Arbigent

Visit Website

Introduction

Arbigent (Arbiter-Agent) is an open-source AI agent testing framework for modern applications, maintained by takahirom on GitHub under the Apache-2.0 licence. The README describes it as a way to get “from zero to AI agent testing in minutes” for Android, iOS and web apps.

Its main idea is scenario breakdown: complex tasks are split into smaller, dependent scenarios so that the AI agent stays on track. The author’s motivation is that AI agents often open other apps or click the wrong button when a task is too large.

Features

1. Scenario breakdown: Split complex tasks into dependent scenarios to keep agents reliable.
2. UI and code: An intuitive UI for building scenarios and a code interface for engineers.
3. Multi-platform: Android, iOS and web targets.
4. Handles UI churn: Aimed at A/B tests, changed tutorials, unexpected dialogs and dynamic content that break classic tests.
5. Model choice: Works with configured AI model providers; see the README for supported options.
6. Project files: Scenarios are saved as project files you can version.

Pros

1. Free and open: Apache-2.0, no account.
2. Pragmatic design: Scenario decomposition is a sensible answer to agent drift.
3. Mobile first: Built with Android and iOS in mind.
4. Easy start: The README promises setup in minutes.
5. Active: Commits within the last two months at review time.

Cons

1. Single-maintainer project: Community and support are smaller than vendor tools.
2. Bring your own model: You manage and pay for model access.
3. Website is GitHub: There is no separate docs site; documentation lives in the README.
4. Impersonation warning: The README warns of a spam account posing as Arbigent; use the official links.
5. Not a device cloud: You supply emulators or devices.

Usage and Links (Example)

Follow the README: download the UI app or add the library, connect an emulator, simulator or browser, and create scenarios. Break a large goal into small dependent scenarios, for example “open app”, then “complete onboarding”, then “search for an item”, and run them.

Resources:

• GitHub repository

Pricing

Arbigent is free and open source under Apache-2.0. Model usage is paid to your chosen provider.

Recommended for?

1. Android and iOS developers experimenting with AI agent tests.
2. Teams fighting UI churn such as A/B tests and dynamic content.
3. Engineers who want open-source tooling they can extend.
4. Not a fit for teams needing a vendor SLA.

Conclusion

Arbigent is a thoughtful open-source framework that makes AI agent tests more dependable by breaking tasks into scenarios.

It suits mobile teams comfortable with community tooling and their own model keys.