Managing dataset state for multi-step transformations in Semantic Kernel

[Daniel (soderdani)]

How to keep dataset context across multiple transformation steps in Semantic Kernel?

I’m building an agent in Semantic Kernel that can fetch, modify, and transform data. One of my plugins generates charts, but each chart type requires specific inputs (for example, a bar chart needs x and y values).

I’d like the agent to start with a dataset containing multiple columns, then iteratively refine it using different transformation functions or tools until it has just the right shape for the chart. For example:

• Load dataset
• Extract relevant columns
• Group products to calculate quantity
• Send final dataset to chart plugin

If the user asks for a bar chart showing the quantity of each product, the agent should follow a pipeline like the above, step by step.

My question: How can I maintain the context of the dataset across these multiple transformation steps within Semantic Kernel?