Tutorials

Tutorial: let an agent build a dashboard

In this tutorial an AI agent (Claude, ChatGPT, Codex, Cursor…) designs a schema, imports data, builds a dashboard and publishes a report, while you stay in control through approvals.

Before you start

  • An MCP endpoint connected to your client (Connect an agent).
  • A workspace where the agent may work.

Step 1 – Create a project

Ask your agent:

Create a Basemind project called “Marketing analytics”.

The agent calls create_project. Open Projects in the sidebar to see it. Leave the autonomy level on Assisted for now.

Step 2 – Record the goal in memory

Remember as a decision: we measure campaigns by cost per signup, weekly.

The agent calls remember. Because the project is Assisted, the entry waits in Agent activity. Open the tab, read the preview and click Approve. The memory now appears in the Memory tab.

Step 3 – Design the schema

In the Marketing analytics project, design a database “Marketing” with a Campaigns table (name, channel: Google/Meta/LinkedIn, start date, budget) and a Weekly results table (campaign link, week, spend, clicks, signups). Show me the plan first.

The agent calls schema_diff and shows a plan such as create table Campaigns, create field Channel (single select)…. When you agree:

Apply it.

apply_schema creates everything in one step. Approve it in Agent activity. If you do not like the result, click Undo and everything the call created disappears.

Step 4 – Add sample data

Add 4 campaigns and 8 weeks of realistic results for each.

The agent uses create_rows. Switch the project to Autonomous first if you do not want to approve each batch, or give the endpoint Full access.

Step 5 – Ask questions

Which channel had the lowest cost per signup in the last 4 weeks?

The agent uses records_query with a group_by on channel and sum aggregations, then explains the answer. Ask it to remember the finding:

Remember this as a fact.

Step 6 – Build the dashboard

Create a dashboard “Marketing” with a summary of total signups, a bar chart of spend per channel and a line chart of signups per week.

The agent calls create_dashboard_widget three times. Open the dashboard, rearrange the widgets and resize them as you like.

Step 7 – Publish a report

Create a report “Weekly marketing” with the key metrics, the signups per week chart and a table of campaigns sorted by cost per signup, then publish it with a secret link.

The agent calls create_report, render_report to check it and publish_report. Open the returned public_url in a private window: the page works without signing in, for anyone who has the link.

Step 8 – Next session

In a new chat, ask:

What do we know about the Marketing analytics project?

The agent calls get_project_context and starts with your decisions, facts and reports: the context survived the chat.

What you learned

  • Keep agents in Assisted mode while learning and approve from Agent activity.
  • Use schema_diff → apply_schema for safe, reviewable schema changes.
  • Store decisions and findings in memory so every session starts informed.
  • Undo reverts a whole agent action in one click.