The Technology Behind the Magic

Understanding how IrisAI transforms your website into an intelligent assistant.

From Content to Conversations in Three Steps

Building Your Knowledge Base

When you activate IrisAI, here’s what happens:

1. Content Discovery

The plugin scans your WordPress database for:

• Published pages
• Blog posts
• Custom post types
• WooCommerce products (if installed)

2. Text Extraction

Content is processed to extract:

• Titles and headings
• Main body text
• Excerpts and metadata
• URLs and structure

3. Vector Conversion

Using OpenAI’s embedding model, each piece of content is transformed into a 1536-dimension vector—a mathematical representation that captures meaning and context.

4. Storage

Vectors are stored in your WordPress database with efficient indexing for lightning-fast searches.

Understanding Questions, Not Just Keywords

When a visitor asks a question:

1. Question Analysis

The question is converted into the same vector space as your content.

2. Similarity Matching

IrisAI calculates cosine similarity between the question vector and all content vectors.

Mathematical formula: similarity = (A · B) / (||A|| × ||B||)

3. Hybrid Ranking

Results are ranked using both:

• Vector similarity (semantic meaning)
• Keyword matching (exact term matches)
• Relevance weighting (title matches rank higher)

4. Top Results

The most relevant content chunks are selected as context for the AI response.

Natural Conversations with Context

With relevant content found, IrisAI generates responses:

1. Context Assembly

The top matching content chunks are combined with:

• The user’s question
• Previous chat history
• System instructions

2. Proxy Communication

Your request flows through the secure proxy system:

3. Response Formatting

The AI response includes:

• Natural, conversational answer
• Source citations with URLs
• Clickable reference cards

4. History Persistence

The conversation is saved in browser storage for continuity across the chat session.

Why Use the IrisAI Proxy?

The proxy system sits between your WordPress site and OpenAI, providing:

Authentication & Authorization

• Verifies your API key with every request
• Confirms your domain is registered and active
• Prevents unauthorized usage

Billing & Credits

• Tracks token usage per request
• Deducts credits from your account balance
• Provides transparent pricing
• No surprise OpenAI bills

Rate Limiting

• Prevents abuse and runaway costs
• Configurable limits per domain
• Protects against bot attacks

Multi-Domain Management

• One account, up to 5 WordPress sites
• Separate usage tracking per domain
• Centralized credit management

Usage Analytics

• Real-time dashboard showing requests
• Token consumption history
• Cost breakdowns
• Performance metrics

Security

• HTTPS encryption for all traffic
• API keys never exposed to clients
• Domain validation on every request
• Request logging for audit trails

Getting Started:

1. Sign up at irisai.cloud
2. Register your WordPress domain
3. Purchase credits via PayPal
4. Copy your proxy URL and API key
5. Enter in IrisAI plugin settings

Bring Your Own Key (Advanced)

Prefer to connect directly to OpenAI?

How BYO Mode Works:

• Uses your personal OpenAI API key
• Direct connection to OpenAI (no proxy)
• You manage billing with OpenAI directly
• Full control over model selection

When to Use BYO:

• You already have OpenAI credits
• You need custom model configurations
• You want to avoid the proxy layer
• You have high-volume usage

Setup:

1. Get API key from platform.openai.com
2. Enable BYO mode in General tab
3. Enter your OpenAI API key
4. Configure model preferences

Note: Your API key is stored securely on your server and never sent to browsers.

For the Technically Curious

Vector Embeddings Explained

Text: "I love WordPress"
(OpenAI text-embedding-3-small)
Vector: [0.123, -0.456, 0.789, ... 1536 dimensions]

Similar text has similar vectors:

• “WordPress is great” – Similar vector
• “I hate spinach” – Very different vector

Cosine Similarity Formula

similarity = cos(θ) = (A · B) / (||A|| × ||B||)

Where:

• A and B are vectors
• · is dot product
• || || is magnitude

Range: -1 (opposite) to 1 (identical)

Database Structure

wp_irisai_vectors
- id (primary key)
- post_id (WordPress post ID)
- chunk_index (for split content)
- content_text (original text)
- embedding (1536-float array)
- metadata (JSON: title, url, type)
- created_at

Performance Optimization

• Vectors stored as binary data
• Indexed by post_id for fast lookups
• Chunking prevents token limit issues
• Results cached for repeated queries

Token Usage

• Average embedding: ~500 tokens
• Average query: ~50 tokens
• Average response: ~800 tokens
• Total per interaction: ~1,350 tokens

Frequently Asked Questions

How accurate is semantic search?

Typically 85-95% accuracy for finding relevant content, significantly better than keyword-only search.

How long does indexing take?

Approximately 2-3 seconds per page. A 100-page site takes about 5 minutes.

Does it slow down my site?

No. Indexing happens once in the admin. Chat requests are asynchronous and don’t block page loads.

Can I exclude pages from indexing?

Yes, filter by post type or add custom exclusion logic via hooks.

How much do credits cost?

Proxy mode: approximately $0.002 per average chat interaction. See Pricing page for details.

Is my data private?

Yes. Content is only processed for embedding generation. OpenAI doesn’t store your content. The proxy doesn’t retain message data.

What happens if I run out of credits?

The chat will stop working until you top up. You’ll see low balance warnings in your proxy dashboard.

Can I use this on a localhost development site?

Yes, but you’ll need to register localhost as one of your domains or use BYO mode.