AI Technology

Dual-AI architecture
for speed and depth.

ExtractVibe uses two AI systems working together: Cloudflare Workers AI for fast, free bulk analysis, and OpenRouter for deep brand voice and personality understanding. The result is near-zero AI cost with high-quality output.

Try it free API docs

Architecture

Two AI systems, one extraction

Each AI system handles what it does best. Workers AI runs free models at the edge for structured data extraction. OpenRouter connects to frontier models for nuanced language analysis.

Primary

Cloudflare Workers AI

Runs Meta Llama 3.1 8B Instruct at the edge with zero additional cost. Handles structured extraction tasks: parsing HTML/CSS, extracting color values, detecting font families, and identifying component patterns.

Free tier with generous limits
Edge execution for low latency
Handles 80% of extraction tasks
Structured JSON output

Fallback

OpenRouter

Connects to high-quality models like Gemini Flash 2.0 for tasks requiring nuanced language understanding. Used for brand voice analysis, personality profiling, and vibe synthesis where depth matters more than speed.

Access to frontier models
Deep language understanding
Used for voice and vibe analysis
Cost-optimized model selection

Extraction Pipeline

Five-step extraction pipeline

Every brand extraction follows a durable, multi-step workflow. Each step runs independently with automatic retries, ensuring reliable results even on complex websites.

Extraction Pipeline

URL Input
     |
     v
+-------------------+
| 1. FETCH & RENDER |  Cloudflare Browser Rendering
|   Puppeteer        |  Full-page screenshot + DOM
+-------------------+
     |
     v
+-------------------+
| 2. PARSE ASSETS   |  Workers AI (Llama 3.1 8B)
|   HTML/CSS/Images  |  Extract colors, fonts, logos
+-------------------+
     |
     v
+-------------------+
| 3. ANALYZE VOICE  |  OpenRouter (Gemini Flash 2.0)
|   Copy & Content   |  Tone, personality, style
+-------------------+
     |
     v
+-------------------+
| 4. SYNTHESIZE     |  OpenRouter (Gemini Flash 2.0)
|   Brand Profile    |  Combine visual + verbal
+-------------------+
     |
     v
+-------------------+
| 5. SCORE & PACK   |  Workers AI (Llama 3.1 8B)
|   Brand Kit JSON   |  Quality score + export
+-------------------+
     |
     v
  Brand Kit Output

01 / Fetch & Render

Browser rendering

Cloudflare Browser Rendering launches a headless browser to render the full page, execute JavaScript, and capture a screenshot. The complete DOM with computed styles is then available for analysis.

02 / Parse Assets

Visual identity extraction

Workers AI analyzes computed styles on every DOM element to extract colors, fonts, spacing, gradients, border radii, and shadow values. Logos and icons are identified and downloaded to R2 storage.

03 / Analyze Voice

Brand voice analysis

OpenRouter sends the extracted text content to Gemini Flash 2.0 for deep language analysis. The model classifies tone, identifies personality traits, and generates writing style guidelines.

04 / Synthesize

Vibe synthesis

Visual and verbal data are combined into a holistic brand profile. The AI synthesizes a "vibe" description capturing the overall brand feel, and generates actionable do's and don'ts.