nexu-io/open-design

The 1st Collaborative AI Design Agent Workspace. Local-first and open source. Create motion, prototypes, websites, apps, dashboards, presentations, images, and videos with your team and AI agents. Share work, exchange feedback, and iterate together in one workspace.

OpenDesign – Collaborative AI‑powered design workspace

What it is – OpenDesign is a desktop‑first application that lets teams work with large‑language‑model (LLM) coding agents to create, iterate, and ship design artefacts such as web prototypes, slide decks, images, dashboards, and motion‑graphics videos. The product does not ship its own AI model; instead it connects to any installed “coding agent” (Claude, Codex, DeepSeek, Cursor, Copilot, etc.) via a plug‑in / MCP system, letting the user pick the model they already trust.

Key capabilities

  • Workspace & studio – a single project view where a brief (written in DESIGN.md) drives the agent to generate live‑preview HTML, React/Vue code, slide decks, or image/video assets. The UI shows a live iframe, a chat‑style conversation with the agent, and a panel of design‑system tokens.
  • Design systems & templates – over 150 ready‑made design‑system packages (brand assets, CSS tokens, optional component libraries) and dozens of HTML‑based prototype / deck templates that the agent can reference when rendering.
  • Plugins / skills – reusable, version‑controlled directories that expose agent‑driven workflows (e.g., “generate a dashboard”, “create a hyper‑frame video”). Users can install official plugins or author their own.
  • Multi‑model support – built‑in adapters for 26+ coding‑agent CLIs (Claude, Codex, DeepSeek Harness, Cursor, Copilot, etc.). Installation is a one‑liner like od mcp install claude.
  • Local‑first & BYOK – runs entirely on the user’s machine (macOS/Windows, source‑build for Linux). A BYOK proxy lets you point the daemon at any OpenAI‑compatible endpoint, with SSRF protection.
  • Export options – prototypes as sandboxed HTML, decks as HTML/PDF/PPTX/ZIP/Markdown, images from prompt templates, and video via the open‑source HyperFrames engine (HTML + CSS + GSAP → deterministic MP4 using headless Chrome + FFmpeg).

Typical workflow

  1. Create a brief – write a DESIGN.md or pick an example.
  2. Select a coding agent – install the desired CLI (od mcp install codex).
  3. Generate – the agent reads the brief, the active design system, and a template, then streams back code/assets while you watch a live preview.
  4. Iterate – continue the conversation, tweak parameters in the “tweaks panel”, and the iframe updates instantly.
  5. Export / hand‑off – download the final HTML, PPTX, or MP4, or push the generated code to a repo for developers to refine.

Why it matters – By treating the LLM as a design engine rather than a generic chatbot, OpenDesign gives product teams a repeatable, version‑controlled pipeline for brand‑consistent UI/UX artefacts, while keeping data local and model‑agnostic.


Quick start (from the README)

  • Desktop download – pre‑built macOS (Apple Silicon & Intel) and Windows binaries are provided; Linux users can build from source.
  • Three‑command setup – the badge in the README points to a QUICKSTART.md that claims you can get started with just three commands (download, install a coding‑agent, run od daemon start).

License – Apache 2.0.


All details above are taken directly from the repository’s README; no additional features have been inferred.

Related

  • Project
  • Project
  • Project
  • Project
  • Project