The Chrome AI RTD Provider is a Prebid.js Real-Time Data (RTD) module that enhances bidding by leveraging Chrome's built-in AI capabilities. It can automatically detect page language using the Chrome AI Language Detection API and generate page summaries or keywords using the Chrome AI Summarizer API. This information is added to the OpenRTB bid request objects, allowing bid adapters to optimize bids based on content language and context.

• Automatic language detection using the Chrome AI Language Detection API.
• Automatic page summarization or keyword generation using the Chrome AI Summarizer API.
• Caching of detected language and summaries/keywords in localStorage to reduce redundant API calls (configurable for summarizer).
• Configurable options for both language detection (e.g., confidence threshold) and summarization (e.g., type, format, length).
• Flexible ORTB2 path configuration for placing detected data.
• Ability to enable/disable each feature independently.
• Compatible with the Prebid.js RTD framework.

To include the Chrome AI RTD Provider in your Prebid.js build, use the following command:

gulp build --modules=rtdModule,chromeAiRtdProvider

Add the Chrome AI RTD Provider to your Prebid.js configuration:

pbjs.setConfig({
  realTimeData: {
    dataProviders: [{
      name: 'chromeAi',
      waitForIt: true  // Optional: delays the auction until language detection completes
    }]
  }
});

Configure language detection and summarization with additional options:

pbjs.setConfig({
  realTimeData: {
    dataProviders: [{
      name: 'chromeAi',
      waitForIt: true, // Set to true if auction should wait for both enabled features
      params: {
        languageDetector: {
          enabled: true,       // Set to false to disable language detection
          confidence: 0.9,     // Set minimum confidence threshold (0.0 - 1.0)
          ortb2Path: 'site.content.language' // Default path for language
        },
        summarizer: {
          enabled: false,      // Set to true to enable summarization/keyword generation
          type: 'headline',    // 'headline','key-points', 'tldr' or 'teaser'
          format: 'markdown', // 'plain-text' or 'markdown'
          length: 'short',     // 'short', 'medium', or 'long'
          ortb2Path: 'site.content.keywords', // Path for summary/keywords
          cacheInLocalStorage: true // Whether to cache generated summary/keywords
        }
      }
    }]
  }
});

{: .table .table-bordered .table-striped }

The module initializes configured features (language detection, summarization) asynchronously.

1. Data Prioritization: On initialization or when getBidRequestData is called, the module first checks for existing language information in this order:
   • Auction-specific ORTB2 data (from reqBidsConfigObj passed to getBidRequestData).
   • Data cached in localStorage for the current page URL (from a previous detection).
2. API Call: If no language is found and the feature is enabled, it attempts to detect the language of the visible page content using the Chrome AI Language Detection API.
   • The API's availability() method is checked. If 'unavailable', detection is skipped. If 'after-download', the module may proceed if the model downloads.
3. Data Handling: The detected language (if it meets the confidence threshold) is:
   • Stored in localStorage for future page loads on the same URL.
   • Added to the OpenRTB bid requests at the configured languageDetector.ortb2Path (default: site.content.language).

1. Data Prioritization: Similar to language detection, it checks for existing summary/keywords:
   • Auction-specific ORTB2 data.
   • Data cached in localStorage (if cacheInLocalStorage: true).
2. API Call: If no data is found and the feature is enabled, it attempts to generate a summary/keywords from the page content using the Chrome AI Summarizer API.
   • The API's availability() method is checked. If 'unavailable', summarization is skipped. If 'after-download', the module may proceed.
3. Data Handling: The generated summary/keywords are:
   • Stored in localStorage (if cacheInLocalStorage: true).
   • Added to the OpenRTB bid requests at the configured summarizer.ortb2Path (default: site.content.keywords).

If waitForIt: true is set in the RTD config, the auction will be delayed until all enabled and available Chrome AI features complete their processing.

• The browser must support the Chrome AI APIs being used (Language Detection, Summarizer).
• The specific Chrome AI models (e.g., for language detection or summarization) must be 'available' or become 'available-after-download'. The module handles these states.
• Sufficient text content must be available on the page (minimum 20 characters for language detection and summarization).
• If using the waitForIt: true option, consider the potential impact on auction latency.

• Relies on browser support for Chrome AI APIs.
• Requires sufficient and meaningful visible text content on the page for accurate results.
• Language detection may not be accurate for pages with multiple languages mixed together.
• Summarization quality depends on the page content and the capabilities of the underlying Chrome AI model.

Language Detector API Support

pbjs.setConfig({
  realTimeData: {
    dataProviders: [{
      name: 'chromeAi',
      waitForIt: true
    }]
  }
});

pbjs.setConfig({
  realTimeData: {
    dataProviders: [{
      name: 'chromeAi',
      params: {
        languageDetector: {
          enabled: false
        }
      }
    }]
  }
});

pbjs.setConfig({
  realTimeData: {
    dataProviders: [{
      name: 'chromeAi',
      waitForIt: true,
      params: {
        languageDetector: {
          enabled: true,
          confidence: 0.95  // Only use high-confidence detections
        }
      }
    }]
  }
});

pbjs.setConfig({
  realTimeData: {
    dataProviders: [{
      name: 'chromeAi',
      waitForIt: true,
      params: {
        languageDetector: {
          enabled: false // Example: only using summarizer
        },
        summarizer: {
          enabled: true,
          type: 'teaser',
          format: 'markdown', // In markdown format
          length: 'medium',
          ortb2Path: 'site.ext.data.summary', // Custom ORTB2 path
        }
      }
    }]
  }
});