> ## Documentation Index
> Fetch the complete documentation index at: https://docs.appreply.co/llms.txt
> Use this file to discover all available pages before exploring further.

# Chats and answers

> Ask RADAR a question, read an answer with charts, attach files and approve the changes it proposes

Everything you do with RADAR happens in a chat. You type a question, RADAR works on it, and the answer appears with charts, review cards and, when you asked for a change, an approval card.

## Ask a question

Open a new chat from the RADAR area, from the side panel (**Chat options**, then **New chat**), or from Home. The new chat screen asks "What do you want to know about your reviews?" and has a message field below it.

![The new chat screen with the question, the message field with its Quality and Auto buttons, and the Recommendations cards below](https://mintcdn.com/appreply/bnS5T3Kw2jFffd-K/images/product/assistant/01-new-chat.png?fit=max\&auto=format\&n=bnS5T3Kw2jFffd-K\&q=85\&s=2c93ce37b2c2e79f36bf6d05fddfc083)

**1. Type your question**

Write it as you would say it to a colleague, for example "What changed this period, and why?". Press **Enter** to send. Press **Shift + Enter** to add a new line.

**2. Pick a suggestion instead**

On Home, under the field are suggestion chips about the page, such as "What changed this period, and why?" or "Which of my apps needs attention first?". Click one to put its words in the field. Nothing is sent yet, so you can edit the question first, then send it. On the RADAR new chat page, a **Recommendations** section under the field shows up to two [skills](/assistant/skills.md) worth starting now, such as the daily triage. Click one to start it.

**3. Choose the quality and the Auto setting**

The buttons in the bottom row of the field set how hard RADAR works and whether it may make small changes without asking. See the [overview](/assistant/overview.md#quality-and-auto-in-the-message-field).

You need at least one connected app and a paid plan or a running trial. Without a plan, RADAR shows "RADAR is part of the paid plans" with a **See plans** button. RADAR answers from the reviews of the apps in your workspace, and from the public app stores when you ask about a competitor.

In the side panel, the new chat screen shows the three newest chats of the workspace, teammates' included, under **Recent chats**, with a **View all** link.

## The side panel

The **Agent** button in the page bar opens RADAR in a panel beside the page. Beside the page, drag the panel's left edge (or focus it and use the arrow keys) to make it wider or narrower. On a screen narrower than 1024 pixels it is a sheet over the page instead. **Open in full page** moves the chat to RADAR's own page, **Chat options** then **New chat** starts a new one, and **Close the panel** hides it. The chat stays as you move between pages. The button is not shown on RADAR's own pages.

In the panel, RADAR is told which page you are on, with its apps and period, so "this" and "here" mean that page. Beside the Reviews feed it also sees your filters, sort and the reviews on screen.

On the Reviews page, RADAR can set the filters, sort and page for you. This is applied at once and shows as a chip such as "Filters set: unreplied, last 30 days", with **Undo**. It needs no approval because it changes the view, not your data.

### Point at content

Beside a dashboard page, click **Point at content** in the message field (or press Cmd+J, Ctrl+J on Windows), then click a chart, a row or a review. It is added to your message as a chip that starts with @. You can add up to 3 per message. Press **Esc**, or click the cross in the **Pick content to ask about** bar, to stop.

## Read an answer

A good answer starts with the bottom line in bold. Charts and short readings of them follow, then anything RADAR could not check, and one or two next steps.

![An answer with a bold bottom line, a line chart of the Android store rating with release markers, a table comparing three builds, and a suggested next step](https://mintcdn.com/appreply/bnS5T3Kw2jFffd-K/images/product/assistant/02-answer-with-charts.png?fit=max\&auto=format\&n=bnS5T3Kw2jFffd-K\&q=85\&s=d697adb133d205b207af4cf5010fa516)

### Charts

When you ask about a period of any length, RADAR looks at it over time first. The first chart of the answer is a series over that period. If you have several apps, it draws one series per app, not all apps added together. A comparison of totals, such as this period against the last, comes after it.

Questions about topics, trends and what changed get their charts in the first answer. You do not need to ask again. An answer has at most three charts.

Click a name in a chart's legend to hide or show that series. The last series that is showing cannot be hidden. The **Options** menu on a chart has:

* **View as table** or **View as chart**
* **Expand**, to see it larger
* **Copy**
* **Save as image**
* **Download CSV**, **Download XLSX** and **Download JSON** for the numbers behind the chart
* **Lock scale** or **Unlock scale**, for charts with a value axis

Lists of reviews appear as review cards. When there is more than one, you move through them one at a time.

