# Si: context analysis and Q&A

> Si is DataxLand's AI data assistant — it proposes how to read and map your data for you to approve, and answers questions with approved metrics.

**Si** is one assistant that does two jobs: **Si context analysis** (reads your data and proposes mappings for a person to approve) and **Si Q&A** (answers questions in the app, in Slack, Microsoft Teams and Telegram, and over MCP).

Everywhere in the app it is called Si: the context analysis (for example the **Ask Si to propose** button), the **Ask Si** Q&A screen and the **Ask Si** panel on the right of a dashboard. Where the two jobs need telling apart (admin pages, costs), they appear as **Si · Context analysis** and **Si · Q&A**. Older names such as Copilot, "Ask AI" or AI Agent no longer appear on screen.

## Si context analysis

When you connect a source, Si reads the table structure, the column descriptions and a sample of the data **with personal data masked** (up to 1,000 rows, only with your consent), then works through the steps you see in the **Si is analysing…** panel:

1. Read the table structure and the masked sample
2. Investigate the data
3. Pick a suitable analytics template
4. Understand the business model
5. Map columns to the standard model
6. Normalise values: statuses, channels, fee types
7. Choose the metrics that can be computed
8. Write the questions you need to confirm
9. Propose the menu layout and home page
10. Save the proposal

While investigating, Si may run **read-only** queries on the masked sample or on the source database, within limits on the number of queries and their cost; a query over those limits waits for a technical person to confirm it. The final data model is generated by DataxLand from the plan you approve; queries Si wrote never go into the build.

You help Si understand your data better with the `data_dictionary` sheet, the **Data context for Si** box (notes, description documents) and the [Knowledge](/en/docs/metrics/knowledge) page.

## Si proposes, people approve

Mappings, new workspace metrics and the semantic model are all **drafts** until an Admin or the Owner approves them. Nothing is built until the workspace is **Approved**. Tiles Si pins to a page carry a **Draft** label until you click **Save**. See [Review the proposal and build](/en/docs/review/mapping).

## Si Q&A

Si answers questions in the [app](/en/docs/ask/app), in [Slack](/en/docs/channels/slack), [Microsoft Teams](/en/docs/channels/teams) and [Telegram](/en/docs/channels/telegram), and over [MCP](/en/docs/mcp). For official numbers Si **reads metrics through the semantic layer**: your question is translated into metrics, dimensions, filters, a period and a comparison, and DataxLand's engine builds the query from the approved definitions. Si never runs model-written SQL against those numbers.

For exploratory questions beyond the existing metrics (medians, correlations, ad-hoc joins, custom charts), Si can write and run SQL or Python in an isolated [exploration sandbox](/en/docs/ask/sandbox); those results are always labelled Unconfirmed.

### Confidence levels

| Level | Meaning |
| --- | --- |
| Standard | A standard metric of the industry × department pack |
| Approved | Your workspace's own metric, checked against data and approved |
| Unconfirmed | An exploratory answer on the approved model; check it before using it as an official number |

See [Metrics and the semantic layer](/en/docs/metrics).

## Personal data

Personal-data columns are masked before anything reaches a prompt and hashed in the warehouse. Si never returns buyers' personal data. See [Security and privacy](/en/docs/security-privacy).

## Related

- [Answer Si's questions](/en/docs/review/questions)
- [Number inspector](/en/docs/ask/inspector)
- [Si on the platform page](/en/platform/si)
