Awesome AI Agents › Generative Software Frameworks

embabel/embabel-agent

⭐ 4444 Kotlin added to this list on 2025-06-16 repository created 2025-04-10

Embabel Agent Framework is a sophisticated agent framework designed for the JVM that enables the creation of agentic flows combining large language model (LLM) prompted interactions with code and domain models. It is developed in Kotlin but offers a natural usage model for Java developers. The framework is built around key concepts such as Actions (steps an agent takes), Goals (objectives the agent aims to achieve), Conditions (criteria assessed before actions or goal completion), Domain Models (objects that inform the flow), and Plans (dynamic sequences of actions formulated by the system to achieve goals). The system dynamically replans after each action, adapting to new information and outcomes, effectively implementing an OODA loop. Embabel stands out due to its sophisticated planning capabilities that go beyond traditional finite state machines or sequential execution. It uses a non-LLM AI algorithm for planning, allowing the system to perform tasks it was not explicitly programmed for by combining known steps in novel ways and making runtime decisions such as parallelization. This dynamic planning enhances extensibility and reuse, enabling the addition of new domain objects, actions, goals, and conditions without modifying existing code or FSM definitions. The framework emphasizes strong typing and object orientation, ensuring clean interaction between prompts and manually authored code, with full refactoring support. It abstracts platform internals from the programming model, allowing local or production execution without code changes. Embabel is designed for mixing LLMs to leverage different models' strengths cost-effectively and privately. Built on Spring and the JVM, it integrates easily with enterprise features like dependency injection, persistence, and transaction management. Embabel supports two authoring styles: an annotation-based model similar to Spring MVC and an idiomatic Kotlin DSL. The default planning algorithm is Goal Oriented Action Planning (GOAP), popular in gaming AI, but the framework allows for pluggable planning strategies. Execution modes include Focused, Closed, and Open, with Open mode enabling the platform to find novel paths and combine functionalities dynamically. The framework is designed for testability and aims to support natural language actions and goals in the future.

https://github.com/embabel/embabel-agent

actionsagent-frameworkagentic-aiagentic-flowsagentsai-planningaiagentframeworkannotation-based-modelclosed-modeconditionsdomain-modeldynamic-planningenterprise-integrationextensibilityfocused-modegenerative-aigoal-oriented-action-planninggoalsgoapjavajvmkotlinkotlin-dsllarge-language-modelllmllm-mixingobject-orientationooda-loopopen-modeplanplatform-abstractionreusespringstrong-typingtestability