Utility Agent
The Utility Agent is the simplest type of agent in Flowable.
It allows invoking a Large Language Model (LLM) with a defined input and output in a lightweight, targeted way.
It is ideal for simple AI-driven tasks that do not require orchestration or intent handling.
Use the Utility Agent when you need a direct, single-step AI task - such as rewriting text, summarizing input, or extracting structured data from a short message or value - without involving case context or multiple agents.
Key Features
- Executes a single operation per invocation
- Can be triggered via:
- Supports both structured and unstructured input/output formats
- Multiple operations can be defined and reused
Operations
Utility Agents rely on operations to define how inputs are handled and what outputs are expected.
Each Utility Agent is bound to one specific LLM model configuration.
If you need to support different models or prompt behaviors, you must create multiple Utility Agents with their own configurations.
Only one operation is executed per invocation. This makes the Utility Agent suitable for focused, stateless tasks.
Design-Time Testing
All operations defined in a Utility Agent can be tested directly within Flowable Design, allowing you to quickly validate prompts, outputs, and mappings before deploying them in process or case models.
Summary
- The Utility Agent provides a minimal way to invoke LLMs with defined inputs and outputs.
- It supports single-operation execution, making it ideal for simple AI tasks.
- It integrates into BPMN and CMMN via the AI Agent Task or can be called through the API.
- Each agent uses a specific model and configuration; create multiple agents to support different behaviors.