Open CoDesign

Open CoDesign: Your Prompts, Your Model, Your Laptop — A Local-First Design Studio

Introduction

Open CoDesign is an ambitious open-source desktop tool that turns prompts into polished design artifacts entirely on your laptop. Built to be model-agnostic and local-first, it lets you work with Claude, GPT, Gemini, Ollama, or any OpenAI-compatible endpoint without cloud lock-in or vendor dependency. It’s an MIT-licensed desktop app that puts you in control: bring your own models, manage your own keys, and export real files—HTML, PDF, PPTX, ZIP, or Markdown—without forced cloud workflows. The project presents itself as a powerful alternative to cloud-only design suites, emphasizing speed, privacy, and flexibility.

[Hero image: Open CoDesign — prompt on the left, live artifact on the right]

What Open CoDesign Is

A desktop-native, open-source design studio. It runs locally, keeping your data on-device and giving you control over how you work with AI.
A bridge between prompts and polished artifacts. It plans, writes, self-checks, and ships designs that look like they came from a professional design tool.
A multi-model platform. Claude, GPT, Gemini, Ollama, OpenRouter, DeepSeek, and more are supported, with a dynamic model catalog rather than a fixed, hardcoded list.
A toolkit for rapid prototyping and export. From HTML and CSS to PDFs, PPTX, ZIPs, and Markdown, you can export production-ready assets directly from your local workspace.

Why People Are Excited About It

Runs on your laptop. There’s no mandatory cloud workspace, so you retain privacy and control.
Works with your model. You’re not locked into a single provider; switch between Claude, GPT, Gemini, Ollama, and other compatible endpoints.
Exports real files. The pipeline ends with tangible deliverables that you can share or publish.
Shows its work. The agent’s activity, tool calls, and progress are visible and interruptible, providing transparency into the generation process.

Key Highlights and Visual Dines

Comment, don’t retype. Click any element, drop a pin, and let the model rewrite only that region. This makes precise edits faster and safer.
Image: Comment mode in action
AI-tuned tweaks. The tweaks panel surfaces the parameters worth tweaking—color, spacing, typography—so you refine without re-prompting.
Image: Tweaks panel with color pickers
Your Designs hub. Every iteration is kept locally, letting you switch between recent versions instantly.
Image: Hub of designs
Live agent visibility. A live agent panel shows todos, tool calls, and streaming progress, so you know exactly what the model is doing.
Image: Live agent panel

What it Is Not

It is not a single-vendor, cloud-only design studio. Open CoDesign emphasizes openness, local-first operation, and model-agnostic workflows.
It is not a closed ecosystem. The project invites contributions, model integrations, and user-defined taste.

A Quick Tour: How It Works

Turn prompts into prototypes locally. You provide a brief, and the agent plans, writes, self-checks, and ships a finish-ready artifact with hover states, tabs, and empty states already wired up.
A design-first mindset. The system includes 12 built-in design skill modules to cover common design tasks, from landing pages to dashboards and pricing sections.
No vendor-binding export. You export your work in standard formats (HTML with inlined CSS, PDF via local Chrome, PPTX, ZIP, Markdown) for immediate use in real projects.

The Quickstart Path: Time to First Artifact

Time to first artifact: about 3 minutes.
Requirements: one API key or a local Ollama instance.
Operating systems: macOS 12+ (Monterey or later), Windows 10+, Linux with glibc ≥ 2.31.
Install options:
One-liner (recommended):
- Windows: winget install OpenCoworkAI.OpenCoDesign
- macOS: brew install --cask opencoworkai/tap/open-codesign
Direct download (v0.1.x) from GitHub Releases (look for the appropriate package for your platform).
Add your API key or local model:
Anthropic (sk-ant-…)
OpenAI (sk-…)
Gemini
Any OpenAI-compatible relay (OpenRouter, SiliconFlow, local Ollama) Credentials are stored locally in ~/.config/open-codesign/config.toml with restricted permissions.
Type your first prompt. Choose from fifteen built-in demos (landing page, dashboard, pitch slide, pricing sections, mobile UI, event calendar, blog article, receipts, portfolio, settings panel, etc.) or describe your own. A sandboxed prototype appears in seconds.

Bringing Your Stack: Importing Your Existing Keys and Providers

If you already use Claude Code or Codex, you can import your providers, models, and API keys in a single click—no re-entry of settings required.
A single-click import reduces setup friction and enables you to hit the ground running with your accustomed tools.

