← Back to Discover

agent-qa

Open-source QA agent that runs tests written in plain language for web and mobile, recovers from broken steps, and builds memory from past runs.

Testing & QABrowser automation
Visit website ↗

What it does

agent-qa is a test harness where tests are written in natural language rather than selectors. Agents work from the roles, labels and screen state they can see, and when a click, fill or select fails they re-observe the interface and try another route within the same run. Each run feeds an execution memory built from product, suite and test observations, which is added as context to later runs and also records steps that had to be healed. An action cache reuses validated plans across similar runs to cut planner work and token use. Hooks in Node, Bun, Python or Bash run in sandboxed Docker containers to set up data or call APIs during a test. It is used through a dashboard and CLI, and exposes MCP and skills for coding agents. You choose your own model: OpenAI- and Anthropic-compatible endpoints, Gemini, local or open-source models, and Codex or Claude Code subscriptions. Tests, configs, hooks and memory live in version control as code.

Capabilities

  • Tests written in natural language for web and mobile
  • Self-healing execution that re-observes the UI after a failed action
  • Execution memory that carries lessons into later runs
  • Action cache to reduce planner work, tokens and runtime
  • Sandboxed Docker hooks in Node, Bun, Python or Bash
  • Dashboard, CLI, MCP and skills, with a model of your choice

When to use it

  • End-to-end testing where UI changes would break selector-based tests
  • Letting a coding agent run and check tests through MCP
  • Teams that want tests and memory versioned and reviewable as code