Skip to main content

AG-UI: Stateful agent with customization

This sample shows a richer AG-UI stateful agent that combines user interaction, model tools, backend tool calls, step markers, activity updates, and streaming events.

Choose this when​

Use this sample when you want to see how several agent-facing features work together in a single conversation flow.

What it demonstrates​

  • Asking the user a frontend question before the agent responds.
  • Preparing and synchronizing activity state with ActivitySync nodes.
  • Wrapping the model call in StepStart and StepEnd nodes.
  • Configuring a ModelCall with selected tools.
  • Calling a backend tool and updating activity progress after the model responds.

How it works​

The workflow asks whether the user wants a customized response style, initializes an activity plan, publishes it to the frontend, then runs a model call that can use selected tools. After the model response, Code and backend tool nodes update context values and activity state before the step ends. The frontend question result is kept as workflow context rather than persisted as model chat history.

Setup notes​

Create a connector and model before running this sample, then select that model in the ModelCall node. It needs the Sharpy client sample because it understands the ask_a_question tool and other details needed for the sample. When using Sharpy, make sure you uncheck the Send All Messages checkbox and put the workflow ID into the input with the same name. This sample depends on the 'DemoServer' host code that defines function tools used by the model.