Chat

Chat is a conversation with the built-in assistant, and it is where you build charts. You say what you want to see. The assistant calls your data sources, shapes the rows and shows you the chart.

Chats in the nav covers the whole workspace, so you can build for any dashboard. A dashboard has its own assistant panel, scoped to that dashboard, so the assistant already knows which one you mean. Notebooks and skills have a scoped panel too.

Opening a panel starts a fresh chat, unless you used your last one there in the past 30 minutes. The back arrow at the top of the thread lists your earlier chats.

Your agent, meaning your own MCP client such as Claude Code, can build the same dashboards from outside the app. See Vistazo as an MCP server.

Ask for a chart and a draft appears in the thread, drawn from real rows.

Show monthly revenue for the last year as a bar chart

Ask for a change and the new attempt arrives as its own draft. When it replaces an earlier one, the earlier one shrinks to a grayed-out Replaced line, and its Show the current draft button takes you to the newer one. You can ask for several charts in one message and keep only the ones you want.

Nothing is saved until you press the Add to dashboard button on a draft. In a dashboard's panel, the chart goes onto that dashboard at once. In Chats, the press sends "Save this chart to a dashboard." as your next message and the assistant does the save. It asks which dashboard unless you already named one, and it may save several charts in one go.

Once a draft is added, the dashboard owns the widget. The block in the thread becomes a live view of that widget, with no edit controls of its own. Changes made to the widget show up in the thread, and nothing in the thread flows back to the dashboard. To change the chart, edit it on the dashboard or ask the assistant to change the widget.

An added chart stays tied to the dashboard you added it to. Ask the assistant to put it on another dashboard and it adds a copy there and says so in its reply. In a dashboard's narrow panel, Focus on canvas picks out a widget on that dashboard.

An added chart says what is true of it now. Where it cannot show the live widget, most labels keep the last draft on screen. In a dashboard's own panel, a chart added to that dashboard is a single line that reads Added: the dashboard shows the chart and anything wrong with it.

Type @ to mention a dashboard, a notebook or a person. On a dashboard, @ also reaches that dashboard's widgets. Type / to use a skill. The Mention and Use a skill buttons on the composer type the character for you.

A mention sits inside your sentence, so "what did @Bob change on @Global" reads as one question. Mention a dashboard or widget and the assistant opens it with its own tools, so it can answer about charts it didn't build. Mention a person and they become the subject of the question. They aren't notified, and a chat sends nothing to anyone. The one exception is Request on a chart, which notifies the dashboard's owner and the workspace admins.

Discussion comments use the same composer. There a mention notifies the person, and a / mention is a link to the skill, since no model reads a discussion.

The built-in assistant reads these docs when you ask how something works.

How do I connect Claude Code?

It answers from the published page and links it. The docs know nothing about your dashboards or numbers, so those questions are answered from your data instead. What a metric means in your workspace comes from workspace knowledge, which it learns from your own dashboards. Reading a page costs nothing extra. The conversation counts toward your inference allowance like any other chat. See Usage.

The assistant asks before it does any of these:

• archives a dashboard, removes a widget or removes a parameter
• saves a skill or a note on workspace knowledge
• changes the code of, or rolls back, a library visualization that a dashboard uses
• rewrites or deletes cells in an existing notebook, or turns on a notebook schedule
• shows a new file on a dashboard whose public access is on, which publishes the file
• calls a tool set to Ask first on one of your connections

The question arrives as a card of options, and tapping one is your answer. The server checks the option you tapped, so typing "yes, go ahead" approves nothing. An approval covers only the objects the card named. It allows the action once, in the reply that follows your tap, and never again later in the conversation.

A run carries on if you close the tab or reload mid-answer. Open the conversation again to see where it got to. As a conversation grows, the assistant summarizes its older turns and keeps going.

A conversation runs one answer at a time, and you can have only a few runs going at once across all your chats. The numbers are in Limits.

A chat belongs to the person who started it. Workspace admins can't open it, and there is no admin override.

You can share a chat by naming people as viewers, opening it read-only to the workspace, or turning on public access so anyone with its address can read it. Only its owner ever writes in it. Someone it is shared with can download it or remove it from their list. See Permissions.

Share and the star sit on the chat's header. Its menu holds Rename, Settings, Download and Archive. Settings holds the chat's name and description, its sharing and its embed.