Repolode
Submit
takahirom

takahirom/arbigent

An AI agent testing framework that runs natural-language test goals on Android, iOS, and Web apps via a UI or CLI.

View on GitHub
Stars
654
Forks
63
Open issues
27
Last commit
last month

What is it?

Arbigent is a testing framework where an AI agent operates a real or emulated device or browser to complete goals written in plain language, such as "Complete the tutorial". Tests are broken into dependent scenarios (for example, login before search), so a complex flow can be split into smaller steps that the agent handles one at a time. Scenarios can be built in a visual UI by QA staff, saved as YAML, and executed from the command line or in CI.

The project supports Android, iOS (including physical iPhones via XCTest), and Web, and works with OpenAI, Gemini, Azure OpenAI, Anthropic, and OpenAI-compatible APIs such as Ollama. It can reuse existing Maestro YAML flows for setup steps, connect MCP servers as external tools, and accept app-provided hints through accessibility labels. Replay with fallback can reuse recorded actions from a passing run to reduce AI cost.

The framework is written in Kotlin and licensed under Apache-2.0. Its README notes that speed and per-run AI cost are current limitations.

Key features

  • 01Runs natural-language test goals on Android, iOS, and Web with an AI agent.
  • 02Splits complex tasks into dependent scenarios such as login before search.
  • 03Creates scenarios in a visual UI and runs saved YAML from the CLI or CI.
  • 04Supports OpenAI, Gemini, Azure OpenAI, Anthropic, and OpenAI-compatible providers.
  • 05Reuses Maestro YAML flows as setup steps and connects MCP servers as tools.
  • 06Replays recorded actions from passing runs to reduce AI cost.

Use cases

  • Automating end-to-end UI tests for mobile apps where layouts or dialogs change often.
  • Letting QA engineers write test scenarios in plain language without coding.
  • Running sharded AI-driven test suites in GitHub Actions on Android emulators.
  • Testing TV or D-pad driven interfaces that are hard to script with fixed selectors.

Quick start

  1. 1

    Download the Arbigent UI binary from the Releases page

  2. 2

    Connect your Android, iOS, or Web device to your PC

  3. 3

    In the UI, select the device and enter your AI provider API key

  4. 4

    Define a scenario by entering a natural-language goal for the AI agent

Things to know

  • Requires a paid AI provider API key, and each step incurs AI cost.
  • Speed is limited by AI latency, so it is much slower than traditional tests.
  • Needs a connected emulator or physical device for testing.

In collections

Open-source AI agent frameworks & appsAndroid apps & phone tools

Topics

#ai-agent#mobile-testing#android#ios#web-testing#kotlin#llm#ui-testing#agentic#ai

FAQ

What is arbigent?
An AI agent testing framework that runs natural-language test goals on Android, iOS, and Web apps via a UI or CLI. Arbigent is a testing framework where an AI agent operates a real or emulated device or browser to complete goals written in plain language, such as "Complete the tutorial". Tests are broken into dependent scenarios (for example, login before search), so a complex flow can be split into smaller steps that the agent handles one at a time. Scenarios can be built in a visual UI by QA staff, saved as YAML, and executed from the command line or in CI.
Is arbigent free to use? What license does it have?
arbigent is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is arbigent still maintained?
Updated within 6 months. The last commit to arbigent was on Oct 2, 2026.
What can I use arbigent for?
Automating end-to-end UI tests for mobile apps where layouts or dialogs change often. Letting QA engineers write test scenarios in plain language without coding. Running sharded AI-driven test suites in GitHub Actions on Android emulators. Testing TV or D-pad driven interfaces that are hard to script with fixed selectors.
How do I install arbigent?
Download the Arbigent UI binary from the Releases page Connect your Android, iOS, or Web device to your PC In the UI, select the device and enter your AI provider API key Define a scenario by entering a natural-language goal for the AI agent
Where can I get arbigent?
The source code and installation instructions are on GitHub: https://github.com/takahirom/arbigent

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects