Open Design: The Open-Source Claude Design Alternative Everyone's Talking About

Published: 2026-05-04 · 8 min read · GitHub: nexu-io/open-design · 20,000+ stars in 6 days

⚡ Exploding on GitHub right now — Open Design is the open-source answer to Anthropic's Claude Design. Local-first, runs on 13+ coding agents, 31 composable design skills, 72 brand-grade design systems, and full BYOK (bring your own key). No cloud lock-in, no model lock-in, no vendor lock-in. Here's how to set it up and start generating production-ready designs from your terminal.

What Is Open Design?

When Anthropic released Claude Design on April 17, 2026, it demonstrated something remarkable: an LLM that could stop writing prose and start shipping real design artifacts — HTML/CSS prototypes, slides, image compositions, even video storyboards. But Claude Design came with strings attached: closed-source, paid-only, cloud-only, locked to Anthropic's model stack.

Open Design (by nexu-io) is the open-source alternative that removes every one of those locks. Same artifact-first design loop, same skill-driven workflow, but running entirely on your laptop, using whatever coding agent you already have installed, with your own API keys.

It hit 20,000+ GitHub stars in under a week, making it one of the fastest-growing repos of 2026. This guide walks through installation, key concepts, and how to use it effectively.

Quick Start: 30 Seconds to Running

You need Node.js 20+ and pnpm installed. Pick your agent:

git clone https://github.com/nexu-io/open-design.git
cd open-design
pnpm install
pnpm tools-dev

That's it. Open Design auto-detects any coding agent on your PATH — Claude Code, Codex, Gemini CLI, GitHub Copilot CLI, Cursor, Qwen, Hermes, and more. If you don't have one, you can configure an OpenAI-compatible proxy endpoint as your design engine.

Once the dev server starts, navigate to http://localhost:5173 and type something like:

"Make me a pitch deck for our seed round — modern, magazine-style layout"

The interactive question form pops up, the agent picks a visual direction, and within seconds you're watching a real project folder being built on disk with sandboxed preview in the browser.

Key Features

1. 13+ Coding Agents, One Workflow

Open Design doesn't ship its own model. Instead, it leverages whatever coding agents you already have on your machine — Claude Code, Codex, Gemini CLI, GitHub Copilot CLI, OpenCode, Cursor Agent, Devin, Qwen, Hermes, Kimi, Pi, Kiro, Mistral Vibe. It auto-detects them at startup via PATH scanning and routes design tasks to the strongest available agent. No agent? A BYOK OpenAI-compatible proxy works as a fallback.

2. 31 Composable Skills

Skills are the building blocks. Each skill is a prompt template + file scaffold + checklist that teaches the agent a specific design capability. The current library has 31 skills covering brand identity, presentation decks, UI prototyping, video storyboarding, image generation, and more. Skills compose — you can chain "brand identity" → "product mockup" → "pitch deck" in a single request.

3. 72 Brand-Grade Design Systems

Design systems define the visual language — typography scales, color palettes, spacing grids, component patterns. Open Design ships 72 design systems ranging from minimalist tech styles to elaborate editorial layouts. Each is a deterministic library (not AI-generated fluff), so the agent stays within defined parameters.

4. Local-First + BYOK

Everything runs on your machine. Your design data never touches a third-party cloud. Bring your own API keys for any supported model provider. If you use Claude Code with your own API key, there's no additional billing — the design loop is additive to your existing agent usage.

5. Multi-Format Export

HTML, PDF, PPTX, MP4, ZIP — Open Design's artifact engine can export to five formats. The sandboxed preview uses React 18 + Babel for live rendering, and the export pipeline handles everything from slide decks to video storyboards.

Detailed Installation Guide

Prerequisites

Step-by-Step

# 1. Clone the repo
git clone https://github.com/nexu-io/open-design.git
cd open-design

# 2. Install dependencies
pnpm install

# 3. Start the development environment
pnpm tools-dev

# 4. Open in browser
# → http://localhost:5173

# 5. (Optional) Set your API key if using BYOK proxy
export OPENAI_API_KEY=sk-your-key-here

The first startup may take a minute as it scans for agents and builds the design system cache. Subsequent startups are near-instant.

How It Works Under the Hood

Open Design uses a local daemon architecture. The daemon runs as the only privileged process, responsible for:

When you submit a design request, the flow is:

  1. Intent parsing — The interactive form captures parameters (format, audience, style, constraints)
  2. Direction selection — The agent picks one of five curated visual directions from the design philosophy library
  3. Plan streaming — A TodoWrite plan streams into the UI showing what the agent will build
  4. Project scaffolding — The daemon creates an on-disk project with seed template, layout library, and checklist
  5. Execution — The agent reads the scaffold, runs a pre-flight checklist, then generates the artifact
  6. Self-critique — The agent runs a five-dimensional critique against its own output (consistency, accessibility, aesthetics, completeness, performance)
  7. Rendering — The final artifact renders in the sandboxed iframe for preview and export

Open Design vs Claude Design

FeatureOpen DesignClaude Design
PriceFree (open-source)Paid subscription
Local-first✅❌ Cloud-only
Model choice13+ agents + BYOKClaude only
Design systems72 (customizable)Limited built-in
Export formatsHTML/PDF/PPTX/MP4/ZIPHTML/PDF
Self-host✅ Anywhere❌
Skills library31 (composable)~10 (fixed)
PrivacyData stays localSent to cloud

Tips & Best Practices

Write Clear Prompts

The interactive question form is your friend. Be specific about your audience, format, and visual style. "Make me a pitch deck" is fine, but "Make me a magazine-style pitch deck for a seed-stage AI startup targeting enterprise VCs" produces dramatically better results.

Use the Right Agent for the Job

Open Design auto-detects your agents, but you can pin a specific agent if you know its strengths. Claude Code excels at detailed HTML/CSS artifacts. Gemini CLI handles creative design briefs well. Codex is fast for quick prototypes. Experiment with different agents on the same request to see the variation.

Iterate with Skills

Don't expect perfect output on the first try. Use the skill composability — generate a brand identity first, then ask for a pitch deck using that identity. Each iteration builds on the previous one, and Open Design maintains the project context between requests.

FAQ

Is Open Design production-ready?

It's under active development (version 0.x), but the core artifact loop is stable and hundreds of production designs have been generated with it. The open-source community is growing fast.

Can I use it without a coding agent CLI?

Yes. Configure an OpenAI-compatible proxy endpoint with export OPENAI_API_KEY=sk-... and Open Design will use that as the design engine. The experience is the same — just slower because of API round trips.

Does it require GPU?

No. The heavy lifting is done by the coding agent (which may use GPU if available), but Open Design itself is a lightweight Node.js daemon that orchestrates the workflow. It runs on any laptop.

Can I contribute my own design systems or skills?

Yes. Skills are just directories with markdown prompt templates and file scaffolds. Design systems are JSON palette/typography definitions. The project welcomes community contributions and has clear contributing guidelines.

Last updated: 2026-05-04 · This guide is maintained by the community. Report issues on GitHub.