Chat
A conversation is a thread of messages. Every reply streams from the model as it is written.
Send a message#
- Type in the composer and press Enter. Shift+Enter adds a line. Switch Enter to send off in Settings → General to send with Ctrl/⌘+Enter instead.
- Press Esc or use Chat → Stop Generation to stop a reply. The text written so far stays.
- After the first reply, Assistant writes a title. Click the title in the header to edit it. Switch this off in Settings → General.
- Type
/in an empty composer to see the commands.
Attachments. Use the paper-clip button or drop a file on the composer. Text files (code, Markdown, CSV, JSON, logs) up to 512 KB become part of the message. Images go to a model that can see (a model with the vision flag). Images are reduced to 1280 px before they are sent.
Edit and branch#
- Edit a message of yours and send it again. The old version stays as a branch. Press ↑ in an empty composer to edit your last message.
- Regenerate a reply to get another one under the same message.
- The
< 1/2 >control under a message moves between the versions. The thread follows the version you select. - Delete removes a message and everything under it.
Chat → Activity Log… lists everything that happened in the conversation: messages, edits, setting changes and, in a group, the decisions of the coordinator.
Conversation settings#
View → Conversation Settings Panel opens the panel on the right:
| Setting | Effect |
|---|---|
| Model | Any model of an engine that is online. See Models and engines. |
| Temperature, Top P, Max Tokens, Reasoning Effort | How the model samples its answer. Profile default uses the value of Settings → General. |
| System Prompt | Instructions the model reads before the conversation. |
| Tools for this chat | Which tools the model may call in this conversation. |
| Tags / Folders | Free words to group conversations. The sidebar filters by tag. |
Save current settings as persona turns the panel into a persona.
Thinking and tools#
Some models think before they answer. The thinking appears in a collapsible box above the reply. A tool call appears the same way, with the request, the result and the time it took.
Replies render Markdown: headings, lists, tables, code with highlighting, math and Mermaid diagrams. Use the copy button on a code block. The i button under a reply shows its statistics: tokens, time and speed.
Long conversations#
The meter in the header shows how full the context window of the model is. When it reaches the Compaction threshold (Settings → Engines → Server, 60 % by default), Assistant summarises the older messages and keeps the summary in place of them. You can run this yourself with Chat → Compact Conversation.
Sidebar#
- The search box matches titles, model names and message text.
- Chat → Pin Conversation keeps a conversation at the top.
- A dot marks a conversation with a reply you have not seen, for example from an automation.
- Tag chips filter the list.
Temporary chat#
File → New Temporary Chat opens a chat that is not saved. Nothing goes to the disk or to memory.
Export and copy#
| Item | Where |
|---|---|
| One conversation as Markdown or JSON | File → Export as…, or the conversation panel |
| The whole conversation to the clipboard | File → Copy as Markdown |
| Everything (settings, personas, recipes, conversations, memory) | Settings → General → Data → Export |
See Data and privacy.
Keyboard shortcuts#
| Keys | Action |
|---|---|
| Ctrl/⌘ + ⇧ + O | New chat |
| Ctrl/⌘ + K | Command palette |
| Enter | Send (when Enter to send is on) |
| ↑ | Edit your last message (empty composer) |
| Esc | Stop generation / close panel |
| Ctrl/⌘ + R | Reload |
| Alt + F / C / V / S / H | Open a menu |
Notifications#
In the desktop app, a reply that finishes while the window is minimised or in the tray shows a desktop notification.