Skip to content
Book demo
Back to Learn

Open Design v0.11.0: An Open-Source Claude Design Alternative for Your CLI

Jun 23, 2026
Paula Hingel
Paula Hingel
Open Design v0.11.0: An Open-Source Claude Design Alternative for Your CLI

Three things worth knowing

  • Open Design v0.11.0 ("The Bazaar") shipped June 17, 2026, with 69.7k stars, 7.9k forks, and 382 contributors under Apache-2.0.
  • It rebuilds Anthropic's Claude Design artifact loop as plain files that any coding agent can read and write, and runs across 22 CLIs, including Claude Code, Codex, Cursor, and Copilot.
  • The DESIGN.md brand contract and 150 shipped design systems turn a team's visual conventions into something an agent can enforce at generation time.

Design handoff breaks in predictable ways. A Figma file becomes a screenshot, a screenshot becomes a developer's best guess, and the brand drifts. Every team has a version of this problem, and most solutions add more tooling to the same broken chain.

Open Design takes a different position: if coding agents are already on your developers' machines and can read and write files, make the design system a file. Make the artifact real HTML. Skip the export.

The nexu-io/open-design GitHub repository showing 69.7k stars, 7.9k forks, and a directory listing including the skills, design-systems, and plugins folders.

What Happened

Open Design v0.11.0 shipped June 17, 2026. The nexu-io/open-design repository holds 69.7k stars, 7.9k forks, and 382 contributors across 13 releases under Apache-2.0.

The project is a direct response to Anthropic's Claude Design, which launched in April 2026 and went viral for delivering design artifacts from a chat prompt. Claude Design stayed closed-source, paid, cloud-only, and locked to Anthropic's models. Open Design rebuilds the same artifact-first loop as plain files: skills/, design-systems/, and plugins/ directories that any MCP-compatible agent can read, write, and remix.

The stack is Next.js 16 on the frontend, a Node 24 Express daemon with better-sqlite3 for storage, and an Electron shell for desktop builds. Native apps ship for macOS (Apple Silicon and Intel) and Windows (x64), with a Linux AppImage on an optional release lane.

Key Features

  • One-command MCP install: od mcp install <agent> wires the stdio MCP server into 22+ CLIs including claude, codex, cursor, copilot, gemini, openclaw, and more. --print for a dry run, --uninstall to remove.
  • 150 brand-grade DESIGN.md systems: Each is a single Markdown file with a 9-section schema covering color, type, spacing, motion, voice, and anti-patterns. Shipped brands include Linear, Stripe, Vercel, Apple, and Notion. Drop a folder into design-systems/<brand>/ and every render picks it up.
  • 261 official plugins: Portable agent-skill folders under plugins/_official/, each with a SKILL.md plus an optional open-design.json manifest for marketplace metadata and capability declarations. Categories cover design scenarios, image templates, Figma migration, and codebase refactors.
  • Five artifact types in one Studio: Prototypes (web/mobile/desktop HTML), live dashboards, decks, images, and HyperFrames motion graphics. Decks export to HTML, PDF, PPTX, ZIP, or Markdown; HyperFrames render to MP4 via headless Chrome and FFmpeg.
  • BYOK proxy with SSRF guard: No CLI installed? POST /api/proxy/{anthropic,openai,azure,google,ollama,senseaudio}/stream runs the same loop with any OpenAI-compatible endpoint. Per-target SSRF protection blocks internal IPs at the daemon edge.
  • Local-first security model: The daemon binds to 127.0.0.1 and is read-only by default. LAN exposure requires an explicit OD_BIND_HOST and OD_ALLOWED_ORIGINS.

Why It Matters

When a design system lives as a Markdown file in the repo, it travels with the code. It's versioned, diffable, and readable by any agent on any CLI. Compare that to the theme JSON in Figma: sometimes exported, manually synced, out of date by Thursday.

For teams already running coding agents, the MCP installation is low-friction. One command, 22 CLIs, and an agent in a separate repo can query your design source by name and always get the live file, not a stale export. No zip attachment, no re-upload.

Running across 22 CLIs and a BYOK proxy also means you're not locked to one provider. Claude Design tied you to Anthropic's models. Open Design runs on whatever CLI your team already uses, and if no CLI is installed, the BYOK proxy handles it.

Example Use Case

A PM needs a landing page in the Linear style. Inside Claude Code, they type:

Use open-design to generate a landing page with the Linear design system

The agent reads skills/, picks the matching SKILL.md, binds the Linear DESIGN.md, and emits an artifact previewable at http://localhost:7456. Single-page HTML, rendered in a sandboxed iframe, editable in place rather than regenerated from scratch.

When the direction is locked, the same HTML/CSS is dropped into Cursor or Codex to continue building as a real component. No zip export, no re-attach, no fidelity loss in transit.

I'd walk through this with any team where a designer and an engineer are currently passing Figma links back and forth. The real question isn't whether the tooling works. It's about whether your team will actually commit to keeping a DESIGN.md up to date alongside the codebase.

Competitive Context

Claude Design pioneered the artifact-first loop but remains Anthropic-only: closed-source, no self-hosting, locked to Anthropic's models and skills. Open Design runs the same loop across 22 CLIs plus BYOK, ships under Apache-2.0, and runs on your laptop or your Vercel project.

Open source
augmentcode/augment-swebench-agent873
Star on GitHub

Figma works on a canvas with a theme JSON for tokens. Open Design outputs real CSS, real fonts, and real components shaped by a DESIGN.md. Different medium, different output.

Cursor and Claude Code are in a different category. They're the execution layer Open Design plugs into, not competitors. OD MCP install cursor and OD MCP install claude wire the tools into both, so the artifact loop runs inside whichever agent a developer already prefers.

My Take

382 contributors on a project that's two months old is a fast start. The commit velocity, active CI pipelines, and 244 deployments in the repo suggest this isn't a side project that collected stars and stalled.

The number I'd watch is how widely DESIGN.md gets adopted outside this repo. Right now, it's a clever file format with 150 good examples. If other projects start treating it as a standard, it becomes something you can assume any agent knows how to read. That's a different proposition entirely, and it's not something the maintainer can force.

[ Coming up next ]

The New Code Review Workflow for AI-Native Engineering Teams

See how leading teams keep code review fast and rigorous as AI writes more of the code.

Save your seat
Thu, Jul 9 // 9:45 AM PDT

Written by

Paula Hingel

Paula Hingel

Technical Writer

Paula writes about the patterns that make AI coding agents actually work — spec-driven development, multi-agent orchestration, and the context engineering layer most teams skip. Her guides draw on real build examples and focus on what changes when you move from a single AI assistant to a full agentic codebase.

Get Started

Give your codebase the agents it deserves

Install Augment to get started. Works with codebases of any size, from side projects to enterprise monorepos.