TLDR: Google’s Agent Development Kit (ADK) for Java has released version 0.2.0, significantly expanding its capabilities by integrating with the LangChain4j LLM framework. This update allows Java developers to access a much wider array of Large Language Models (LLMs) beyond Google Gemini and Anthropic Claude, including those from OpenAI, Anthropic, Mistral, and local models via Ollama or Docker Model Runner. The integration facilitates multi-agent scenarios and offers improved tooling, asynchronous operations, and enhanced memory management for AI agent development.
Google has announced a major update to its Agent Development Kit (ADK) for Java, with the release of version 0.2.0. This latest iteration introduces a pivotal integration with the LangChain4j Large Language Model (LLM) framework, dramatically broadening the scope of LLMs available to Java developers for building sophisticated AI agents. Prior to this update, the ADK for Java primarily supported Google Gemini and Anthropic Claude. The new LangChain4j integration now opens the door to a diverse ecosystem of models, including those from OpenAI, Anthropic, and Mistral, as well as local open-weight models accessible through platforms like Ollama and Docker Model Runner (e.g., Gemma, Qwen, Phi).
Guillaume Laforge, a Google developer relations engineer and a key contributor to LangChain4j and its ADK integration, highlighted the strategic advantage of this development. He explained that LangChain4j enables the creation of multi-agent scenarios where different models can be mixed and matched. This allows developers to leverage the ‘best model for the job,’ potentially using a fast model for simple classification tasks to route requests, and a more powerful model like Gemini 2.5 for complex reasoning. Laforge provided a practical example involving a main agent powered by Claude and a tool agent utilizing OpenAI for weather information, demonstrating the flexibility in combining diverse LLMs.
Beyond the headline LangChain4j integration, ADK 0.2.0 brings a suite of other significant enhancements aimed at improving both tooling and agent performance. Developers can now create FunctionTools from object instances, offering greater flexibility compared to previous static method limitations. The update also improves asynchronous tool operations, making agents more responsive, and provides finer programmatic control over agent execution. On the agent logic and memory front, the release strengthens handling through callback chaining and introduces new memory management primitives, giving developers more granular control over how agents store, retrieve, and process information. This includes an InMemoryMemoryService for simple memory management and VertexAiRagRetrieval for advanced RAG patterns.
Also Read:
- Google Introduces AI-Powered Gaming Agents and Competitive Leagues to Android Play Games
- Hugging Face Unveils Smol2Operator: An Open-Source Framework for Developing Agentic GUI Coders from VLMs
This expansion positions the ADK for Java as a more open and versatile framework for AI agent development, allowing developers to tailor their solutions with a wider range of LLMs, whether cloud-hosted or running locally. The ADK for Java, part of Google’s broader Agent Development Kit introduced earlier this year, continues to evolve rapidly, with documentation and template projects available for developers eager to explore its new capabilities.
