Kombai

Kombai is built for the design-to-code loop that happens before and after a mockup exists. Its taste agent identifies the design decisions a task needs — type scale, contrast, hero structure, hover behaviour, scroll reveals — and makes quality choices rather than falling back on the generic defaults AI models produce. Design Creativity Bench, the company's own benchmark, measures how much models repeat themselves across prompts; Kombai reports improvements in originality and creative range when models run through its pipeline.

The workflow combines three surfaces. An infinite canvas lets you explore several design directions in parallel, generate targeted variations, and edit text, styles and layout precisely while running live HTML, states and animations. A deep context layer reads your repo, Figma files, Storybook and npm packages so new work reuses real components, tokens and hooks instead of inventing new ones. A built-in frontend coding agent then produces code in your stack, with best-practice guidance for hundreds of libraries, and an editable browser for refining what was generated.

Kombai is distributed as a desktop app plus extensions for VS Code, Cursor and other IDEs, and it can be used alongside existing coding agents like Claude Code or Codex. Model choice is open: pick from Claude, GPT, Grok, Kimi and others, generate images or video with Veo, Seedance and Nano Banana, or let Kombai's router decide per task. Figma import is native, with cleanup of messy grouping and hidden layers plus asset extraction.