Expo Exact: one app on web, macOS, iOS and Linux, built by agents
Added NewCopyable prompt below
Published Last verified Released
Our verdict
The first cross-platform framework we've seen designed around the agent closing its own loop: the repo ships the scripts an agent uses to build the app, see it and pass one test file on web, macOS and the iOS Simulator, plus a ready-made Claude Code prompt. Expo co-founder Charlie Cheever previewed it on November 9, 2026 and it gained about 1,070 GitHub stars in a day (about 2,400 total). Expo calls it pre-alpha, says it is sharing it for feedback, and the README promises no API stability; Expo's speed claims (120fps scrolling, near-instant startup) are its own. It is not React Native, so treat it as a playground for new side projects, not a port target for a production app yet. We have not run it ourselves.
- Who it's for
- Builders who let Claude Code or Codex write their apps and want one codebase that runs as a real native Mac, iPhone, Linux and web app, and who are comfortable on a fast-moving pre-1.0 stack
- Leverage
- An agent that can build, look at and test its own app on several platforms in one loop, instead of you running each simulator and reporting back what broke
- Effort
- High
- Date added
What it does
Exact (expo/exact, MIT) is a cross-platform application runtime. You describe the interface in Contract, a small declarative language, and write data logic in TypeScript or Rust. The Contract compiler turns the interface into a compact plan: on the web it becomes one ES module over a runtime of about 20 KB that drives the real DOM; on macOS, iOS and Linux a Rust runner executes the same plan, a Rust layout kernel computes boxes with CSS rules, and each platform draws with its own tools (AppKit, UIKit, or a GPU painter on Linux), with no webview and no JavaScript bridge in the UI. Layout covers CSS block, flex and grid; there are virtualized lists, inputs, a Markdown editor, images, video, Canvas 2D and motion. The agent driver (scripts/agent.mjs) runs an app.test.contract file on each host and captures screenshots, and existing Swift apps can embed Exact screens. Expo says Android apps can already be built too. In Expo's own test from the README, a fresh agent built a todo app with tests passing on web, macOS and iOS in about 57 minutes, about 25 of them spent on the first native builds.
How builders use it
- On a Mac with Xcode, Rust, Bun and Chrome installed, paste the README's prompt into Claude Code and let it clone the repo, run setup and scaffold an app with exact new.
- Keep the screen in Contract and the data in app.ts, and have the agent write an app.test.contract that it must pass on web, macOS and the iOS Simulator before calling a task done.
- Use it for a small tool or side project first: builder Quoc Huy (@qhuy_uxheavy) posted an Excalidraw-style app with native iOS controls that he says took about 30 minutes.
- Expect daily breaking changes: pin a commit and re-read the upgrade notes in the README before pulling.
What to expect
A small app that runs natively on web, Mac, iPhone and Linux from one source, with tests your agent ran and screenshots it checked on each platform.
Copy this: Claude Code prompt (from the Exact README)
# Official source: github.com/expo/exact (MIT, pre-alpha, no API stability).
# Needs macOS with Xcode, Rust (rustup), Bun and Google Chrome. Linux works for the web loop and the Linux host.
Clone https://github.com/expo/exact and follow its README to make a new Exact
app with `exact new`. Run `bun scripts/exact.mjs setup` once first. Make a todo
list where I can add items, check them off, delete them,
and see how many are left. Put the view in Contract and keep the list in `app.ts`.
Write an `app.test.contract`, pass it on web, macOS, and the iOS Simulator with
`scripts/agent.mjs`, then open the app for me on all three.Related discoveries
See the newest picks →- ToolVibe-Coding & Prompt-to-App
Impeccable: give your coding agent a design vocabulary and an A.I.-slop detector
An Apache-2.0 design skill with 24 commands and a no-LLM detector that flags A.I.-generated UI tells in files, URLs or CI.
Effort: LowPrompt#Plugin / extension#Open source
- ToolVibe-Coding & Prompt-to-App
Floorplan: a plain-English map of your A.I.-built app
MIT plugin for Claude Code and Codex that reads your codebase and draws an interactive map: every part, file and web request named in plain English, connections traced across repos, plus a health check and a "recent work" view.
NewEffort: LowPrompt#Plugin / extension#Local / BYO model
- ToolVibe-Coding & Prompt-to-App
Modern Web Guidance: stop agents writing legacy web code
An open-source agent skill and CLI (Apache-2.0) from the Chrome team, now backed by Edge, Firefox and MDN, that feeds your coding agent short, expert-reviewed guides on modern HTML, CSS and browser APIs right before it writes front-end code, and scans an existing codebase for legacy patterns worth replacing.
NewEffort: LowPrompt#Plugin / extension#Open source
- WorkflowVibe-Coding & Prompt-to-App
Let your agent click through your app with Playwright MCP
Give the agent a real browser so it can check its own UI work.
Effort: LowPrompt#MCP#Open source
Get the best new A.I. picks in your inbox
The best new agent workflows and tools we’ve vetted, with a clear verdict on each. Short, occasional, useful.