On November 9, Expo co-founder Charlie Cheever posted a preview of Exact on X and open-sourced it under the MIT license at GitHub expo/exact. Exact is an "agent-native" cross-platform app framework written in Rust: the UI is described in Contract, a small declarative language, and data logic is written in TypeScript or Rust. On the web a roughly 20 KB runtime drives the real DOM; on macOS and iOS it uses AppKit and UIKit native controls; on Linux it paints with the GPU. There is no webview and no JavaScript bridge. Agents get 10 operations that work the same on every host, such as tree, screenshot, tap, type and clock, and test files run unchanged across platforms. The README sets budgets such as a 100 ms p50 cold start and says that on the RealWorld benchmark its web runtime used 26 KB of code versus 81 KB for React. Cheever stresses it is pre-alpha with APIs changing daily, and says the Expo framework is not going away.
Key Takeaways
- ✓Positioning: an agent-native app framework written in Rust, at GitHub expo/exact under the MIT license, described by its author as a pre-alpha preview
- ✓Platforms: the README lists 4 hosts: web, macOS, iOS and Linux; the author says iOS is the most mature and Android and Windows are less mature
- ✓Agent interface: 10 operations that are the same on every host (tree / screenshot / tap / type / state / layout / logs / clock / prefer / perf); the agent controls the clock, which avoids timing flakes in tests
- ✓Speed targets (project figures): 100 ms p50 cold start to first frame, and a measured 99 ms from saved edit to updated page; on RealWorld, 26 KB of web code vs 81 KB for React
- ✓Getting started: the README includes a prompt to hand to Claude Code; the project says a fresh agent built a todo app and passed tests on 3 platforms in about 57 minutes
Key Decision Metrics at a Glance
Heavy Claude Code use: compare subscription limits and API bills
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and Pain Points
A growing share of app code is written by AI agents, but mainstream cross-platform frameworks were designed for people. Agents struggle to launch an app, see the screen, tap a button and confirm the result, and animations and async work make automated tests flaky. Exact, a new project from Expo co-founder Charlie Cheever, targets that problem; he says software is now fully in the agentic era.
Architecture Highlights
The UI is written in Contract, a declarative language, and the compiler turns it into a compact plan. On the web a roughly 20 KB runtime drives the real DOM. On macOS, iOS and Linux a Rust runner executes the same plan, a Rust layout kernel applies CSS rules, and AppKit, UIKit or a GPU painter draws the result, with no webview and no JavaScript bridge. TypeScript is compiled to Hermes bytecode at build time, so no app JavaScript runs before the first pixel. Agents drive every host through 10 shared operations and can advance time with the clock command. Cheever also mentions a built-in game engine targeting Metal, Vulkan and WebGPU.
Benchmarks
There are no third-party evaluations yet; all figures come from the project README. The cold start target is 100 ms p50 to the first interactive frame. On a fresh app, a saved edit reached the page in 99 ms. On the RealWorld (Conduit) benchmark, the Exact runtime used 26 KB of code versus 81 KB for React and was ready about 300 ms before the React SSR build finished hydrating. On X, Cheever says it starts faster than SwiftUI and scrolls at a steady 120 fps, but he has not published a full test method.
Getting Started
You need a Mac with Xcode, Rust, Bun and Chrome. Clone expo/exact, run bun scripts/exact.mjs setup once, then create a project with exact new, or start from a template in expo/exact-templates. The README includes a prompt you can hand straight to Claude Code. The project is pre-alpha with no API stability or migration guide, so it suits experiments and feedback rather than production use for now.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.