> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gocrisp.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Artifact refinement

> Iteratively refine charts, documents, tables, and presentations through a chat-based editing interface.

Artifact refinement lets you fine-tune any artifact an agent creates — a chart, a markdown document, a CSV table, or a PowerPoint deck — through natural-language instructions in a split-pane editor.

## Supported artifact types

| Type             | What you can refine                                                                                    |
| ---------------- | ------------------------------------------------------------------------------------------------------ |
| **Charts**       | Colors, labels, legends, axis formatting, chart type, data series                                      |
| **Markdown**     | Content, structure, formatting, tone, and length                                                       |
| **CSV / Tables** | Column selection, sorting, filtering, calculated columns                                               |
| **PowerPoint**   | Slide content, layout, formatting, chart data — edits target the slide you are viewing in the carousel |

## How it works

<Steps>
  <Step title="Open the refinement modal">
    Click the **Refine** button on any supported artifact in your conversation. A split-pane modal opens.
  </Step>

  <Step title="Describe your changes">
    In the chat pane (right side on desktop, Chat tab on mobile), type what you want changed. For example: "Make the title font larger" or "Add a column showing percentage change."
  </Step>

  <Step title="Review the preview">
    The left pane shows a live preview of the artifact with your changes applied. For markdown documents, a track-changes view highlights additions, modifications, and removals.
  </Step>

  <Step title="Iterate">
    Continue sending instructions until the artifact looks right. Each message builds on the previous changes.
  </Step>

  <Step title="Apply or discard">
    Click **Apply Changes** to save the refined artifact back into your session. Click **Cancel** to discard all changes and keep the original.
  </Step>
</Steps>

## Refining a PowerPoint deck

PowerPoint refinement is slide-aware. The preview pane is the same slide carousel that appears in chat, and the agent edits the slide you have selected.

* Navigate to the slide you want to change using the thumbnails, arrow keys, or mouse wheel.
* Ask for the change — for example, "shorten the bullets on this slide" or "swap the chart for a line chart."
* The agent updates only the slide you are viewing and re-renders it in place. Other slides stay untouched.
* Switch slides at any time to refine a different one in the same session.

When you click **Apply Changes**, the updated `.pptx` replaces the original in your session files and the inline carousel refreshes.

## Refining a saved tile

If the chart or table you refine is a saved **tile** — for example, one attached to a [scheduled report](/unfi-ai-agents/schedules#tiles-in-scheduled-reports) — the preview you already have open updates in place after each change. You don't get a new copy in the conversation, and future scheduled runs pick up the updated tile automatically.

A provenance panel on the tile shows which run and parameter values you're looking at, and whether the values came from a scheduled run or an interactive session.

## Desktop vs. mobile layout

* **Desktop** — The modal uses a resizable split pane. Drag the divider to adjust the balance between preview and chat.
* **Mobile** — The modal uses a tabbed layout with **Preview** and **Chat** tabs. An unread-message badge appears on the Chat tab when the agent responds while you are viewing the preview.

## Starting fresh

If you want to begin the refinement from scratch without closing the modal, click **Start new session**. This clears the refinement chat history and resets the artifact to its original state.

## How refinement uses your session agent

The refinement agent is based on the same agent you are using in your session. This means it has access to the same data context, tools, and domain knowledge — so refinement instructions like "use last quarter's data instead" work without additional explanation.