A new open-source project is betting that the awkward part of AI-driven software testing is not getting a model to click buttons — it is making the result cheap, deterministic and worth trusting.
The framework, called e2e and built by the agentic testing platform TesterArmy, describes itself as a "next generation" end-to-end testing framework for web and mobile apps. A test mixes two kinds of steps. In an agent step, you describe a goal in plain language and an agent drives the app to reach it. In an assertion step, you check the result with ordinary locators in the same test file — the pattern Playwright users already know.
The design choice that matters is what happens next. When an agent step is verified by a later assertion, the framework records the actions the agent took, and the following run replays those actions with no model calls at all — until the application changes underneath it. Tests that contain no agent steps need no model in the first place, and teams can bring their own subscription, API key or a local model.
Installation is a single command that asks for an engine, web or mobile, and a model provider, then writes a config and an example test. The project ships as a set of packages: the core e2e SDK, runner and CLI; a web engine covering Chromium, Firefox and WebKit through Playwright; a mobile engine for iOS simulators and Android emulators through agent-device; a GitHub reporter that posts results as a pull request comment; hosted-browser and hosted-simulator packages; and decision-model executors for bounded semantic actions and assertions. The CLI ships its documentation inside the npm package, so coding agents can read it offline.
The repository is Apache-2.0 licensed, is explicit that APIs and configuration can still change before 1.0, and sends anonymous usage telemetry — which commands and engines run, where runs fail — that can be disabled with an environment variable or a CLI flag. It drew more than 1,700 stars in a single day, making it the fastest-rising AI testing project on GitHub trending.
It arrives into a crowded field. The past year has produced a steady stream of open-source "agent skills" libraries that try to give coding agents engineering discipline — TDD loops, debugging protocols, code review — and e2e is a telling next step: taking the same agentic premise and pointing it at the one place software teams traditionally trust least, the test suite.