### Plan and steps

For a longer task, such as a [skill](/assistant/skills.md), RADAR first writes a plan and shows it as a plan card. The card has a header such as **Plan in progress**, a count like 2 / 5, and the tasks. Each task is **Not started**, **In progress**, **Done** or **Skipped**. RADAR updates the plan as it learns, and the card stays open while it works. A simple question gets a direct answer with no plan.

![A finished answer with its plan card showing 4 of 4 tasks done, and the lookups folded into short lines under it](https://mintcdn.com/appreply/bnS5T3Kw2jFffd-K/images/product/assistant/03-plan-and-steps.png?fit=max\&auto=format\&n=bnS5T3Kw2jFffd-K\&q=85\&s=3b211961742ce4176393d481af313795)

Above the answer, one line says what RADAR is doing, for example "Thinking", "Reading the results" or "Writing the answer", with a timer. When it finishes, the line settles to something like "Worked 14s · 3 lookups".

The steps are folded under the answer as a short line, such as "3 lookups". Click it to see each lookup RADAR made. Use this when you want to check where a number came from.

### Checking the answer

After RADAR writes the answer to a longer task, or to a question with charts, it reads the answer once more against what you asked and what its plan promised. While it does this, the plan card and the line above the answer read **Checking the answer**, and nothing in the plan spins.

If something is missing, RADAR goes back to work and writes the answer again. The earlier draft is folded away, and the answer you see is the newest complete one.

### Links

Links in an answer, on a review card or on an approval card open in a new tab, so the chat stays where it is. Only web and email links are made clickable. If the text of a link does not show where it goes, the address is printed beside it.

## Follow up

Keep going in the same chat. RADAR remembers what it just showed you, so a follow-up such as "Show the reviews behind the top topic" works on the same data without starting again.

After an answer, the empty field often shows a suggested follow-up in gray with a **Tab** key hint. Press **Tab** to use it, or just start typing something else.

If you type while RADAR is still working, your message is queued. It shows as **Queued** above the field and is sent when the answer ends. Click **Remove** to take it back.

## Stop an answer

While RADAR works, the send arrow in the field becomes a **Stop** square. Click it to end the answer. The chat then says "Stopped. You stopped this answer." You can also type **/stop**.

An answer can also end without finishing. The line under it says why:

* **Timed out.** RADAR ran out of time. Click **Retry**.
* **Failed.** RADAR ran into a problem. Click **Retry**.
* **Not finished.** It stopped before the end of its answer. Click **Continue**. The line can instead say RADAR reached the spending limit for one answer and wrote up what it had, or that it could not confirm the answer covers everything you asked.
* **Too much data.** The answer needed more data than fits in one go. Ask for a narrower period or fewer apps.

## Attach files

You can attach a file to a message, and RADAR will use it as part of the question.

* Allowed types: **.txt**, **.md**, **.csv**, **.xlsx** (Excel), **.docx** (Word) and **.pdf**
* Up to **3 files** per message
* Up to **60,000 characters** of text in each file, and up to **5 MB** for a spreadsheet, Word file or PDF

**1. Add the file**

Do any one of these: click **+** in the field and choose **Add file**, drag the file onto the chat (the field says "Drop to attach .txt, .md, .csv, .xlsx, .docx or .pdf"), or paste it into the field.

**2. Check the chip**

Each file appears as a chip with its name. Hold the pointer over a chip to see its length. Remove a chip to take the file off.

**3. Say what you want done**

Type your question next to the file, for example "Which of these customer emails mention a crash?" and send.

How a file is used:

* RADAR reads the text of the file as material to work with, not as instructions. What you want done goes in your message. Instructions written inside the file are not followed.
* The file is read in your browser and sent with the message as text. It is not saved anywhere as a separate file. A spreadsheet arrives as its sheets in order, each as rows of values; a Word file as its paragraphs and tables; a PDF as the text of its pages.
* Every file is checked for safety before RADAR reads it. A file that tries to give RADAR instructions, such as asking it to ignore its rules or reveal its settings, is refused with the message it came with.
* Text that is hidden in a Word file is left out, because you would not see it when you open the file. A hidden sheet in a spreadsheet is read and marked as hidden.
* If a file does not fit, it is not added, and a short message names the file and the reason. This happens for another file type, a file with no text (such as a scanned PDF), a file over its size or character limit, a fourth file, or a name that starts with a dot. A spreadsheet or Word file with macros, a password-protected file, a PDF with scripts, and a file whose contents do not match its name are also refused.
* In a long chat, files from much earlier messages may be left out of what RADAR reads to keep the conversation short. RADAR then asks you to attach the file again.

