Architecture

OpenVizAI separates AI reasoning from deterministic rendering.

Pipeline

1. Input: user prompt + dataset rows
2. Sampling: the runtime selects a limited row sample
3. AI decision: model selects chart type and field mappings
4. Runtime transform: deterministic code maps full dataset using the chart spec
5. Render: React components render with your selected chart library (apexcharts or chartjs)

Why This Scales

• Token cost is stable because row sampling is bounded
• Full dataset transformation remains deterministic and auditable
• Rendering logic is isolated from provider/model changes
• Rendering library is explicit at runtime via chartLibrary