# Meet cherry
> cherry is sanda's data agent. Ask it about your numbers, and it builds metrics, views and reports when you say so.
cherry is the one conversation in sanda. You ask a question of your data and cherry answers it from your semantic fluid, with the queries behind the answer one click away. You ask it to build something (a metric, a view, a report) and it does the work in the same conversation, holding each change for your confirmation unless your workspace has chosen otherwise.
cherry is included on every edition. What it can change for you depends on your role and on what your workspace has switched on.
## Where cherry lives
There is one conversation, and you can reach it from four places. A thread you start in one is the same thread in the others.
- **The pane.** On every page except the two below, cherry sits in a column at the right of the window. Open and close it with the **cherry** button at the right of the top bar, or with ⌘ J (Ctrl J on Windows). Drag its left edge to make it wider. It stays open as you move between pages, so you can ask about a report while you look at it. A dot on the button tells you cherry has replied while the pane was closed.
- **cherry chat.** The **cherry chat** entry in the left navigation opens the conversation full screen, with your earlier threads down the left. The full screen button in the pane's header takes you there.
- **The overview.** The overview opens on a greeting and a box to ask in. Asking there takes you to cherry chat with your question already sent.
- **A phone.** cherry is a sheet over the page. It opens only when you ask it to, and only on the page you asked from.
On cherry chat and on the overview, ⌘ J puts the cursor in the box instead, because those pages are cherry already. See [keyboard shortcuts](https://docs.sanda-os.com.au/reference/keyboard-shortcuts) for the rest.
cherry knows which page you are on. On a report it knows which report is open, so "put a revenue tile on this" means that report. On the modelling page it knows which warehouse you are looking at.
## What cherry does
- **Answers questions** about your numbers: compare two periods, break a figure down, find the biggest movements. See [ask questions](https://docs.sanda-os.com.au/cherry/ask).
- **Builds in the semantic fluid.** It puts tables on the map, defines metrics, dimensions, filters and relationships, and reviews what learn from my data proposed.
- **Builds in the warehouse.** It writes views and procedures and schedules them.
- **Builds reports.** It creates a report, places tiles, charts, tables and written analysis, and changes a report that already exists. See [build reports with cherry](https://docs.sanda-os.com.au/reports/build-with-cherry).
- **Runs the moving parts.** It starts a sync, runs learn from my data, and schedules a report to be emailed.
- **Sets up a new workspace**, one step at a time, with a checklist you press through. See [set up with cherry](https://docs.sanda-os.com.au/get-started/setup-with-cherry).
- **Answers how-to questions** about sanda from these docs, with a link to the page.
- **Takes you to the right page** when you ask, and the conversation goes with you.
The full list, and which parts depend on your role, is in [what cherry can do](https://docs.sanda-os.com.au/cherry/capabilities).
## Grounded in your semantic fluid
cherry does not guess at your business. It is given the names of every metric, dimension, named filter and table on your [semantic fluid](https://docs.sanda-os.com.au/semantic-fluid), and it answers by asking for those things by name. sanda then composes the query from the definitions your team agreed. The number cherry quotes is your definition of revenue, not its own.
When the fluid cannot express a question, cherry can write read-only SQL of its own. It says in a sentence why it needed to, and the statement is shown to you. If it leans on a column nobody has defined, it names the column, so you know which term to define next.
cherry is instructed to quote only numbers that came back from a query in the conversation, and to pass on a refusal in plain words rather than work around it. You do not have to take that on trust: every answer can show its work.
## It shows its work
Under each reply you can see:
- **The steps it took**, as one quiet line between its sentences: what it looked up, how many steps, how long, and whether any failed.
- **SQL and rows**, the statement behind the answer and the rows it returned.
- **What it changed or proposed**, as cards with the exact detail.
- **What the reply cost**, as a small dollar figure under the reply.
## What cherry sees
To answer, the model behind cherry needs enough to be useful. A request can carry:
- your question and the recent messages in the conversation;
- the names of your tables and columns, and the definitions on your semantic fluid;
- a small sample of real values from your columns, up to six per column, so it can tell what a column holds;
- the rows a query returns, up to 50 at a time;
- your name, your role and your workspace's name.
Columns whose names suggest a secret, such as passwords and tokens, are never sampled or returned. Other columns are not filtered by content. Requests go only to model endpoints that do not keep what they receive and do not train on it. Your [privacy page](https://docs.sanda-os.com.au/workspace/privacy) sets this out in full.
## What cherry does not do
- It does not act on its own initiative in setup. It does not provision a warehouse, connect a source, sync, learn or build a report unless you ask.
- It never takes a credential in chat. You type a password or token into the connection wizard, never into cherry.
- It does not change your plan, billing, team or settings.
- It does not create or manage search services. You do that in the console. It can search an index once one exists. See [use search](https://docs.sanda-os.com.au/search/use-search).
- It does not change your data with the SQL it writes to answer a question. That SQL is read-only.
- It does not email a report without asking you first, whatever mode your workspace is in.
## Who can use it
Everyone who can open the console can ask cherry questions. What it can change depends on your role: a viewer's cherry answers questions and changes nothing, and some actions are for owners and admins only. See [what cherry can do](https://docs.sanda-os.com.au/cherry/capabilities).
People who read reports on the reports portal get a smaller cherry that only answers questions about what has been published to them. It cannot build, change or send anything.
## Cost
Every AI call sanda makes for your workspace is billed, cherry's included, and counts towards your daily AI limit. The meter in the top bar shows where you are. See [models and AI usage](https://docs.sanda-os.com.au/cherry/models-and-usage).
:::links
- [Ask questions](https://docs.sanda-os.com.au/cherry/ask): How to ask well, and what a reply contains.
- [What cherry can do](https://docs.sanda-os.com.au/cherry/capabilities): The full list, by role.
- [Actions and approvals](https://docs.sanda-os.com.au/cherry/actions-and-approvals): Ask first, Autonomous, and what always asks.
- [Memory and conversations](https://docs.sanda-os.com.au/cherry/memory-and-history): Threads, history and what cherry remembers.
- [Models and AI usage](https://docs.sanda-os.com.au/cherry/models-and-usage): Choosing a model and the daily AI limit.
:::