What is e2e?
e2e, from the testing company TesterArmy, is a test framework in which a step can be a goal in natural language, such as "upgrade the workspace to the Pro plan", that an agent carries out in the app. The same test can mix those agent steps with ordinary locators and assertions. Web tests run on Chromium, Firefox and WebKit through Playwright; mobile tests run on iOS simulators and Android emulators.
The replay cache is what separates it from a browser agent. When a later check confirms that an agent step worked, the runner records its actions, and the next run replays them with no model call until the screen no longer matches, at which point the agent takes over. Tests without agent steps need no model at all, and models come from your own ChatGPT, GitHub Copilot, OpenCode or SuperGrok subscription, an API key, or a local OpenAI-compatible server.
It is pre-1.0, and the README warns that APIs and config can change between minor releases.
Is e2e worth it?
e2e solves the problem that sinks most AI testing tools: paying a model to rediscover the same clicks on every run. The repository, created in July 2026, has about 12,100 GitHub stars; the npm package shipped e2e@0.21.3 on 23 October 2026 and recorded about 380,000 downloads in the week to 26 October.
The cache rules are strict, which is the point. An agent step is recorded only when a later locator assertion or agent check passes, so an unverified step runs live every time and a step whose replay fails is evicted. agent.assert, agent.waitFor and agent.extract always call the model, so a suite still has a running cost, and the run summary counts replayed, handed-off and missed steps. Recordings are JSON files under .e2e/cache, and CI defaults to read-only so a pipeline does not rewrite them.
The cache key is brittle by design. Renaming a test, rewording a goal, moving the file or switching engine all cause a miss; timestamps and fresh email addresses miss on every run unless wrapped in unique(). Changing the model does not, so a new model's behaviour has to be checked with --no-cache. Text typed into the app is stored as typed, and the docs warn to review recordings before sharing them.
Two defaults are worth knowing. The CLI sends anonymous usage telemetry unless you opt out, and several of the open issues concern replay-cache edge cases. Against Stagehand or Browser Use, which drive a browser for any job, e2e is narrower and better for the one job of a test suite that has to pass on every pull request.
Where it earns its place
A four-engineer SaaS team with a flaky Playwright suite
Stop rewriting selectors every time the billing page changes
Replace brittle click sequences with agent.act goals followed by a locator assertion; once verified, steps replay without the model until the page changes, then the agent repairs and re-records. Model spend falls to the assertions and the changed steps.
A mobile lead at a React Native agency
Run the same checkout test on iOS and Android for each client build
Use the mobile engine on local simulators, or hosted devices through Limrun, EAS or TestMu AI, with the GitHub reporter posting results on the pull request. Free apart from model and device costs.
A QA manager without engineers to spare
Get regression tests running on every pull request without building the framework
TesterArmy's hosted platform runs natural-language tests on each pull request or a schedule: Hobby is $99 a month for 250 runs and Startup $299 for 1,000, each run capped at 20 minutes. The open-source runner is the self-managed route.
Before you commit
- Setup
- `npx e2e init` asks for web or mobile and a model provider, then writes a config and an example test that makes no model calls. Needs Node.js 24.8+ (or 22.22.3+ on Node 22); Windows only through WSL; mobile needs Xcode or the Android SDK. A pasted prompt lets Claude Code, Codex or Cursor do the setup.
- Integrations
- Playwright for Chromium, Firefox and WebKit; agent-device for iOS and Android; hosted browsers and devices through Kernel, Lightpanda, smol machines, Limrun, EAS and TestMu AI; a GitHub pull-request reporter; an MCP server; plugins for Claude Code, Codex and Cursor; any AI SDK provider, OpenRouter or Vercel AI Gateway.
- Data and privacy
- Model input is the redacted screen, the instruction and requested screenshots; OpenAI Responses calls carry store: false. Cache entries hold actions and typed text but no prompts, screenshots or secret values. The CLI sends anonymous telemetry by default; `E2E_TELEMETRY_DISABLED=1` or `DO_NOT_TRACK=1` turns it off.
- Cost at scale
- The runner is free; cost is model tokens for agent assertions and uncached steps, which a cached suite keeps small, plus CI or hosted device time. TesterArmy's hosted plans run from $99 a month for 250 runs to $299 for 1,000, with Enterprise (SSO, self-hosting) custom.
- Limits
- Pre-1.0 with releases most weeks; 47 open issues, several on replay-cache edge cases; agent checks and extracts never replay; a goal that changes nothing visible is never cached.
What's good
- Verified agent steps replay with no model calls
- Mixes plain-English goals with Playwright-style assertions
- Web and iOS/Android in one runner
- Runs on ChatGPT, Copilot or local models
What's not
- Pre-1.0; APIs change between minor releases
- Renaming a test or goal discards its cache
- Assertions by the agent always cost a model call
- Telemetry on by default
Best for
Product teams that want readable end-to-end tests across web and mobile without hand-maintaining every selector.
e2e alternatives
e2e: common questions
- Is e2e worth it?
- e2e is the first agentic test framework that treats model calls as something to remove, and its replay cache is what makes plain-English tests affordable in CI. We rate it 4.0 out of 5.
- How much does e2e cost?
- Free under Apache-2.0; you pay for your own model. TesterArmy's hosted platform is separate: Free (5 runs), Hobby $99/month (250 runs), Startup $299/month (1,000 runs), Enterprise custom.
- What are the downsides of e2e?
- Pre-1.0; APIs change between minor releases; renaming a test or goal discards its cache; assertions by the agent always cost a model call; telemetry on by default.
- Who is e2e best for?
- Product teams that want readable end-to-end tests across web and mobile without hand-maintaining every selector.
- How much does e2e cost for a team or in production?
- The runner is free; cost is model tokens for agent assertions and uncached steps, which a cached suite keeps small, plus CI or hosted device time. TesterArmy's hosted plans run from $99 a month for 250 runs to $299 for 1,000, with Enterprise (SSO, self-hosting) custom.
- Is e2e safe for company data?
- Model input is the redacted screen, the instruction and requested screenshots; OpenAI Responses calls carry store: false. Cache entries hold actions and typed text but no prompts, screenshots or secret values. The CLI sends anonymous telemetry by default; `E2E_TELEMETRY_DISABLED=1` or `DO_NOT_TRACK=1` turns it off.
- What are the best e2e alternatives?
- The closest alternatives we have reviewed are Stagehand, Browser Use and Skyvern.
Answers checked Oct 28, 2026.
Make e2e? Show this rating on your site
