Bug0 QA Agent pairs AI-driven browser tests with cached Playwright replay

As reported on November 19, 2026, Linearbytes had released Bug0 QA Agent, a Playwright library for browser regression testing. It lets developers write test steps in plain English, then uses an AI agent to connect those instructions to actions in the user interface.

On an initial run, the agent examines the page’s accessibility tree and screenshots to carry out a step. When a single-action step succeeds, the framework converts it into a browser action and caches it in Redis. Later runs can replay that action through Playwright without another round of language-model inference. If a UI change breaks the replay, Bug0 clears the cached action and invokes the agent again. Complex, multi-action instructions are not cached and run through the agent each time.

Assertions take a different path: they are evaluated by visual models on every run. Claude and Gemini assess each assertion independently, with a third model used if they disagree. The framework also supports disposable-email testing through adapters for services including mailsink and Mailinator, allowing tests to extract one-time codes or magic links from emails.

The approach aims to limit the latency and API costs of invoking a model at every browser step while making tests less dependent on selectors that can change without altering the intended behaviour. It does not eliminate the broader challenges of autonomous testing. A paper by Chevrot and colleagues, published in April 2025, reported that its best autonomous testing architecture agreed with professional human testers on only about 60% of test runs; that figure is not a performance result for Bug0.

Although Bug0 was described as open source, its stated license is Functional Source License 1.1 under the Av2 future-license option. It is a source-available, not an open-source, license and imposes a non-compete restriction during the first two years.