Built-in Taste: Designing with Purpose

Twelve built-in design skill modules. These cover slide decks, dashboards, landing pages, SVG charts, glassmorphism, editorial typography, hero sections, pricing blocks, footers, chat UIs, data tables, and calendars.
A design taste layer guides the model toward better typography, thoughtful whitespace, and purposeful color choices. This is applied before any CSS is written, ensuring the design respects layout intent, design-system coherence, and contrast.
Extendable taste. You can add a SKILL.md to any project to teach the model your own design preferences.

What You Get: Models, Generation, and Editing

Unified provider model. Works with Anthropic, OpenAI, Gemini, DeepSeek, OpenRouter, Silicon Flow, local Ollama, and more. Supports keyless proxies as needed.
One-click import from Claude Code and Codex configurations. Bring your existing providers in without rekeying.
Dynamic model picker. Each provider exposes its real catalog of models, rather than a fixed, short list.
Prompt-to-prototype generation. The prototype is rendered in a sandboxed iframe (on-device React 18 + Babel).
Demos and skill modules. Fifteen demos plus twelve design skill modules provide ready-to-edit starting points for common briefs.
Live agent panel. Tool calls stream in real time so you can observe the model’s workflow as it edits files.
AI image generation. Opt-in for bitmaps for heroes, product shots, backgrounds, and illustrations.
AI-generated sliders. The model emits parameters worth tweaking—color, spacing, font—so fine-tuning is precise and efficient.
Comment mode. Click an element, drop a pin, and have the model redo only that region.
Generation cancellation. You can stop a generation mid-stream without losing prior turns.

Preview and Workflow: Seeing Is Believing

Multi-device previews. True responsive frames for phone, tablet, and desktop; switch with one click.
Files panel. Inspect multi-file artifacts (HTML, CSS, JS) before export.
Instant design switching. The last five designs keep their preview iframes alive, preserving context and enabling seamless Hub ↔ Workspace navigation.
Connection diagnostic panel. One-click tests for any provider with actionable error messages.
Per-generation token counter. See exactly how many tokens a given generation consumes, in the workspace sidebar.
Comprehensive Settings. Four tabs—Models, Appearance, Storage, and Advanced. Includes light and dark themes and EN + 简体中文 UI with live toggle.
Local-first export. All exports are designed to be production-ready and easy to integrate into existing workflows.

Export and Packaging: Ready for Real Projects

Five export formats:
HTML (with inlined CSS)
PDF (via local Chrome)
PPTX
ZIP
Markdown
GitHub release pipeline. Generates unsigned DMG for macOS, EXE for Windows, and AppImage for Linux. Future releases will include signed builds and auto-update features.
Roadmap alignment. The current v0.1.4 focuses on AI image generation, OAuth login for ChatGPT Plus/Codex users, CLIProxyAPI import, and improved API diagnostics. v0.2.0 will introduce Agentic Design, turning Open CoDesign into a working workspace with session history and design tools on demand.

Roadmap: From Generator to Design Agent

Current release: v0.1.4. Now shipping AI image generation, OAuth login for ChatGPT Plus/Codex, and fortified API diagnostics.
Next: v0.2.0 (Agentic Design). Aims to transform Open CoDesign from a one-shot generator into a true local design agent with:
Design as a session. Each design is a JSONL-backed session with a persistent workspace on disk.
Permissioned agent loop. A pi-based internal loop for read, write, edit, bash, grep, find, and ls—guarded by a robust permission UI.
On-demand design tools. Ask, scaffold, skill, preview, gen_image, tweaks, todos, and done for flexible workflow control.
DESIGN.md as shared memory. Brand tokens and design-system decisions become editable files rather than model memory.
Migration path. Existing SQLite designs can migrate into workspaces and session histories.
Later: v0.2.x and beyond
Renderer improvements for new tools and process-management surfaces.
Cost transparency with pre-generation estimates and weekly budgets.
Version snapshots and side-by-side diffs for design comparisons.
Three-style parallel exploration to broaden design exploration.
Codebase-to-design system work (token extraction) and a path toward code-signing with a future auto-update feature.
Figma layer export and other evolving features.

Built On: The Foundation

Technologies: Electron, React 19, Vite 6, Tailwind v4.
Core primitives: @mariozechner/pi-ai and pi-coding-agent for model/provider integration and the agent loop.
Database and packaging: better-sqlite3, electron-builder.
The architecture is designed to be extendable, with room to incorporate new providers and evolving design modules.

Community and Collaboration: Getting Involved

Community channels
GitHub Discussions for show-and-tell, Q&A, and ideas.
Linux Do community for Chinese-speaking users and feedback.
GitHub Issues for bug reports and reproducible problems.
WeChat group for Chinese-speaking users
QR code is posted in the project repository. Note that the QR rotates every 7 days and is valid for a limited window.
If the code expires, you can request a refresh via GitHub Issues.
Contributing
See CONTRIBUTING.md for guidance.
Open an issue before writing code and run lint, typecheck, and tests before a PR.
Licensing
MIT license. You’re free to fork, ship, and modify the project, with third-party notices preserved in NOTICE.