## Questions from RADAR

Sometimes RADAR needs a choice from you before it goes on, for example which app to use. It shows a question card with up to four questions.

Click an option to mark it, or type your own answer in the box. With several questions, use **Next** and **Back** to move between them. Nothing is sent until you click **Confirm**. A question you leave unanswered counts as skipped, and RADAR takes the recommended option.

## Approval cards

When you ask RADAR to change something in your workspace, it does not make the change. It shows the change as an approval card in the chat. The exceptions are the view of the Reviews page, described under [The side panel](#the-side-panel), and a memory tidy that an Admin or Owner asks for, which only removes entries and keeps the earlier text (see [Context](/assistant/context.md)). The card says what will change, often as a before and after, and the buttons are at the bottom.

![An approval card that switches off three auto-reply rules, with a Reject button and an approve button named Switch off 3 rules](https://mintcdn.com/appreply/bnS5T3Kw2jFffd-K/images/product/assistant/04-approval-card.png?fit=max\&auto=format\&n=bnS5T3Kw2jFffd-K\&q=85\&s=9c2bfeede86abc767a755c625131f131)

* **Approve.** The button carries the name of the action, such as **Send reply**, **Delete topic** or **Save workflow**. Clicking it makes the change, using your own account and role. The card then becomes one line that says what was done, with a link to the page when there is one. RADAR goes on from there, and the chat notes it as "You approved: ..." so you can see what you decided.
* **Reject.** Nothing changes. The card becomes one line that says "Not approved." and names the change.
* **Who can approve.** Only the person who asked for the change can approve it. A card also needs a role high enough for the change. If you are not allowed, the card says who is, for example "Needs an Admin to approve. You are a Member."
* **Expiry.** A card nobody decides on expires after 24 hours. Ask again to get a new one. In the last 6 hours the card shows how long is left.
* **Several changes.** When RADAR proposes a group of changes, they are stacked together under a heading such as "4 changes waiting", with buttons such as **Approve all 4**, **Approve 3 of 4** and **Reject all 4**. A change that needs its own careful look is left out of the approve button and shows **Review** instead.
* **A batch of replies.** Each draft is a text field you can edit before sending. **Undo edit** puts the draft back, and **Show original** shows the review in its own language when it was translated. Tick the replies you want to send and leave the rest. One card holds at most 12 replies.
* **One change over many items.** A card such as "Switch off 3 auto-reply rules" lists each item with a tick box, and its button names the count. Untick an item to leave it as it is. A change that cannot be undone also asks you to tick a confirmation first. Such a card is never approved automatically.
* **Auto-approve.** With **Auto on**, small changes that are easy to undo are approved for you. The card then reads "Approved automatically." Replies, rules, alerts and settings always wait for you. See [Auto setting](/assistant/overview.md#auto-setting).
* **Roles and connectors.** A Member sees "Waits for an Admin." (or "Waits for the Owner.") on a change that needs a higher role. Approving needs the **AppReply actions** connector to be on. Reject always works.

If a change fails when you approve it, the card collapses to a line that starts with **Failed.** and gives the reason. If it did not finish in time, the card says "This may or may not have been applied. Check the page before asking again."

## Manage your chats

The chat list shows every chat in the workspace, newest first, under **Today**, **Previous 7 days** and **Older**. Chats started by a teammate show their initials after the title. The person who started a chat, or an Admin, can use its menu to **Rename**, **Archive** or **Delete** it. Delete asks you to confirm, because it cannot be undone. Archiving shows a short message with **Undo**. An archived chat leaves the list and cannot take new messages; open it from a link and choose **Restore** in its menu, or type **/restore**.

The **+** button in the field opens **Add context**: **Add file**, **Skills** and **Sources**, with a line such as "Usage: 3 of 30 turns today". **Sources** lists the connectors an Admin has switched on; switching one off there changes this chat only.

A few commands save clicks. Type **/** in the field to see the ones that can run now: **/new**, **/stop**, **/retry**, **/quality**, **/auto**, **/attach**, **/sources**, **/rename**, **/archive**, **/skills** and **/usage**.

> [!NOTE]
>
> RADAR has limits on how much it can be used, and the field tells you when one is close or reached. Today, each person has 30 turns and $20 per 24 hours, and a workspace has 600 turns and $20 per 24 hours and \$50 per rolling 30 days. At most 2 answers per person and 3 per workspace can run at once. When a limit is reached, the message says which one it is, whose it is, and when a message can be sent again. The [Usage](/assistant/usage.md) page shows what has been used.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.