Easel: A Codex-First macOS Workspace for AI-Assisted Design

What Easel Actually Is

Easel is an open-source, Codex-first macOS workspace for AI-assisted product design and frontend iteration. Built by James Rochabrun as a native SwiftUI app, it bundles several things that usually live in separate windows: a local project library, design-system setup, a Codex chat interface, a live web preview with a point-and-click inspector, and slide tooling.

The author describes it as "Claude Design for Codex"

esstyle design surface, but pointed at OpenAI's Codex SDK rather than Anthropic's tools. At the time of writing the repo sits at a modest 121 stars and 6 forks, so this is early-stage software, not a polished commercial product. It ships as an Xcode project you build yourself, which means you'll need a Mac and a willingness to work with developer tooling.

The core loop is straightforward: you set up a design system, chat with a Codex-powered agent to generate or modify frontend code, and watch the result render in a live preview. The point-and-click inspector lets you select elements directly in that preview, which is the part that distinguishes it from a plain chat-to-code workflow.

How the Design System Setup Works

The most practical idea in Easel is how it handles design systems. You have three options: import one from Figma, bring your own, or have the agent generate one as a `design.md` file. That last option is worth pausing on, because a markdown-based design system is something an AI agent can actually read, reason about, and apply consistently across a project.

If you've ever tried to get a model to produce on-brand UI, you know the failure mode: each generation drifts in spacing, color, and typography because there's no shared source of truth. A committed `design.md` gives the agent a reference it can keep coming back to. That's a sensible pattern, and it's portable beyond Easel itself \u2014 a structured design doc is useful anywhere you're prompting models to build interfaces.

From there, the workspace is meant for building prototypes and slide decks. The combination of a live preview and an inspector suggests the intended audience is designers and frontend-minded builders who want to iterate visually rather than reading through generated code diffs. That's a reasonable niche, since a lot of AI coding tools assume you're comfortable living in a terminal or editor.

Where It Fits and What to Watch

Easel is Codex-first today, with Claude and other AI providers plus local models listed as planned for future releases. That roadmap matters: right now you're committing to OpenAI's Codex SDK, so if you prefer a different provider you'll be waiting on multi-provider support that isn't shipped yet. Treat the multi-model promise as a direction, not a current feature.

Compared to the broader landscape, Easel sits in a different category than general code agents and the design-focused features inside chat tools. It's narrower and more opinionated: a native Mac app focused on the design-to-frontend loop, with a visual inspector as the centerpiece. If you mostly work in the browser or want cross-platform support, this isn't it. If you live on macOS and want a dedicated surface for AI-assisted UI iteration, it's worth a look.

For automation-minded readers, the relevance is indirect but real. Easel itself isn't a workflow engine \u2014 you won't be wiring it into Zapier, Make, or n8n. But the patterns it leans on are the ones that make AI design work reliably at scale: a machine-readable design system, a tight feedback loop between generation and preview, and a consistent agent context. Those ideas transfer to any pipeline where you're asking a model from OpenAI or another provider to produce frontend artifacts. If you're building internal tooling that generates UI or marketing decks, Easel is a useful reference implementation even if you don't adopt the app directly.

A few caveats to keep in mind. As an early open-source project with a small contributor base, expect rough edges, breaking changes, and incomplete documentation. You'll need to build it from source via Xcode, and you'll need Codex access to use the core chat. There are no published benchmarks or formal release commitments here \u2014 just a working app and a clear roadmap. That's normal for projects at this stage, but it means Easel is better suited to experimentation than to anything you'd put in a production design workflow today.

The honest summary: Easel is a thoughtful, focused take on AI-assisted design that gets the design-system-as-context idea right and packages the generate-preview-inspect loop into one native app. Whether it becomes a daily tool depends on how the multi-provider support and stability evolve. For now it's a promising project to clone, try, and learn from.

If you want help putting AI-assisted design or frontend generation to work in your own stack, browse the provider directory to find someone who can build it with you.