IntelliWP settings configuration guide, RAG + Fine-tuning WordPress AI agent setup
IntelliWP represents a significant evolution in artificial intelligence integration within the WordPress ecosystem. Its Settings configuration page constitutes the control hub from where developers can implement a sophisticated WordPress AI agent that goes beyond traditional chatbots.
The distinctive characteristic that positions IntelliWP above other plugins lies in its dual RAG + Fine-tuning architecture. This flexibility allows users to choose the most suitable methodology according to their specific needs, something no other WordPress plugin offers with this technical sophistication.
OpenAI API configuration, the foundation of connectivity
API key management in IntelliWP implements an intelligent priority system that prioritizes security over convenience. The system searches for the API key in hierarchical order: first in server environment variables, then in wp-config.php, and finally in the database as fallback.
When the key loads from external sources, the field appears disabled with the message “This field is read-only because the API key is loaded from an external source”, confirming the system operates in secure mode. In this case, a key preview is shown for verification without exposing complete credentials.
For manual configuration, the field allows direct API key input with placeholder “Enter your OpenAI API key”. The system automatically validates connectivity with OpenAI and displays connection status.
The prominent terms of service warning is not merely informative: it reminds that although IntelliWP is GPL, OpenAI API usage is subject to specific OpenAI terms, establishing clear responsibilities regarding content processing.
Operation mode selection, Fine-tuning vs RAG
This is perhaps the most critical decision in the entire configuration. IntelliWP offers two completely different methodologies for creating your Smart WordPress AI agent, each with specific advantages depending on the use case.
Fine-tuning (Classic Mode) creates a model specifically trained with your content. The technical description explains: “Creates a model specifically trained with your content. Requires retraining when content changes.” This method is ideal for sites with relatively static content where precision and consistency are priorities.
RAG (Retrieval-Augmented Generation) uses vector search to retrieve relevant content for each query. The description indicates: “Uses vector search to retrieve relevant content for each query. Automatically updates when content changes.” When you select this option, a “Configure more” link appears dynamically, leading to specific WordPress RAG agent configurations.
The choice between these methods completely determines the system’s subsequent behavior, including operational costs, response speed, and flexibility for content updates.
Content sources management
IntelliWP provides granular control over what content is included in the training or indexing of the WordPress intelligent agent.
The “Include posts and pages in training” option comes activated by default and extracts content from all published posts and pages on the site. This is the primary knowledge source for most implementations.
“Include files from the media library” allows extracting text from files in the media library. This feature comes disabled by default with the warning “This can significantly increase training time”, reflecting the impact on performance and costs.
The “Use AI to generate enhanced training datasets” feature uses OpenAI API to generate high-quality question-answer pairs. The technical description specifies: “Uses the OpenAI API to generate high-quality question and answer pairs” with the explicit warning “This option consumes additional API tokens and may increase costs.”
Base model selection and configuration
The model selection dropdown organizes options into two clearly differentiated categories. “Base Models” include official models like GPT-3.5 Turbo, GPT-4, and GPT-4 Turbo, which always appear available when the API key is valid.
“Fine-Tuned Models” show previously trained models by the user. If no custom models exist, it displays “No fine-tuned models available”, indicating that at least one training cycle must be completed before having additional options.
Base model selection determines the agent’s fundamental capabilities, including contextual understanding, response quality, and operational costs.
Advanced configuration and cost control
IntelliWP demonstrates exceptional transparency in cost handling through detailed pricing information. Approximate costs are clearly presented: “Training: $0.008 per 1K tokens” and “Fine-tuned model usage: $0.012 per 1K tokens (input), $0.016 per 1K tokens (output)”.
The direct link to OpenAI’s official pricing page allows verifying updated information, acknowledging that prices may vary.
The “Content limit for AI enhancement” field controls the maximum number of content items processed with OpenAI API. The default value of 100 represents a balance between dataset quality and reasonable costs. The description explains: “Maximum number of content items that will be processed with the OpenAI API. A higher number improves dataset quality but increases cost.”
Rate limiting protection
The protection system against excessive usage includes the “Protect against API overuse” checkbox that prevents uncontrolled API calls. This feature is essential for avoiding unexpected costs from accidental mass usage.
The “Requests per minute” field allows configuring the maximum number of API requests per minute, with a default value of 50 and recommendation of 20 for conservative usage. This control is particularly important on high-traffic sites where multiple users could simultaneously interact with the agent.
External access control and security
The allowed domains configuration includes five specific fields for authorized URLs. Only domains listed here can access the chat API from external JavaScript applications, establishing a clear security perimeter.
The security note explains that “Only domains listed here will be able to access the Chat API from external JavaScript applications”, emphasizing the importance of correctly configuring these fields before implementing external integrations.
The “Enable CORS headers for API responses” checkbox enables cross-origin requests from authorized domains. This configuration is mandatory for external JavaScript applications and must be activated along with the allowed domains list.
JavaScript snippet integration
IntelliWP facilitates external implementation through a pre-configured JavaScript snippet. The large text field contains ready-to-copy-and-paste code for external sites, with the instruction “Copy this code and paste it into your external website’s HTML.”
The “Easy Integration” functionality includes options for customizing theme and responsive configurations, along with a “Copy JavaScript snippet” button that facilitates implementation without transcription errors.
Security notices and final validation
Final security notices establish three critical requirements for correct functionality: snippets only work on domains from the allowed list, it’s necessary to add domains to the allowed list, and CORS headers must be enabled if using JavaScript snippet.
The fourth point recommends testing the integration on the target domain before considering the implementation complete.
Distinctive features of intelliWP
The dual architecture allows switching between Fine-tuning WordPress AI agent and WordPress RAG agent from the same interface through radio buttons in the Operation Mode section. Once RAG is selected, the “Configure more” link appears dynamically to access specific configurations.
The API key management system implements hierarchical search: environment variables, wp-config.php, and database as fallback. When external configuration is detected, the field appears disabled showing “This field is read-only because the API key is loaded from an external source.”
Cost transparency includes detailed information: “Training: $0.008 per 1K tokens” and “Fine-tuned model usage: $0.012 per 1K tokens (input), $0.016 per 1K tokens (output)” with direct link to OpenAI’s official pricing page.
The IntelliWP Settings page concentrates all strategic decisions that determine the behavior, performance, and costs of the AI system. Mastering these configurations provides complete control over implementation, allowing optimization of the balance between functionality, performance, and costs according to specific project requirements.