arbigent  by takahirom

AI agent testing framework for modern apps

created 7 months ago
366 stars

Top 78.1% on sourcepulse

GitHubView on GitHub
Project Summary

Arbigent is an AI-powered testing framework designed for modern Android, iOS, and Web applications. It addresses the brittleness of traditional UI testing by leveraging AI agents, offering a hybrid UI and code-based approach for scenario creation and execution. This framework aims to make AI agent testing practical, scalable, and accessible to both QA engineers and developers.

How It Works

Arbigent breaks down complex testing goals into smaller, dependent scenarios, managed by an orchestrator. It utilizes an interceptor pattern for extensive customization, allowing users to integrate various AI providers and adapt to different operating systems and form factors. The framework supports a hybrid workflow: scenarios can be visually designed via a UI or programmatically defined in YAML files, enabling integration into CI/CD pipelines.

Quick Start & Requirements

  • Installation: Download the Arbigent UI binary from the Release page. For CLI usage, install via Homebrew: brew tap takahirom/homebrew-repo && brew install takahirom/repo/arbigent.
  • Prerequisites: Requires a connected Android/iOS device or simulator, and API keys for supported AI providers (OpenAI, Gemini, Ollama).
  • Setup Time: Described as "5 Minutes" to get started.
  • Documentation: Basic Structure, CLI Usage, GitHub Actions Sample.

Highlighted Details

  • Supports scenario dependencies for complex task decomposition.
  • Offers AI-powered image assertion for verifying AI decisions.
  • Includes stuck screen detection and self-correction mechanisms.
  • Provides initial support for Model Context Protocol (MCP) for external tool integration.

Maintenance & Community

The project is maintained by Takahiro. Community contributions are welcomed. Links to social media are provided: X and X.

Licensing & Compatibility

The project is open-source, allowing free use, modification, and distribution. Specific license details are not explicitly stated in the README, but the open-source nature implies broad compatibility for commercial use and closed-source linking.

Limitations & Caveats

The YAML format for project files and the code interface are noted as being under development and subject to change. The framework's speed is dependent on AI model performance and real-time UI interaction, making it slower than traditional tests. Resource utilization (device and AI costs) can be significant.

Health Check
Last commit

2 days ago

Responsiveness

1 day

Pull Requests (30d)
20
Issues (30d)
3
Star History
110 stars in the last 90 days

Explore Similar Projects

Feedback? Help us improve.