TestDriver

TestDriver is an AI QA engineer for GitHub that runs each pull request in a real desktop sandbox, generates end-to-end Vitest tests, self-heals when the UI changes, and opens a PR with the new coverage.

TestDriver is an AI-driven end-to-end testing service built around GitHub pull requests. Instead of only reading a diff, it launches the application in a real desktop sandbox, interacts with the interface the way a person would, and reports whether the changed behavior actually works. Tagging @testdriverai in any GitHub repo triggers the agent, which then writes UI tests and flags regressions before code is merged. The workflow starts with a vision agent that reads the pull request diff to understand what changed. It then generates tests from that content, clicking, typing and reading the screen without relying on selectors or DOM access. Once a test passes, the agent opens a separate pull request containing the generated test file, so the team reviews and approves coverage like any other code change. The tests are plain Vitest files, which means they run locally and in existing CI pipelines across web, desktop and browser extension targets on real virtual machines. Maintenance is handled through self-healing. When a cached element no longer matches because the layout or styling moved, the tool asks the AI to locate it again, updates the cache and continues. For failures, the dashboard offers video replays, network logs with headers and payloads, CPU and memory metrics, and a step-by-step action log. Pass rates, flaky tests and failure trends are tracked over time, and results can be exported as JUnit XML for other reporters. The product suits engineering teams that host code on GitHub and want automated QA on every change, particularly those testing apps they cannot easily instrument: third-party web apps, Chrome extensions, native desktop applications, VS Code extensions, canvas and video content, and OAuth flows. Compared with AI code review tools, it executes the app rather than inspecting text. Compared with Playwright-based agents or screenshot diffing, it covers desktop and mobile-style flows and adapts to UI changes. Pricing is per seat plus metered testing minutes, with a custom pilot for enterprise needs.