The Chat Window

The Chat Window

The screen you spend all your time on. Everything here is reachable from the composer at the bottom, and most of it has a keyboard route.

The composer

The box you type in, and the row of controls under it.

Control	What it does
+	Attach files, same as dragging them in
Microphone	Dictates into the box. Not the same as Voice Mode
Fast / Balanced / Deep	How hard the model thinks. Click to cycle
Web: Auto / On / Off	Whether the model may search. Click to cycle
Code	Greyed out until you enable code execution in Settings → Tools
Model name	The model answering. Click it, or press Ctrl+K, to change
Send	Enabled once there is something to send

The two mode pills matter more than they look. Fast turns reasoning off rather than asking for less of it, which is the difference between a one word answer arriving now and arriving after seven seconds of silent thinking. Deep is genuinely slower and only earns it on a hard question. Both are explained in Settings.

The microphone is dictation: it records, transcribes on your machine, and drops the text into the box for you to edit before sending. Voice mode is a conversation instead, and lives behind the sidebar's Talk button.

Slash commands

Type / as the first character and a menu appears. Arrow keys move, Tab or Enter runs the highlighted one, Escape clears the box.

Command	What it does
/new	Start a new conversation
/model	Open the model picker
/web	Cycle the web search mode
/think	Cycle the thinking mode
/voice	Start dictation
/code	Jump to the code execution setting
/files	Open the file picker
/settings	Open settings

A space closes the menu, so "/dev/null is not a file" is a sentence and Enter sends it rather than running something.

Attaching files

Drag them in, paste them, or use +. Draggy reads them itself rather than handing a path to the model, so it works with a model that has never heard of your filesystem.

Kind	Extensions	Limit
Documents	docx, pptx, xlsx, pdf	25 MB
Images	png, jpg, jpeg, gif, webp, bmp	20 MB
Text and code	txt, md, csv, json, yaml, py, ts, rs and around sixty more	1 MB

Images need a model with vision. If yours has none you are told so, rather than given a confident description of an image nobody looked at. A PDF that is a scan has no text to extract, and Draggy says that instead of guessing, because it does not run OCR.

While an answer is arriving

Thinking appears first, in a collapsible block above the reply, and tool calls appear as they happen. Escape stops generation and keeps what has arrived.

Under a finished reply is a row of numbers: tokens per second, how many were generated, how long the first one took, how much of the context window is in use, and how much of the model is on the GPU. That last one is the number to watch. A model reporting 100% on GPU answers at reading speed; at 60% it works but several times slower, and stepping down a size is usually better than living with it.

Changing what was said

• Copy puts the reply on the clipboard as Markdown.
• Regenerate asks again. The old answer is kept: arrows appear above the reply and you can move between versions.
• Edit, on your own messages, rewrites the question and runs the conversation again from that point.

Exporting saves the version you were looking at, not the last one generated.

When a conversation gets long

A model can only hold so much at once. As a conversation approaches that limit Draggy condenses the earlier part into notes and carries on, and marks the place with Earlier messages condensed.

Nothing is deleted. Every message stays on screen and stays searchable; the model is working from a summary of the early part rather than the words. The difference between "it forgot" and "it condensed" is worth knowing when an answer suddenly loses a detail from an hour ago: scroll up, quote the line, and it is back.

History, search and export

Chat History in the sidebar lists every conversation, newest first, with a search box that matches titles and message text.

Each row has two buttons, and neither needs a hover to find:

• Export as Markdown writes the conversation to created_files/ and shows you the file, named after the conversation and the date.
• Delete removes it immediately. There is no copy anywhere else.

An export is what you read: both sides, under headings, with the names of anything you attached. Thinking, tool steps and timings are left out, because they are how the answer was reached rather than the answer, and none of them mean anything outside Draggy.

Keyboard

Key	Does
Enter	Send
Shift+Enter	New line
Ctrl/Cmd+N	New conversation
Ctrl/Cmd+K	Model picker
Escape	Stop generating, or close the model picker
/	Slash menu, then arrows and Tab

The sidebar

Collapsed to icons until you point at it. New Chat, Chat History, Created files (everything the assistant wrote, see Tools and Capabilities), Talk (Voice Mode) and Settings.