Embabel Agent Framework for Java AI Agents Reaches 1.0 General Availability

Embabel 1.0 Release

Embabel, a framework designed for constructing AI agents within Java and Kotlin environments, has officially reached its 1.0 general availability milestone. This release provides developers with tools to define agents using typed domain objects, which include goals, actions, and the conditions linking them. This approach contrasts with the traditional method of manually coding sequences of prompts and tool calls.

Dynamic Planning with GOAP

The framework incorporates a planning step that utilizes Goal-Oriented Action Planning (GOAP), a concept borrowed from video game AI. Instead of relying on predefined scripts, agents are given a set of actions with preconditions and effects. A planner then determines the optimal sequence of actions to achieve a goal. This allows the agent to adapt and find new paths if conditions change during a task, such as a tool call failing or new information becoming available.

Integration with Spring AI

Embabel is built on top of Spring AI, the Spring team's library for model interaction, embedding management, and tool invocation. This layered architecture means Embabel does not replace Spring AI but rather enhances it, providing a higher-level abstraction for agent development. Rod Johnson, co-creator of Embabel and founder of the Spring Framework, drew a parallel to Spring MVC's relationship with raw servlets, where Spring MVC provided a structured layer over basic servlet functionality.

Flexible Model Support

Due to its foundation on Spring AI, Embabel inherits support for a wide range of model providers, including OpenAI, Anthropic, Gemini, Bedrock, Mistral, and DeepSeek. It also supports local and self-hosted options via Ollama, Docker, or OpenAI-compatible LMStudio endpoints. Developers have the flexibility to pin individual actions within an agent to specific models, allowing for fine-grained control over model choice.