# Send messages and watch a bot work

The chat screen, the message box, sending and stopping, what a bot shows while it works, and how replies look.

Canonical page: https://themachine.myapps.ai/docs/desktop/chat/sending-messages

## The chat screen

### The chat header

![The chat header: bot name, find, export, All threads, usage chip, model chip, and the bot's computer button](/images/docs/desktop/chat-header.webp)

Across the top of every chat, from left to right:

| Item | What it does |
|---|---|
| **Avatar** | Opens the bot's profile card. See [below](#the-profile-card). |
| **Name** | Click it to open the bot's [settings](/docs/desktop/bots/bot-settings). The pencil next to it renames the bot. |
| **Live** | Talk with the bot out loud, in real time. See [Make a Live call](/docs/desktop/live/make-a-call). |
| **Find in conversation** (magnifying glass) | Search this chat. See [Find, share, and export](/docs/desktop/chat/find-share-export). |
| **Export conversation** (share icon) | Compact, share, or export. Same page. |
| **All threads** | See, search, rename, and switch this bot's threads, or start a new one. |
| **Usage chip** | What this thread has cost, and how full its context is. See [below](#the-usage-chip). |
| **Model chip** | Which AI model this thread uses. See [Choosing a model](/docs/desktop/models/choosing-a-model). |
| **Record a meeting** (notepad icon) | Record and transcribe a meeting, then ask the bot for notes. See [Voice](/docs/desktop/chat/voice#record-a-meeting). |
| **Bot's computer** (monitor icon) | Look inside the computer this thread runs on. See [The bot's computer](/docs/desktop/cloud-computer/overview). |

When the chat is narrow, for example with the bot's computer open beside it,
the least-used icons step aside to make room: first the usage chip and
**Record a meeting**, then **Export** and **All threads**. In the phone layout,
they move into a single **…** menu.

A few other things can show in the header:

- **Where it works.** On a [This PC](/docs/desktop/this-pc/overview) thread, a
  line under the bot's name shows the computer and folder it works in, and
  what it's doing there right now.
- **Kick.** When a turn has been running for a long time, a **Kick** button
  appears next to the name, in case the bot is stuck rather than slow. See
  [The bot's computer](/docs/desktop/cloud-computer/overview).
- **Routine runs.** When you open a thread an
  [automation](/docs/desktop/automations/overview) ran in, a strip under the
  header says **Run details** with the routine's name, plus **Back to
  results** and **Run logs**.

> **Simple mode**
> With **Simple mode** on, the header has no model chip, and the chat hides the
> bot's tool steps. Your bots work exactly the same. Turn it on or off in
> [Settings → General](/docs/desktop/settings/general).

### The profile card

![The profile card that opens from the bot's avatar](/images/docs/desktop/bot-profile-card.webp)

Click the bot's avatar to see who it is at a glance:

- its face, name, and role,
- what it's doing: **Working**, **Ready**, or **Last run failed**,
- **Looks after**: the goals it takes care of, if it has any.

Underneath are the things you'd reach for next:

| Item | What it does |
|---|---|
| **New thread** | Start a fresh conversation with this bot. |
| **Responsibilities** | Goals, schedule, and what wakes it. |
| **Kick** | Unstick a turn that seems frozen. |
| **All settings** | Open the bot's [settings](/docs/desktop/bots/bot-settings). |

### The usage chip

Shows how much this thread has cost so far, and how much of the model's
**context** is used. Context is how much of the conversation the model can
keep in mind at once.

- Hover it for details: number of turns, words in and out, and context used.
- It turns **amber**, then **red**, as the thread gets long.
- When it's amber or red, the model is re-reading a lot every time, which is
  slower and more expensive. Start a new thread, or [compact this
  one](/docs/desktop/chat/find-share-export#compact-a-long-thread).
- Click it to open this bot's **Usage** settings.