WeChat Group QR Code

For the Chinese-speaking community, the WeChat group QR code is provided in the repository.
Image: WeChat QR code
Important note: The QR code rotates weekly and is valid until the stated date. If you encounter expiration, reach out via GitHub Issues to refresh the image.

Star History and Popularity

A visual star-history chart traces the project’s popularity over time.
Image: Star history
This dynamic visualization provides a tangible measure of community interest and ongoing development momentum.

Community Voice: Why It Matters

The project emphasizes privacy and autonomy. By keeping design processing local, it reduces dependency on cloud services and subscription models.
It champions openness and collaboration. The MIT license and open development model invite researchers, designers, and developers to contribute improvements, plugins, and integrations.
It fosters practical design production. Not just prompts, but production-ready assets—exported files ready for real-world usage.

Citing Open CoDesign in Your Work

If you reference Open CoDesign in papers or articles, a recommended citation is provided in the repository’s CITATION.cff and the BibTeX entry:
bibtex entry is provided in the README, including author, title, year, and URL.

Caveats and Practical Tips

Dependency on local hardware: because the tool runs locally and supports multiple providers, your experience depends on your machine’s resources and available models.
Verifying builds: for users seeking verified builds, compiling from source is an option. See CONTRIBUTING.md for details.
macOS Gatekeeper notes: early builds may require running commands to bypass Gatekeeper on macOS Sequoia 15+ systems. The one-liner xattr approach is included in the quickstart notes for a smoother install.

Accessibility and Language Support

The UI is designed to be bilingual with English and 简体中文 (Simplified Chinese) UI options, with a live toggle between languages. This makes the tool accessible to a broader audience and supports teams with mixed-language workflows.

Notes on Documentation and Release History

The main repository pages include a changelog and a detailed release history. The “What’s New” section highlights changes across versions, including v0.2.0 preparations and v0.1.4 release notes.
Documentation is accessible via the Docs section of the project website, a Quickstart guide, and a changelog with entry-by-entry notes.

Conclusion: A New Path for Desktop AI Design Tools

Open CoDesign represents a thoughtful convergence of prompts, models, and production-ready design artifacts, all operating locally. It’s designed for designers and developers who want speed, privacy, and control without sacrificing quality or collaboration. With an emphasis on multi-model flexibility, a rich kit of design skills, and a transparent, live-diagnostic workflow, it invites users to shape their own design processes rather than conform to a single cloud-based paradigm.

Gallery: Images from the Input

Hero showcase: Open CoDesign — prompt on the left, live artifact on the right
Comment mode: Click, pin, and rewrite a region
AI tweaks: Sliders for color and typography
Hub of designs: Your iterations, kept locally
Live agent activity: Todos and tool calls streaming
WeChat community: QR code for the Chinese-speaking group

Appendix: Quick Reference Commands and Links

Quickstart install (one-liner):
Windows: winget install OpenCoworkAI.OpenCoDesign
macOS: brew install --cask opencoworkai/tap/open-codesign
Quickstart manual install references:
macOS arm64: open-codesign-*-arm64.dmg
macOS x64: open-codesign-*-x64.dmg
Windows x64: open-codesign-*-x64-setup.exe
Windows ARM64: open-codesign-*-arm64-setup.exe
Linux AppImage: open-codesign-*-x64.AppImage
Linux Debian/Ubuntu: open-codesign-*-x64.deb
Linux Fedora/RHEL: open-codesign-*-x64.rpm
API key adding:
Anthropic: sk-ant-…
OpenAI: sk-…
Gemini
OpenAI-compatible relays (OpenRouter, SiliconFlow, Ollama)
Documentation and community:
Website: https://opencoworkai.github.io/open-codesign/
Quickstart: #quickstart
Discussions: https://github.com/OpenCoworkAI/open-codesign/discussions
Contributing: ./CONTRIBUTING.md
Security: ./SECURITY.md

Coda: The Vision Ahead

Open CoDesign is not merely a tool; it’s a design philosophy for AI-assisted product creation. It foregrounds local-first operation, model-agnostic workflows, and an extensible design system that evolves with the user’s taste and project needs. By blending prompt-driven generation with structured design skill modules, it aims to make high-quality design artifacts accessible directly from your laptop—without mandatory cloud subscriptions or vendor lock-ins. As the project moves toward Agentic Design, you can expect a more integrated workspace, richer collaboration capabilities, and deeper control over the design process—while keeping the authenticity and privacy that local-first design tools promise.