GPT-Vis

Overview

GPT-Vis is an AI-native visualization library from the AntV team, built for applications where an LLM produces the chart. Instead of hand-written chart code, a model generates a short, markdown-like "vis syntax" string (for example vis line followed by data rows), and GPT-Vis renders it. It also accepts plain config objects.

The core package, @antv/gpt-vis, is framework-agnostic and works with React 18+, Vue 3 or vanilla JavaScript. The shadcn registry wraps it in a React component plus one item per chart type, 27 items in total, covering everything from line and bar charts to sankey diagrams, mind maps and organization charts. It is released under the MIT license.

Features

• Natural syntax: A simple, markdown-like format that LLMs can generate with high accuracy.
• 26 chart types: Line, pie, bar, column, area, scatter, radar, funnel, waterfall, histogram, boxplot, violin, dual axes, liquid, word cloud, sankey, treemap, venn and more.
• Diagrams and trees: Mind map, fishbone diagram, flow diagram, network graph, organization chart and indented tree components.
• Streaming support: Renders in real time as output streams from an AI model.
• Fault tolerant: Handles incomplete or malformed data during generation.
• Intelligent defaults: Automatic data detection, color schemes and adaptive layouts.
• Agent skill: A chart recommendation and syntax skill installable with npx skills add https://github.com/antvis/GPT-Vis.

Tech stack

• @antv/gpt-vis
• React (registry wrapper), with Vue 3 and vanilla JS supported by the core package
• clsx, tailwind-merge

Registry

Install via the shadcn CLI registry:

npx shadcn@latest add @gpt-vis/component-name