Insights

Data analyst with a live desktop. Connect Insights to a database or drop in a spreadsheet and it queries, designs metrics, and builds an interactive ECharts dashboard — rendered and visible in real time via VNC.

## What it's good at

- **Database analysis** — connect to MySQL or PostgreSQL (read-only) and run queries against your live data
- **Spreadsheet analysis** — read Excel (.xlsx) and CSV files directly; no manual import needed
- **Metric design** — Insights thinks about what to measure, not just how to display it
- **Interactive dashboards** — output is an ECharts + HTML dashboard, not a static image; charts are clickable and filterable
- **Self-verification** — Insights previews the dashboard in a built-in browser and takes a screenshot to confirm it renders correctly before delivering

## How it works

Insights runs in a persistent sandbox with 4 GB RAM, 2 vCPUs, and a full VNC desktop environment. When you provide a data source, it connects, reads, and explores the data to understand its structure before asking what you want to analyze.

It writes Python or JavaScript analysis code, generates ECharts configuration, builds an HTML page, and starts a local HTTP server so you can preview the result live in the VNC browser. Once the layout looks right, it delivers the final dashboard files to your workspace.

The sandbox uses a `longrunning` lifecycle — it pauses when idle and resumes when you return. The workspace is permanently stored; your dashboards, data files, and scripts are there next time you open the conversation.

If Docker sandbox is not enabled, tasks execute directly on the machine where Yao Engine is installed.

## How to write a good prompt

**Lead with the data source.** Tell Insights where the data is: a database connection, a file attachment, or both. The more specific you are, the faster it gets to the analysis.

**Describe the question, not the chart type.** "Show me which products are driving revenue growth" works better than "make a bar chart." Insights will choose the right visualization.

**Specify the audience if it matters.** "This is for a weekly exec review" or "the team just needs a quick overview" shapes the level of detail and layout.

**Examples:**

> Connect to MySQL at localhost:3306, database `sales`, and analyze revenue trends by region for Q1–Q2 2025.

> Analyze this CSV file. I want to understand customer churn — which segments are leaving most.

> Build a KPI dashboard for this Excel export: total orders, avg order value, return rate, and customer lifetime value.

> Connect to my PostgreSQL database and find the top 10 underperforming SKUs by margin.

## Capabilities at a glance

| Capability | Details |
|------------|---------|
| Data sources | MySQL, PostgreSQL (read-only), Excel (.xlsx), CSV |
| Output | ECharts + HTML interactive dashboard, delivered to workspace |
| Preview | Built-in browser via VNC; auto-screenshot for verification |
| Sandbox | `claude-desktop` image, 4 GB RAM, 2 vCPUs, VNC enabled |
| Lifecycle | Long-running — pauses on idle (1 h), max 24 h per session |
| Workspace | Permanently stored — dashboards and data files persist indefinitely |

## What's next

- [Transformer](/docs/en-us/built-in-agents/transformer) — clean and restructure raw data before handing it to Insights
- [Report Writer](/docs/en-us/built-in-agents/report-writer) — turn Insights findings into a formal written report
- [AI Experts: Find the Right One](/docs/en-us/getting-started/ai-experts-intro) — see all built-in experts