### The message box

![The message box with its controls](/images/docs/desktop/composer.webp)

At the bottom of the chat, from left to right:

| Control | What it does |
|---|---|
| **Paperclip** | Attach files. See [Attachments](/docs/desktop/chat/attachments). |
| **Approval button** | Whether the bot asks before acting. See [Approvals](/docs/desktop/models/approvals). |
| **Text box** | Type here. It grows as you write. Type `/` for [commands](/docs/desktop/chat/message-actions#slash-commands) or `@` to [mention a bot](/docs/desktop/chat/message-actions#mentions). |
| **Stop** (square) | Appears while the bot works. Stops the current turn. |
| **Effort bars** | How hard the model thinks (some models only). See [Reasoning effort](/docs/desktop/models/reasoning-effort). |
| **Voice message** (sound wave) and **Dictate** (microphone) | Talk instead of type. See [Voice](/docs/desktop/chat/voice). |
| **Send** (arrow) | Appears once you've typed. While the bot works, it shows a clock instead: see [below](#send-while-the-bot-is-working). |

Above the box you may see a chip showing where the thread works: **Cloud**
or **This PC**. See [This PC](/docs/desktop/this-pc/overview).

On a phone, a **Live** button sits next to the message box while it's empty.

## Send a message

![Typing a request and pressing Enter; the bot starts working](/images/docs/desktop/anim-send.webp)

Type and press **Enter**. To start a new line without sending, press
**Shift+Enter**.

- **Not sure where to start?** A brand-new thread offers three starters:
  **What can you do for me?**, **Show me what's on your computer**, and
  **Help me set you up for my work**. Clicking one puts it in the message box
  for you to send or change.
- **Fix your last message:** with the box empty, press **↑** to edit the last
  message you sent.
- **Drafts are saved per thread.** Switch away halfway through typing and your
  text is still there when you come back.
- **If a message fails to send,** your text goes back in the box, so nothing
  is lost. If you've already started typing something new, a red strip says
  **Not sent:** with your message, a **Retry** button, and **×** to dismiss it.
- **While files upload,** the box says *Attaching files…* and waits until
  they're done.

### Send while the bot is working

You don't have to wait for the bot to finish. Type and send at any time, for
example to add a detail you forgot.

While a bot works, the box says *"Scout is working — sends when this turn
finishes"* (with your bot's name), and the send button shows a **clock**
instead of an arrow. Your message goes into the chat and the bot picks it up
as soon as the current turn ends.

In a [group chat](/docs/desktop/bots/group-chats), the box says *"… is working
— Enter queues your message"*.

## Stop a bot

Any of these stops the current turn:

- the **Stop** button on the work card in the chat (see [below](#what-its-doing-right-now)),
- the square **Stop** button in the message box,
- on a phone, the red **Stop** button in the chat header,
- sending a message that only says **stop**, **cancel**, **abort**, **halt**,
  **quit**, or **enough** (also *"please stop"* or *"stop it now"*). It works
  exactly like the button: your message shows in the chat, but the bot never
  reads it.

Stopping only stops the current turn. The thread, its files, and its computer
are kept, and you can carry on with your next message.

If a bot looks stuck and Stop doesn't help, use **Kick**. See [The bot's
computer](/docs/desktop/cloud-computer/overview).

## Watch a bot work

![A bot working: the Stop button, a running tool chip, and the working label with its timer](/images/docs/desktop/chat-working.webp)

### What it's doing right now

While a bot works, a card in the chat says **Scout is working** (with your
bot's name), a timer, and a line saying what it's doing right now. The card
has two buttons:

| Button | What it does |
|---|---|
| **Watch** | Opens the bot's computer in the [Live view](/docs/desktop/live/live-view), so you can see it work. |
| **Stop** | Stops the current turn. |

On a phone, the bot shows a chat bubble with moving dots instead, and Stop is
in the header.

The line on the card can say:

| Label | Meaning |
|---|---|
| *Setting up this bot's computer…* | A brand-new thread is getting its computer. Takes a little longer. |
| *Waking its computer…* | The thread's computer was asleep. Usually a few seconds, occasionally a minute or two. |
| *Thinking* | Working out what to do. |
| *Searching the web* | Looking something up online. |
| *Reading a file* / *Editing a file* | Working with files on its computer. |
| *Running a command* | Running a program on its computer. |
| *Using the computer* / *Looking at the screen* | Working with apps or the screen, usually on [This PC](/docs/desktop/this-pc/overview). |
| *Listening to your audio…* / *Watching your video…* | Taking in an audio or video attachment. |
| *Writing…* | Writing its reply. |

The bot's dot in the sidebar pulses blue while it works.

### Tool chips

![The steps of a finished turn, with one tool chip expanded to show its input and result](/images/docs/desktop/tool-chips.webp)

Every action a bot takes, like searching, reading a file, or running a
command, appears as a small **tool chip** in the chat with its status:
running, completed, or failed.

