Neura Market
vostride/agent-qa logo

vostride/agent-qa

Free

自我改进的 Agentic QA 测试工具，通过自然语言编写 Web 和移动端测试，并用执行记忆和自修复动作排查回归。

FreeFree tier
Type
Open Source
Company
vostride
Visit Website GitHub (40 stars)

About vostride/agent-qa

agent-qa is a self-improving, Agentic QA harness with memory designed for software teams. It allows users to write web and mobile tests in natural language, which agents execute using visible roles, labels, and screen state. The system features self-healing test execution: when a sub-action like click or fill fails, it re-observes the UI and attempts alternative paths within the same run. It builds execution memory from product, suite, and test observations across runs, curating knowledge from healed steps to avoid future mistakes. agent-qa offers a polished dashboard and CLI for developers, along with MCP and skills for coding agents. It includes a smart action cache that reuses validated plans to reduce runtime and token usage, and supports sandboxed hooks (Node, Bun, Python, Bash) in isolated Docker containers for environment setup and data seeding. The entire test suite, including configs, hooks, and memory, is version-controlled and open source, enabling review and sharing. Users can bring their own LLM via OpenAI-, Anthropic-compatible endpoints, Gemini, local models, or subscriptions like Codex and Claude Code.

Key Features

Write tests in natural language for web and mobile using actions and assertions based on visible roles, labels, and screen state
Self-healing test execution: re-observes UI and tries alternative paths when sub-actions fail, recovering from UI drift
Self-improves with Memory: builds execution memory from past runs and curates knowledge from healed steps to improve future tests
Built for humans and machines: provides a polished dashboard, CLI, MCP, and skills for coding agents
Smart Cache: reuses validated action plans across similar test runs to reduce planner work, token usage, and runtime overhead
Sandboxed hooks: runs Node, Bun, Python, or Bash hooks in isolated Docker containers for setup, teardown, and data seeding
Open source: all tests, configs, hooks, memory, and suite logic are version-controlled for review, reuse, and sharing
Bring your own LLM: supports OpenAI, Anthropic, Gemini, local/open-source models, Codex, and Claude Code via compatible endpoints

Pros & Cons

Pros
  • Tests are written in natural language, reducing the learning curve and maintenance overhead
  • Self-healing execution minimizes test flakiness and avoids failures due to minor UI changes
  • Execution memory improves test reliability over time by learning from past runs and healed steps
  • Smart cache reduces runtime and token usage, making test runs faster and cheaper
  • Supports multiple LLM providers, including local models, for flexibility and cost control
  • Open source with version-controlled configurations enables transparency, review, and collaboration
Cons
  • Requires Docker to run sandboxed hooks, adding a dependency for test setup
  • Initial installation and configuration may be complex for teams unfamiliar with Node.js and Docker
  • Relies on LLM API calls, which can incur costs depending on the provider and usage volume
  • Self-healing may not always succeed on complex UI changes or interactions requiring deep context

Best For

Writing and maintaining UI regression tests for web and mobile applications using natural languageCatching regressions before shipping by learning from test execution history and adapting to UI changesAutomating QA workflows with sandboxed hooks for environment setup, API calls, fixture seeding, and state cleanupCollaborative test development with version-controlled test suites that can be diffed, reviewed, and shared across teamsRunning tests with various LLM backends for flexibility in cost, performance, and privacy

FAQ

What is agent-qa?
agent-qa is a self-improving, open-source QA harness that uses AI agents to write and execute web and mobile tests written in natural language. It learns from past executions, adapts to UI changes, and catches regressions.
How does self-healing work?
When a sub-action like click, fill, or select fails during a test run, agent-qa re-observes the current UI state and attempts a different path to accomplish the same goal within the same run, recovering from UI drift or flaky interactions.
What LLMs does it support?
agent-qa supports any LLM accessible via OpenAI- and Anthropic-compatible endpoints, including Gemini, local or open-source models, and subscriptions like Codex and Claude Code.
Do I need Docker?
Docker is required only if you plan to use sandboxed hooks (Node, Bun, Python, Bash) for test setup, teardown, or seeding data. The core test execution does not require Docker.
Is agent-qa free?
Yes, agent-qa is open source and free to use. You only pay for the LLM API calls if you use a commercial provider.