Summary

Open Design addresses the limitations of closed-source, cloud-only AI design tools like Anthropic's Claude Design. It provides a local-first, open-source alternative enabling users to leverage their existing coding agents for design artifact generation. This empowers developers and designers with greater control, flexibility, and cost-efficiency by avoiding vendor lock-in and enabling self-hosting.

How It Works

The system comprises a web frontend and a local daemon. The daemon automatically detects and utilizes installed coding agent CLIs (e.g., Claude Code, Copilot CLI) as the design engine. It orchestrates a workflow driven by 19 composable "Skills" and 71 integrated "Design Systems." Users interact via a structured prompt stack that includes an initial question form, a visual direction selector, and a five-dimensional self-critique mechanism, ensuring a deterministic and high-quality output process.

Quick Start & Requirements

• Primary Install/Run: Clone the repository, install dependencies with pnpm install, and start the development server with pnpm dev:all.
• Prerequisites: Node.js (v22 recommended via .nvmrc), pnpm, and at least one supported coding agent CLI (e.g., claude, codex, copilot) must be installed and accessible in the system's PATH.
• Resource Footprint: A local runtime directory (.od/) is automatically created and managed by the daemon for project data and configurations.
• Links: Repository: https://github.com/nexu-io/open-design.git. Demo: http://localhost:5173.

Highlighted Details

• Agent Agnosticism: Supports numerous coding agents (Claude Code, Codex, Cursor, Gemini CLI, Copilot, etc.) via PATH detection, with Anthropic API as a BYOK fallback.
• Extensive Design Systems: Integrates 71 brand-grade design systems, sourced from awesome-design-md, providing diverse styling options.
• Composable Skills: Offers 19 pre-built skills for generating various artifacts like prototypes, presentations, documentation, and marketing pages.
• Structured Design Process: Features a mandatory turn-1 question form, a curated 5-direction visual picker, live todo progress tracking, and a rigorous 5-dimensional self-critique before artifact emission.
• Local Development Environment: The agent operates with full filesystem access (Read, Write, Bash, WebFetch) within a project's dedicated directory.
• Multi-Format Export: Supports exporting designs as HTML, PDF, PPTX, ZIP, and Markdown.

Maintenance & Community

The project draws inspiration from several key open-source initiatives. A roadmap detailing future development, including features like comment-mode edits and an AI-emitted tweaks panel, is available at docs/roadmap.md. Specific community channels or active contributor lists are not detailed in the README.

Licensing & Compatibility

The project is licensed under Apache-2.0. The guizang-ppt-skill component retains its original MIT license. The Apache-2.0 license generally permits commercial use and integration into closed-source projects.

Limitations & Caveats

This is described as an "early implementation." Key features like comment-mode surgical edits and an AI-emitted tweaks panel are still on the roadmap. The local runtime directory (.od/) path is currently hard-coded relative to the repository root and cannot be easily relocated. The system relies on coding agent CLIs being correctly configured in the user's system PATH for automatic detection.