AI Search

AI Search for WordPress enhances the search experience by:
– Replacing the default WordPress search with an AI-powered intelligent search system.
– Generating embeddings for posts using OpenAI’s text-embedding-3-small model via an external Node.js service (Open AI account isn’t necessary).
– Managing embeddings centrally with per-origin quota control.
– Caching results locally using WordPress transients.
– Ranking posts based on similarity using cosine similarity metrics.

Features

• Intelligent Search: AI-powered context inference.
• Remote Embedding Service: Embeddings stored in a central MySQL database through a Node.js API.
• Admin Settings: Manage global token and service URL.
• Bulk Embedding Generation: Generate embeddings for multiple posts or custom post types.
• Cache Management: Clear embedding cache and post meta embeddings.
• Enhanced Similarity Control: Fine-tuned similarity threshold with 3-decimal precision.
• Caching: Local transient caching.

Usage

• Simply use the default WordPress search, and AI Search will enhance it.
• Manage your OpenAI API key and plugin settings via the Settings > AI Search menu.

External Service

This plugin connects with:

• OpenAI Embeddings API: https://platform.openai.com/docs/guides/embeddings
• Custom Node.js Embedding Service

Please read more here:
https://platform.openai.com/docs/guides/embeddings

1. Install and Activate the plugin through the ‘Plugins’ menu in WordPress.
2. Go to Settings > AI Search and configure what you need.
3. Run Bulk Embedding Generation via admin panel (the current posts have to be regenerated).

1.9.0

• Enhanced Fallback Search System: Introduced intelligent 4-tier fallback mechanism for zero-result searches:
  • WordPress default search fallback with result count display
  • Broader semantic search with relaxed similarity thresholds
  • Unprocessed content search for posts without embeddings
  • Smart query suggestion system for completely failed searches
• User Feedback Notifications: Real-time search feedback system with dismissible notices:
  • Visual indicators showing which fallback method was used
  • Search result counts for transparency
  • Automatic cleanup of feedback transients
  • Contextual messaging for different search scenarios
• Interactive Threshold Demo: Professional threshold demonstration interface:
  • Real-time product similarity visualization
  • Sample clothing store dataset with summer search scenario
  • Dynamic filtering based on threshold adjustments
  • Performance indicators showing recall rates and precision levels
  • Color-coded similarity scores and visual feedback
• Token Validation System: Comprehensive service token management:
  • One-click token validation with detailed usage statistics
  • Real-time quota monitoring (daily/total limits)
  • Client information display including origin verification
  • Error handling for invalid or expired tokens
• Setup Wizard Enhancement: Improved first-time user experience:
  • Step-by-step configuration process
  • Provider comparison with clear recommendations
  • Visual progress indicators and smooth transitions
  • Comprehensive completion flow

1.8.0

• Service Token Management: Added manual service token field in General Settings for better control and transparency
• Token Validation System: New “Validate Token” button with real-time validation and detailed usage statistics
• Enhanced Admin Interface: Service token field automatically shows/hides based on provider selection
• Comprehensive Validation Results: Displays client ID, origin, daily/total limits, current usage, and quota status
• Improved Error Handling: Better error messages for token validation failures and connection issues
• User Experience Improvements: Loading spinner, button states, and clearer user guidance
• Debugging Enhancements: Added error logging for better troubleshooting and API response monitoring

1.7.0

• Major Performance & UX Improvements
• Service Registration Optimization: Moved AI service registration from plugin initialization to settings save, eliminating unnecessary API calls during normal page loads
• Cache Management Tab: Added comprehensive cache management interface with three clearing options:
  • Clear embedding cache (transients) – safe regeneration
  • Clear post meta embeddings – requires manual regeneration
  • Clear all embedding data – complete reset option
• Enhanced Similarity Threshold**: Improved precision and range:
  • Range changed from 0.0-1.0 to 0.5-1.0 (eliminates low-relevance results)
  • Precision increased from 0.01 to 0.001 steps (3 decimal places)
  • Better user guidance and formatting
• Modular Code Architecture: Refactored service registration into isolated, reusable methods for better maintainability
• Real-time Cache Statistics: Display current cache size and post embedding counts
• Enhanced Error Handling**: Better user feedback when AI service is unavailable with automatic fallback
• Security Improvements: Enhanced nonce verification and user capability checks

1.6

• Replaced internal embedding storage with external service integration.
• Token-based authentication per origin.
• Restore original search in case there isn’t any AI results
• Solve fatal errors in case the vectors don’t have the same size

1.5

• Bulk embedding generation based on custom number of posts

1.4

• Settings UI Revamp: Introduced tabbed navigation to separate “General Settings” and “Generate Embeddings” for better organization.
• Batch Embedding Generation: Users can now trigger embedding generation for up to 50 posts at a time.
• Custom Post Type Selection: A dropdown was added, allowing users to choose which CPT they want to generate embeddings for.
• Embedding Optimization: Now only processes posts that do not yet have embeddings to avoid redundant API calls and improve performance.
• Security Enhancements: Added nonce verification to protect settings updates.

This update significantly improves usability, optimizes performance, and gives users more control over AI-powered search embeddings.

1.3

• Similarity Threshold Control: Added a range input field (0-1) in settings to allow users to adjust the similarity threshold dynamically.

1.2

• Cache Search embeddings, saved in transients for one day

1.1

• text-davinci-003 (deprecated) has been replaced bytext-embedding-ada-002 model
• db manipulation has been replaced by custom_post_meta approach

1.0

• Initial release, adding the settings page and the search filter.