# Ask Si in the app

> Ask Si in plain language on the Ask Si screen or in the dashboard's Ask Si panel; read the source and confirmation level, edit how the question was understood, use deep thinking.

The **Ask Si** screen (under **Ask and explore**) is where you ask Si. Si only uses metrics from the **Metric handbook**, and every answer has a **Sources** section. You need the Analyst role or above.

## Ask

1. Type your question or click one of the **Suggested questions**.
2. Click **Ask**.

Good questions name the metric, the period and the breakdown:

| Instead of | Ask |
| --- | --- |
| "How are sales?" | "Net revenue last month by channel" |
| "Profit?" | "Profit after fees for the last 28 days vs the previous period" |
| "Best products?" | "Top 10 products by gross profit this month" |

## Read the answer

- **Sources**: the metric and version used, the period, and the date the data runs to.
- **Confirmation level**: **Standard** (a metric of the industry × department pack), **Approved** (a workspace metric that was checked and approved) or **Unconfirmed** (an exploratory answer; check it before using it as an official number).
- **Tools called**: the lookups Si made.
- When the metric does not exist or data is missing, Si says **Couldn't answer this question** instead of guessing.

For standard numbers, Si does not write SQL: the question is translated into metric, dimension, filter, period and comparison, and DataxLand's engine builds the query from approved definitions. Exploratory questions beyond the metric library (medians, correlations, custom joins) run as SQL or Python in the [exploration sandbox](/en/docs/ask/sandbox) and are always labelled Unconfirmed.

## Answer language

Si answers in the language you ask in: ask in English and you get an English answer with a **Source** line; ask in Vietnamese, with or without accents, and you get Vietnamese. When a question is too short to tell (for example "GMV?"), Si follows the language of your previous question in the conversation, then the workspace's **Number and date format** setting ([time zone, currency and format](/en/docs/workspace/region)): an English format means English answers. Ask "answer in Vietnamese" or "answer in English" to switch. Numbers and money keep the workspace format, so they match your dashboards.

## Edit how the question was understood

Under the answer, **How the question was understood** lists chips for metric, dimension, filter, period, comparison and top-N. Click one, change it and click **Run again**. The re-run goes straight to the semantic layer without asking the AI again. From the result you can **Pin to page**.

## Deep thinking

Turn on **Deep thinking** for multi-step questions ("why did profit drop this month, which channel, which product?"). It is slower, can take about a minute and counts as 3 questions; **Reasoning** shows the steps. Not on the Free plan.

## Ask about a dashboard page

On a dashboard, click **Ask Si** in the top bar. Si knows which page you are on and suggests "Summarize this page" or "Anything unusual?". Clicking **Why?** on a number opens the inspector, which explains the change by formula and by dimension, computed by the engine, not written by the AI.

## Quotas

Each question counts towards the plan's monthly AI questions (Free 30, Starter 300, Small Business 1,500). When they run out, asking pauses until the next period; dashboards and reports keep running.

## Related

- [Ask Si in Slack](/en/docs/channels/slack) · [Telegram](/en/docs/channels/telegram)
- [MCP server](/en/docs/mcp)
- [Quickstart](/en/docs/quickstart)
