What You Can Do with ChartBrick in Claude or ChatGPT

The assistant reads your live datasets — no exports, no copy-pasting:

• Inventory — "What datasets do I have, and what columns are in each?"
• Aggregate — "Total revenue by month." "Average order value by country."
• Filter — "Only where channel is Online and amount is above 500."
• Compare — "Online vs retail by quarter — which grows faster?"
• Totals — "What's the grand total of ad spend this year?"
• Explain existing charts — "What does my 'Spend by platform' chart show?" It reads the chart's real configuration and current data.

Answers always come from the latest refresh of each dataset. For very large datasets, ask for a breakdown (by month, by category…) rather than the raw table — results are trimmed to a reasonable size.

Ask for any chart — "show me the sales chart", "render the traffic chart in dark mode" — and in Claude and ChatGPT it appears as an image right in the conversation, with an Open in ChartBrick link to jump into the editor. It matches the light or dark look of your chat automatically.

Good to know:

• Works for any chart with data, whether or not it's published. Tables are the one exception — ask for the data instead.
• The in-chat image link is valid for about 15 minutes; ask again for a fresh render.
• Assistants without inline display (e.g. Perplexity) get a link to the rendered image instead — whether it's shown depends on the assistant.

With the create-and-edit permission, the assistant builds real charts in your workspace:

• Create — "Make a bar chart of total revenue by month from the Monthly performance dataset." You get the chart, a link to the editor, and it stays private until you publish it.
• Pick from all your chart types — bar, line, area, pie, donut, scatter, heatmap, funnel, KPI, and more.
• Edit — "Rename it to Q3 Revenue, switch it to a line chart, and filter out refunds." Only what you mention changes.
• Combine measures — "Revenue as bars and new customers as a line on a secondary axis" — a combination chart with two scales, so a small metric isn't flattened next to a big one.

Whatever it builds, you can open in the editor and refine by hand — it's a normal ChartBrick chart.

You can ask for the visual details in plain language, at creation time or later:

• Titles & captions — title, subtitle, and a footnote (e.g. a data-source note).
• Colors — one of the named palettes ("use the Ocean palette") or exact brand colors as hex codes.
• Labels — show values on the bars/points; on pie and donut charts they can be percentages.
• Number format — prefix/suffix ("€", " kg"), decimals, compact notation (1.2M). One format applies to the whole chart.
• Layout — legend position, stacked or 100%-stacked bars/areas, smooth or stepped lines.

Example: "Create an area chart of new and returning signups by week, titled 'Growth', stacked, smooth lines, values shown on the chart, using our brand colors #1B7F79 and #F4A261."

Also behind the create-and-edit permission:

• Dashboards — "Create a dashboard called KPIs and add my revenue and churn charts to it."
• Publish / unpublish — "Publish that chart and give me the public link." Publishing makes the chart or dashboard reachable by anyone with the link, same as the Publish button in the app.
• Add data — paste CSV or JSON into the chat: "Add this as a dataset called Q3 survey." The assistant creates the dataset and can chart it immediately.
• Refresh — "Refresh the Airtable dataset before we look at it." Works for datasets whose source supports refresh.

Regardless of permissions, the connector can never:

• Delete anything — no datasets, charts, or dashboards.
• Leave its workspace — it only sees the one workspace you picked when connecting.
• Manage your account — no members, billing, passwords, or workspace settings.
• Act as read-write when you said read-only — without the checkbox, every editing request is refused.

And you can cut access at any moment from Settings → Account → Connected apps → Revoke — the very next request from the assistant is refused.