Skip to content
GitHubDiscordSlack

Alumnium v0.15 with TypeScript client, HTTP support, and Haiku 4.5

Published by Alex Rodionov's avatar Alex Rodionov
release notes

Alumnium v0.15 is a significant release that brings first-class TypeScript support and enables distributed testing architectures. Highlights include a full-featured TypeScript client, HTTP client for Python, upgrade to Haiku 4.5 with structured planning, and iOS locator improvements.

This release introduces a complete TypeScript client for the Alumnium server, bringing AI-powered test automation to JavaScript and TypeScript environments. The client provides native integrations with three major testing frameworks:

  • Appium for mobile and web applications via officially recommended WebdriverIO
  • Playwright and Selenium for web testing

The SDK includes comprehensive tooling for accessibility tree parsing across Chromium, Android, and iOS platforms, centralized logging with environment-driven configuration, and support for all standard automation actions including clicks, text input, keyboard operations, hover, drag-and-drop, and dropdown selections.

The package is published to NPM with automated CI/CD pipelines and includes practical examples demonstrating calculator tests, table extraction, navigation patterns, and drag-and-drop interactions.

The new HTTP client enables Python users to connect to a remote Alumnium server by passing a url parameter to the Alumni() constructor. This implementation is functionally identical to the native in-process client but enables distributed testing architectures, such as server-side deployment of the whole LLM communication engine with a centralized cache management.

Alumnium has migrated from Haiku 3 to Haiku 4.5, enabling adoption of structured output planning similar to implementations used in OpenAI and Google models. The structured approach allocates dedicated output token space for reasoning before generating final responses, improving overall accuracy.

The upgrade eliminated various workarounds specific to the older Haiku 3 implementation, including special-case handling for tool calls and separator processing. While the new model brings enhanced capabilities and improved reasoning, it comes with a 4x price increase (from $0.25 to $1 per million tokens) and more restrictive Amazon Bedrock rate limits.

Future iterations will leverage additional capabilities like thinking mode available in this model version.

The Appium implementation has migrated from XPath to iOS Predicate locators for element location strategies. iOS Predicate locators offer improved performance compared to XPath, following Appium’s official recommendations on locator strategies.

With TypeScript and Python clients both supporting HTTP communication, the foundation is set for a truly polyglot testing ecosystem, allowing us to explore supporting other languages.

The structured planning capabilities in Haiku 4.5 pave the way for integrating more advanced thinking models across all supported platforms.

We’re also focused on getting Android platform support via Appium fully working and are playing around with an Alumnium MCP server that allows general purpose agents such as Claude Code to automate web and mobile applications.