Open Design is an open-source alternative to Anthropic's Claude Design that turns your existing coding agents into a full design engine, and it has over 34,000 GitHub stars.
Open Design is an open-source alternative to Anthropic's Claude Design that turns your existing coding agents into a full design engine, and it has over 34,000 GitHub stars.
Open Design is an open-source project that transforms coding agents you already have installed, like Claude Code, Cursor, Gemini CLI, and others, into a capable design tool. Instead of building a new AI from scratch, it provides a layer of composable skills and brand-grade design systems that sit on top of your existing agent infrastructure.
The project supports 16 different coding agent CLIs out of the box, including Claude Code, Devin, Cursor Agent, Gemini CLI, Gemini 2 Flash, OpenCode, Qwen, Qoder CLI, Kiro, Kimi, Pi, Hermes, and more. It ships with 31 design skills, 72 pre-built design systems modeled after brands like Linear, Stripe, Vercel, Airbnb, Notion, Apple, and Anthropic, and 5 visual direction presets. It also includes device frame mockups, a BYOK proxy for connecting to multiple AI providers, direct import of Claude Design ZIP files, and local-first data storage via SQLite. The project is released under the Apache 2.0 license.
Open Design automatically detects which coding agent CLIs are installed on your machine and integrates with them. When you want to create a design artifact, you select a skill, such as web prototype, dashboard, or presentation, and a design system, such as Linear or Stripe. The skill defines what kind of output to produce, and the design system defines the visual language, including colors, typography, spacing, and component styles.
The 5 visual directions provide higher-level aesthetic choices: Editorial Monocle for magazine-style layouts, Modern Minimal for clean and simple designs, Warm Soft for friendly and approachable interfaces, Tech Utility for developer-focused aesthetics, and Brutalist for bold and raw visual statements. These use deterministic OKLch color palettes, which means the colors are generated consistently rather than varying between runs.
Device frames let you preview designs in context, with support for iPhone 15 Pro, Pixel, iPad, MacBook, and browser chrome. The BYOK proxy handles routing to your preferred AI provider. Everything is stored locally in SQLite, so your projects and data stay on your machine.
The breadth of supported agents and skills is impressive but also means quality will vary. Not all 16 agents will produce equally good results with all 31 skills. Some combinations will work well; others will produce mediocre output. You will need to experiment to find the combinations that work for your needs.
The 72 design systems are modeled after real brands, but "modeled after" is not the same as "endorsed by." The fidelity of these systems to the actual brand guidelines will vary, and using them in production work for real companies requires your own quality review.
The project is ambitious in scope, which means the surface area for bugs and inconsistencies is large. With 16 agents, 31 skills, 72 design systems, and 5 visual directions, the number of possible combinations is enormous, and not all of them will be well-tested.
Finally, this is a tool that generates design artifacts, not a tool that replaces design thinking. It can produce a polished dashboard mockup quickly, but deciding what should go on that dashboard, why, and for whom remains a human job.
Open Design is for developers and small teams who want to produce professional design artifacts quickly using AI coding agents they already have. If you are a developer who cannot afford a designer, a startup that needs rapid prototyping, or a team that wants to standardize on a set of design systems without building them from scratch, this is built for you. Professional designers may find it useful for rapid iteration but will likely want to refine the output in their own tools.
The takeaway: Open Design is one of the most ambitious open-source design tools to emerge in the AI era, and its community traction suggests it is solving a real problem. If you use any of the supported coding agents, it is worth trying.