e2e by TesterArmy
Open source AI testing framework in TypeScript: natural-language agent steps plus Playwright-style locators for web, iOS and Android end-to-end tests.
Overview
e2e is an open source end-to-end testing framework for web and mobile apps, built by Y Combinator-backed TesterArmy. Tests are TypeScript files in your repo that mix normal locators and assertions with natural-language agent steps: you describe a goal and an agent drives the app to reach it. Once a later assertion verifies an agent step, its actions are recorded and replayed with no model calls until the app changes. Web tests run on Chromium, Firefox and WebKit through Playwright; mobile tests run on iOS simulators and Android emulators. You bring your own model: a subscription, an API key, a local model, or your coding agent (Claude Code, Codex) over the Agent Client Protocol.
The Verdict
Who Should Use e2e?
Best For
- Teams that want AI-written test steps without leaving code-based, version-controlled tests
- Projects testing one product on web and on iOS/Android with a single framework
- Developers who want to reuse an existing Claude Code or Codex subscription for test runs
- CI pipelines that want PR comments and test selection based on what a pull request changed
Not Ideal For
- Teams that need a stable API today (the project is pre-1.0)
- Non-developers who want plain-language tests in a dashboard (that is the hosted TesterArmy product)
- Projects stuck on older Node.js versions
What's Great
- Agent steps and deterministic locators live in the same test, so you choose where AI is used
- Verified agent steps replay without model calls, cutting cost and flakiness on repeat runs
- Tests without agent steps need no model at all
- One framework for browsers (Playwright engines) and mobile simulators/emulators
- Bring your own model or coding agent; integrations for BrowserStack, Kernel, Lightpanda and other hosted browsers/devices
- Apache-2.0 licensed, very actively developed
Watch Out For
- Still in active development on the way to 1.0: APIs and config can change between minor releases
- Telemetry is on by default (opt-out via CLI command or environment variable)
- Requires Node.js 24.8 or newer (22.22.3 or newer on Node.js 22)
- Agent steps add model cost and latency until they are recorded
Pricing
View all features & details
Key Features
- Natural-language agent actions and agent assertions
- Record-and-replay of verified agent steps with no model calls
- Web engine on Chromium, Firefox and WebKit via Playwright
- Mobile engine for iOS simulators and Android emulators
- Decision-model filter that skips tests a PR is unlikely to affect
- GitHub reporter that posts results as a PR comment
- Run agent steps on Claude Code or Codex over the Agent Client Protocol
- Examples for Next.js, Vite, Astro, Angular, Expo, SwiftUI, Jetpack Compose, Flutter and more
Use Cases
- Writing end-to-end tests faster by describing goals instead of every click
- Cross-platform regression tests for apps shipped on web and mobile
- PR-level test runs in CI with focused test selection
- Teams evaluating the hosted TesterArmy platform with an open source starting point
How It Compares
| Feature | e2e | Playwright | TesterArmy (hosted) |
|---|---|---|---|
| Natural-language steps | Yes, mixed with code | No | Yes, plain language only |
| Tests live in your repo | Yes (TypeScript) | Yes | No (dashboard) |
| Mobile apps | iOS + Android emulators | No (web only) | iOS + Android |
| Model required | Only for agent steps | No | Included |
| Price | Free (Apache-2.0) | Free (Apache-2.0) | Free trial, then $99/mo+ |