.innerHTML += `
      <div class="message assistant loading">
        <div class="typing-indicator">Thinking...</div>
        <button onClick="${stop}">Stop</button>
      </div>
    `;
  }
  
  if (error) {
    messagesContainer.innerHTML += `
      <div class="error">
        <p>Error: ${error.message}</p>
        <button onClick="${clearError}">Dismiss</button>
      </div>
    `;
  }
  
  // Render suggestions
  const suggestionsContainer = document.querySelector('#chat-suggestions');
  if (suggestions && suggestions.length > 0) {
    suggestionsContainer.innerHTML = `
      <div class="suggestions">
        <p>You might also ask:</p>
        ${suggestions.map(suggestion => `
          <button onClick="${() => {
            sendMessage(suggestion);
          }}">
            ${suggestion}
          </button>
        `).join('')}
      </div>
    `;
  } else {
    suggestionsContainer.innerHTML = '';
  }
};

// Filter suggestions
const renderFilterSuggestions = (renderOptions) => {
  const { suggestions, isLoading, refine } = renderOptions;
  const container = document.querySelector('#filter-suggestions');
  
  if (isLoading) {
    container.innerHTML = `
      <div class="filter-suggestions loading">
        <div class="skeleton"></div>
        <div class="skeleton"></div>
        <div class="skeleton"></div>
      </div>
    `;
    return;
  }
  
  if (!suggestions.length) {
    container.innerHTML = '';
    return;
  }
  
  container.innerHTML = `
    <div class="filter-suggestions">
      <h4>Suggested Filters</h4>
      ${suggestions.map(({ attribute, value, label, count }) => `
        <button
          class="suggestion-chip"
          onClick="${() => refine(attribute, value)}"
        >
          <span class="label">${label}</span>
          <span class="count">${count.toLocaleString()}</span>
        </button>
      `).join('')}
    </div>
  `;
};

// Initialize widgets
const chatWidget = connectChat(renderChat);
const filterSuggestionsWidget = connectFilterSuggestions(renderFilterSuggestions);

search.addWidgets([
  searchBox({ container: '#searchbox' }),
  
  // AI Chat
  chatWidget({
    agentId: 'your-agent-id',
    tools: {
      search_index: {
        onToolCall({ toolCall, addToolResult, applyFilters }) {
          // Apply filters from AI
          applyFilters(toolCall.args);
          
          // Return results to AI
          addToolResult({
            output: 'Filters applied',
          });
        },
      },
    },
  }),
  
  // Filter Suggestions
  filterSuggestionsWidget({
    agentId: 'your-agent-id',
    maxSuggestions: 3,
    debounceMs: 400,
    attributes: ['brand', 'category', 'color', 'size'],
    hitsToSample: 5,
  }),
  
  // Standard widgets
  hits({
    container: '#hits',
    templates: {
      item: (hit, { html, components }) => html`
        <article>
          <img src="${hit.image}" alt="${hit.name}" />
          <h3>${components.Highlight({ hit, attribute: 'name' })}</h3>
          <p>$${hit.price}</p>
        </article>
      `,
    },
  }),
  
  refinementList({
    container: '#brand',
    attribute: 'brand',
  }),
]);

search.start();

Best Practices

Provide Context

Send relevant hits and current refinements for better AI suggestions.

Handle Errors

Always implement error handling and display user-friendly messages.

Debounce Requests

Use appropriate debounce delays to avoid excessive API calls.

Show Loading States

Provide visual feedback during AI operations.

Styling

/* Chat Interface */
.message {
  padding: 12px;
  margin: 8px 0;
  border-radius: 8px;
}

.message.user {
  background: #e3f2fd;
  margin-left: 40px;
}

.message.assistant {
  background: #f5f5f5;
  margin-right: 40px;
}

.typing-indicator {
  display: flex;
  align-items: center;
  gap: 4px;
}

/* Filter Suggestions */
.filter-suggestions {
  margin: 16px 0;
  padding: 16px;
  background: white;
  border-radius: 8px;
  box-shadow: 0 2px 4px rgba(0,0,0,0.1);
}

.suggestion-chip {
  display: inline-flex;
  align-items: center;
  gap: 8px;
  padding: 8px 16px;
  margin: 4px;
  background: #f5f5f5;
  border: 1px solid #ddd;
  border-radius: 16px;
  cursor: pointer;
  transition: all 0.2s;
}

.suggestion-chip:hover {
  background: #e0e0e0;
  transform: translateY(-1px);
}

.suggestion-chip .count {
  background: white;
  padding: 2px 8px;
  border-radius: 12px;
  font-size: 0.875em;
}

/* Loading skeleton */
.skeleton {
  height: 32px;
  margin: 8px 0;
  background: linear-gradient(
    90deg,
    #f0f0f0 25%,
    #e0e0e0 50%,
    #f0f0f0 75%
  );
  background-size: 200% 100%;
  animation: loading 1.5s infinite;
  border-radius: 16px;
}

@keyframes loading {
  0% { background-position: 200% 0; }
  100% { background-position: -200% 0; }
}

Getting Started

Core Concepts

Vanilla JavaScript

React

Vue

Advanced Guides

Overview

Chat

Basic Setup

Chat with Custom Transport

Chat State Management

Tool Integration

Apply Filters from Chat

Chat with Suggestions

Filter Suggestions

Basic Setup

How Filter Suggestions Work

Custom Transport

Debouncing

Transform Suggestions

Complete AI Search Example

Best Practices

Provide Context

Handle Errors

Debounce Requests

Show Loading States

Styling

Build docs developers (and LLMs) love

Getting Started

Core Concepts

Vanilla JavaScript

React

Vue

Advanced Guides

​Overview

​Chat

​Basic Setup

​Chat with Custom Transport

​Chat State Management

​Tool Integration

​Apply Filters from Chat

​Chat with Suggestions

​Filter Suggestions

​Basic Setup

​How Filter Suggestions Work

​Custom Transport

​Debouncing

​Transform Suggestions

​Complete AI Search Example

​Best Practices

Provide Context

Handle Errors

Debounce Requests

Show Loading States

​Styling

​Related Resources

Build docs developers (and LLMs) love

Overview

Chat

Basic Setup

Chat with Custom Transport

Chat State Management

Tool Integration

Apply Filters from Chat

Chat with Suggestions

Filter Suggestions

Basic Setup

How Filter Suggestions Work

Custom Transport

Debouncing

Transform Suggestions

Complete AI Search Example

Best Practices

Styling

Related Resources