Embabel Agent Framework Reaches 1.0

InfoQ reports the release of Embabel Agent Framework at version 1.0, a framework developed for building AI agents using Java and Kotlin. The framework enables developers to define agents in terms of typed domain objects, such as goals, actions, and conditions, rather than manually crafting sequences of prompts and tool calls. Planning for the agent's actions is carried out using a process known as Goal-Oriented Action Planning (GOAP), drawing inspiration from video game AI.

This method allows the agent to search for a sequence of actions that meet the goal, even if the environment changes during execution. The Embabel framework is an addition to Spring AI, a library provided by the Spring team for interacting with models, managing embeddings, and invoking tools. It is not a replacement for Spring AI but rather an extension built on top of it.

Rod Johnson, who is the creator of the Spring Framework, announced the release of Embabel 1.0.0 GA, stating his excitement about the development. Johnson is known for founding Spring MVC, a project launched in 2003, which was designed to simplify and streamline the process of building web applications. Similar to Spring MVC, Embabel is essentially a layer on top of Spring AI, providing developers with a way to declare goals and available actions while leaving the framework to determine the sequence of steps needed to achieve those goals.

The framework supports various providers, including OpenAI, Anthropic, Gemini, Bedrock, Mistral, DeepSeek, and local or self-hosted options via Ollama, Docker, or OpenAI-compatible LMStudio endpoints. Developers can also choose specific models for individual actions or define role aliases in configuration to automatically route steps to the most suitable model based on cost, privacy, or capability requirements.

The planning step that distinguishes Embabel from LangGraph, another graph-oriented orchestration layer built by LangChain, is the runtime search for a path through available typed actions. This differs from LangGraph, where the developer would need to wire the graph beforehand. Embabel supports a combination of GOAP planning and explicit state machines within the same agent, allowing for flexibility in the agent's design.

Additionally, Embabel can be integrated with Akka, a toolkit developed by Lightbend around the actor model, which focuses on distributed systems. In Akka, agents run as isolated actors with their own state and mailbox, enabling them to survive failures and be distributed across a cluster. In comparison, Embabel offers a programming model for declaring goals and typed actions, while Akka provides infrastructure-level support for persistence, fault tolerance, and distribution.

JetBrains' Koog takes a different approach by leveraging Kotlin's language features instead of a runtime or declarative programming model. Each of these approaches represents a unique strategy for developing AI agents, with Embabel focusing on the type system, Akka on the runtime, and Koog on the language itself. For developers already using Spring Boot services, Embabel can now be evaluated as a potential option for implementing AI agents, building upon their existing infrastructure.