Agent Perspective Overview

Agents ship with v2.0.0; before that release, Dashboardx Designer had no AI at all.

The Agent Perspective is where you stop building the dashboard by hand and start describing it. You say what question the dashboard should answer; the agents read your data, choose the charts, lay out the canvas, style it, and check their own work — with you approving anything that matters.

It is not one chatbot with a text box. It is a small team, and you can see who is doing what.

What is in the perspective

Part	What it does	See
Agent workspace	Full-screen conversation view — the whole dashboard is your context	This page
Agent Dock	The same conversation as a tab beside your spreadsheet or canvas	Agent Dock
Plan Mode	The agent proposes a design and waits for your approval before touching anything	Plan Mode
Thinking trace	The agent's working-out, available when you want to inspect it	Thinking & Reasoning

The dock and the full-screen perspective share one conversation. Switching between them does not restart anything.

The agent team

Six built-in agents cover the work. You talk to one; it delegates to the others when a specialist would do better.

Agent	Responsible for
Dashboard Master	Your entry point. Understands the request, plans, delegates, applies changes to the canvas, and reports back.
Data Analyst	Turns raw sheets into something plottable — finds the data region, cleans and derives columns, creates datasets.
Visualization Professor	Decides what should be on the canvas: which chart answers the question, how it should be sized, how slices should interact.
Dashboard Stylist	Theme, colour, KPI thresholds, and the overall coherence of the canvas.
Data Reviewer	Checks the numbers before they reach the canvas.
Visualization Reviewer	Checks the finished layout against your request and flags what should change.

You can talk to any of them directly from the agent selector, and set a preferred model for each in Settings → Agent.

How a request flows

1. You say what you want. For example: "Regional sales trend by month, with the top five products, and give me a channel filter."
2. The master breaks it into steps. It delegates where that helps — data shape to the Data Analyst, chart choice to the Visualization Professor.
3. The work lands on the spreadsheet and canvas. Anything that changes your files asks first. See Settings → Permissions.
4. Reviewers check it. If something is wrong, the master fixes it and re-checks rather than declaring victory.
5. You can take over at any point. Every change the agents made is a normal change — edit, move, or undo it afterwards.

While it runs, the conversation shows which step is in flight, and the thinking trace explains the reasoning behind it — you see progress, not an unexplained spinner.

Approvals are part of the flow

The agents do not get a blank cheque. Anything that writes to your files asks first, and you choose whether to allow once, always allow, or refuse. Operations that delete data always ask, even when you have turned on auto-accept. See Settings → Permissions.

Choosing a model

Reasoning-capable models show their working-out and tend to make better design decisions. Faster models are fine for mechanical edits. Set defaults per agent in Settings → Agent and override per conversation in the message box. See Settings → Models.

Agents need a model that is configured and verified — set that up first in Settings → Providers.

What agents can reach

Over a hundred capabilities are available to the agents across the spreadsheet, canvas, datasets, widgets, and presentation — reading and writing ranges, creating and binding datasets, adding and arranging widgets, applying themes, and driving presentation mode. They work through the same operations you would, which is why their output is fully editable afterwards.

Where to go next

• Quick Start — your first dashboard, end to end
• Agent Dock — keep the conversation beside your work
• Plan Mode — agree on a design before anything is built