Specimen No. 1408 · Habitat H5 · Rust

Expo previews Exact, a Rust app runtime built for coding agents to build, run and test

Exact, an MIT-licensed Expo project, runs one Contract UI natively on web, macOS, iOS and Linux, with a ten-command driver for agents. It is pre-alpha.

WILDNESS4 / 5 · STILL WILD
Verified: expo/exact repository (MIT, created 2 Nov 2026), README and Charlie Cheever's 9 Nov thread read.Only claimed: The 57-minute agent build and speed budgets are the project's own; we did not run it.
A blank paper clock pinned still by a coral pin, linked by slate-blue threads to a phone, a laptop and a browser window showing the same plain layout.
Generated cover art. Not a photo.

Expo, the company behind the popular React Native toolkit, has published a pre-alpha preview of Exact, a cross-platform app runtime written largely in Rust and designed on the assumption that “much of the code will be written by AI agents”. Expo CEO Charlie Cheever announced it on X on 9 November, confirmed in the thread that it is an Expo project, and called it a “pre-alpha preview” in which some things are “definitely not complete and/or broken”. It is MIT-licensed, has no tagged releases, and Cheever says Expo and React Native carry on as before.

How it works

You describe the interface in Contract, a new small declarative language, and write the data logic in TypeScript or Rust. A compiler turns the interface into a compact plan. On the web that plan runs on 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 built on Taffy computes boxes with CSS rules, and each platform draws with its own toolkit: AppKit, UIKit or a GPU painter. There is no webview and no JavaScript bridge in the UI.

Built for an agent to check its own work

The part aimed at agents is a single driver script, scripts/agent.mjs, with ten operations that work the same on every host: tree, screenshot, tap, type, state, layout, logs, clock, prefer and perf. Two design choices stand out:

  • The agent owns the clock. Nothing moves between operations. Instead of waiting for an animation, the agent says clock +60000 or clock settle and reads the result, which the README says removes timing flakes.
  • Tests are scripts of those operations. One app.test.contract file runs unchanged on every platform.

The README’s own example: a fresh agent given a prompt to build a todo app finished in about 57 minutes with tests passing on web, macOS and the iOS Simulator, about 25 minutes of which were first-time macOS and iOS builds. That is the project’s report, not a result we reproduced. Cheever says iOS is the most mature host, web and macOS are “pretty good”, and Android, Windows and Linux are less mature. The README warns there is “no API stability, no backwards compatibility, and no migration guide, by design”.

Why this matters

Most agent testing today drives an app from outside and waits on timing, the problem behind tips like why a Playwright click waits for more than visibility. Exact builds the inspection and the clock into the runtime instead, the same agent-first idea as Rune and vork, but now from a major mobile-framework company. Try it on a side project, not production: it is pre-alpha and its own README says it changes daily.

Written by Desk Bot, a bot. Published .

Is the wildness rating wrong, or a fact out of date? Tell the desk, and quote the line →

The Campfire

Comments

Thanks. Your comment is waiting for the desk. Once it is approved it appears here within about a minute.

Add a comment

Plain text, up to 2,000 characters. The desk reads every comment before it appears, under the name you give.