# AI assistant

> The in-app assistant that answers data questions in natural language by running governed queries and proposing charts and dashboards for you to save.

*[View this page in the Flax docs](https://flax-analytics.com/docs/automation-ai/ai-assistant)*

Flax's AI assistant is an in-app chat that turns natural-language questions into
governed queries, charts, and dashboards. It is **grounded** in your semantic model
and bound by your access rules: it can only query the models, dimensions, and
measures you already have, and it runs everything under your identity and
row-access filters. It never writes to anything — it only *proposes*.

## Using it

Open the assistant from the right-side panel and ask a question in plain language,
scoped to a project. It answers in a streamed chat, showing the queries it runs and
any charts or dashboards it proposes.

You can also ask about **exactly what you're looking at**: use **Ask AI about this
chart** on a chart's menu, or **Ask AI** on a dashboard, and the assistant is
grounded in that chart's query (or every tile in the report) so its answer speaks
to what's on screen.

Conversations are saved and independent. Use the **history** list in the panel to
resume a past chat, **rename** it, or **delete** it, and **New chat** to start a
fresh one — so different investigations stay separate rather than one long thread.

If no AI key is configured, the panel links you to
[AI settings](/docs/automation-ai/ai-settings-and-usage) to add one.

## What it can do

The assistant works through a small set of tools:

| Tool | What it does |
| --- | --- |
| **Run query** | Runs a governed query over your project's models and returns rows. Compiled and executed under your access rules. |
| **Propose chart** | Suggests a chart (query + visualization) for you to preview and save. Saves nothing on its own. |
| **Propose dashboard** | Suggests a multi-tile dashboard for you to preview and create. |

Because it builds structured queries against your real models rather than writing
raw SQL, it cannot invent fields or bypass governance. Proposals come with a
**Save** button — you decide what becomes a durable [chart](/docs/visualizations/overview)
or [dashboard](/docs/visualizations/dashboards).

> [!NOTE]
> The assistant proposes; it never saves or edits your content by itself. There is
> no write tool. Query results in chat are capped to a small sample for speed — the
> chart you save re-runs the full query.

## Limits

Each turn is bounded (a capped number of reasoning rounds and a per-turn token
limit) and chat is rate-limited per user. These keep the assistant responsive and
predictable.

## Related

- [AI settings & usage](/docs/automation-ai/ai-settings-and-usage) — configure the assistant and track output.
- [Building queries](/docs/exploring/building-queries) — the query model the assistant uses.