Click a chip to see:

- **Input**: exactly what the bot asked for,
- **Result**: what came back.

This is the best way to understand what a bot actually did.

The model's own reasoning, when shown, appears as a fold-out block. While the
bot is thinking, the block stays open as **Thinking**, with a live preview
that follows the newest thought (scroll up in it to stop following). When the
thinking ends, it folds itself to **Thought for 12s** (or just **Thought**).
Click it to open or close it; once you do, it stays the way you left it.

Don't want to see them? Turn off **Settings → Appearance → Tool calls**, or
turn on Simple mode.

When a bot saves something to its [memory](/docs/desktop/bots/skills-and-memory),
a small note says **Saved to memory** with the file name, and *"— say /forget
if it shouldn't keep this"*. This note shows even when tool chips are hidden.

### When the turn ends

![Expanding a finished turn's steps, then opening one tool chip](/images/docs/desktop/anim-tool-chip.webp)

- The bot's in-between notes fold into **Worked for 2m 10s** (or however long
  it took).
- Its actions fold into a row like **6 steps · web search, write, bash**.
- The final reply appears below.

Click either folded line to open it again.

Replies appear word by word as they're written. A soft pop plays when a bot
finishes (turn it off in **Settings → General → Reply sound**). If Machine
Desktop is in the background, you get a [notification](/docs/desktop/settings/notifications).

### The line under a finished reply

![The line under a finished reply, with Replay and Share](/images/docs/desktop/run-footer.webp)

When a turn did some work (at least one step), a quiet line under the bot's
latest reply sums it up, for example:

*Worked 2m 10s · 6 steps · $0.04*

It shows how long the turn took, how many steps it made, and what the turn
cost, when that's known. (It doesn't name the model: the model chip in the
chat header shows that.) Then two buttons:

| Button | What it does |
|---|---|
| **Replay** | Opens the bot's computer, so you can watch the steps it took in the [Live view](/docs/desktop/live/live-view). |
| **Share** | Saves the run as one web page (an `.html` file) anyone can open: what you asked, each step with the screenshots the bot saw, and its answer. No account or link needed. |

### If a turn fails

If something goes wrong, a red box with a warning sign explains what
happened, and the bot's sidebar dot turns red.

- On the latest error, click **Retry** to try again.
- Click **×** to hide an error you've dealt with.

Common errors are covered in [Troubleshooting](/docs/desktop/troubleshooting).
You can also use **Regenerate response** or edit your message. See [Message
actions](/docs/desktop/chat/message-actions).

## How replies look

- **Tables, lists, checklists, and math** display properly. Hover a table and
  click **Copy as CSV** to paste it into a spreadsheet.
- **Code** appears in boxes showing the language and number of lines, with
  **Copy**, **Download snippet as file**, and **Wrap long lines**.
- **Long messages you send** fold behind **Show full message**.
- **Spoilers:** text between `~~` marks is hidden until you click **Reveal
  spoiler**.
- **Live website previews:** when a bot starts a website on its computer, a
  preview appears in the chat, with **Reload preview** and **Open in
  browser**. When the page is an HTML file the bot wrote, a **File** /
  **Server** switch also appears: **File** (the default) shows the file
  itself, so it still shows after the server has stopped; **Server** shows
  the running server, for pages that need it.
- **Images from the web** are hidden until you click **Load image**. This
  stops websites from knowing you opened the thread.
- **Files the bot made** appear as chips with **Save a copy**. A file on your
  own computer (on a [This PC](/docs/desktop/this-pc/overview) thread) also
  has **Show in folder**, which opens its folder.
- **Messages from other bots** are labelled with the bot's name and how they
  arrived. See [Message actions](/docs/desktop/chat/message-actions#messages-from-other-bots).
- **Confetti:** when a bot finishes something worth celebrating, it may throw
  a little. (With reduced motion on, you get a 🎉 instead.)

### Moving around a long chat

- Scroll up to read earlier messages. **Show earlier messages** (or **Load
  earlier messages**) brings in more.
- After you jump to an old message, for example from a search, **Show later
  messages** brings back the ones after it.
- When you scroll up, the chat stops following new messages. Click **Jump to
  latest** to go back to the bottom.
- Hover a message to see when it was sent. Days are separated by **Today**,
  **Yesterday**, or a date.
