---
product: "AG Studio"
title: "Working with AI"
description: "The chat panel turns everyday language into dashboards and data queries."
framework: angular
version: "3.0.0"
llms: "https://www.ag-grid.com/studio/archive/3.0.0/llms.txt"
---

# Working with AI

The chat panel turns everyday language into dashboards and data queries.

AG Studio includes a built-in chat panel that connects to an AI service your team sets up - such as OpenAI or Anthropic. Through it, Widgets can be created, Filters added, and the dashboard rearranged just by describing what you want, without needing to dig through configuration panels.

## How It Works

A request is not handled by one responder. One part of the AI coordinates the work, and specialised agents plan the dashboard, query the data, arrange the page and configure each Widget. What you type becomes actions in the dashboard, such as adding a Widget or changing a Filter.

The AI itself comes from a service your team connects during setup, such as OpenAI or Anthropic. AG Studio does not provide or host the AI - which means the quality and behaviour of responses depends on the service your team has chosen. It also means your team can switch to a different AI service at any time.

The chat panel will not appear in the Studio until your team has completed the setup. If it is not visible, check with whoever manages your AG Studio installation.

Once configured, the chat panel is accessible from the left side by default. When open, it appears as a conversational interface alongside the dashboard. For an overview of how the panel is organised, see [User Interface](https://www.ag-grid.com/studio/archive/3.0.0/angular/user-interface/).

## Threads

Each conversation lives in a thread. Multiple threads can be created - for example, one for building a sales dashboard and another for exploring inventory data.

Threads are not part of the dashboard itself, so saving the dashboard does not save the conversation. Whether threads are still there tomorrow depends on how your team set the chat panel up: if they have connected somewhere to store conversations, threads are kept between visits. If they have not, threads last only as long as the page is open, and are gone when it is reloaded.

To start a new thread, click the New button in the chat panel header. To switch between threads, click the thread name and select from the list.

Messages are sent using the input field at the bottom of the panel. Press Enter to send, or Shift+Enter to insert a new line without sending.

## Asking Effective Questions

The chat panel responds to natural language requests like:

- "Create a dashboard with KPIs for total revenue, average order value, and customer count"
- "Add a bar chart showing sales by product category"
- "Show me [Field] by [Category]"
- "What are the trends in [Measure] over time?"
- "Move the pie chart to the top right corner"
- "Add a date filter for the last 12 months"

> **Note**
>
> More specific requests produce better results. Instead of "Make me a dashboard", try "Build a sales dashboard showing revenue by region, a top-10 products table, and monthly sales trends". Referencing actual Field names helps the AI understand the data more precisely.

Complex dashboards often build better through multiple requests rather than trying to describe everything at once. When the first result does not match expectations, describing the adjustment needed is more effective than starting over - for example, "Change the bar chart to a line chart" or "Add a quarterly breakdown".

## Understanding Progress

When a message is sent, the AI works through the request in stages. A thinking indicator is shown while it is processing. For complex requests it may create a structured plan before executing it, and delegate sub-tasks to specialised agents for layout, widget configuration, and data querying. When finished, it summarises what was done.

## Limitations and Considerations

The AI works best when the data structure is well documented with clear Field names and descriptions, and when requests are framed in business terms rather than technical terms. It can only work with the data sources currently loaded in the Studio.

It may not always choose the most optimal visualisation on the first attempt, and unusual or complex business logic may need to be explained explicitly. How quickly it responds, and how accurate its suggestions are, depends on the AI service your team has connected - not on AG Studio itself. If a result isn't quite right, following up with a more specific request is a normal and effective way to get there.