Embabel

An agentic AI framework for Java and the JVM, created by Rod Johnson, founder of the Spring Framework. DICE ships as a module of it, and Embabel DICE is that module’s implementation.

• Framework: https://github.com/embabel/embabel-agent
• DICE module: https://github.com/embabel/dice
• Impromptu (reference sample): https://github.com/embabel/impromptu

---

Shape of the API

From the worked example in Agents That Extract and Use Preferences from Conversations, the programming model is annotation-driven in the Spring idiom:

• @Action methods, optionally with canRerun and a trigger type, run in response to typed events such as a UserMessage.
• An OperationContext / ActionContext carries process state, including the identity of the user the process is bound to.
• LLM calls are composed fluently — context.ai().withLlmByRole(...).withReferences(...).withSystemPrompt(...).respond(...).

Models are selected by role, not by name. The sample binds the responding call to a CHEAPEST role, proposition revision to STANDARD, and the embedder to FAST, with the mapping configured in application.yml. That is model tiering as a first-class configuration concern — the same discipline Agentic Memory for Search Personalization argues for, expressed in the framework rather than bolted on.

References — RAG sources, memory — are passed to a call as a list, so agent memory enters the prompt through the same mechanism as retrieved documents.

Related Tools

• Embabel DICE — the DICE module: proposition extraction, entity resolution, memory projection
• django-dice — an independent Python implementation of the same DICE design

Related Concepts

• DICE · Agentic Memory · Context Engineering · RAG

Articles

• Agents That Extract and Use Preferences from Conversations — Jettro Coenradie; a chatbot built on Embabel with DICE preference extraction wired in