Chat settings control how new conversations start and how replies are shown. They apply to future chats and do not rewrite existing messages.
Where to find it#
- Open the user menu at the bottom of the sidebar.
- Choose Settings.
- Select Chat.
The page is grouped into Conversation basics, Input and sending, Display and rendering, Context management, File processing, and AI preference memory.
Choose the default model for new chats#
- Find Default model for new chats under Conversation basics.
- Open the model list and select a model, or keep System default.
- Leave it as System default to use the first available system-recommended model.
Only chat-capable models appear in the list. If a model is missing, it is not enabled, not allowed for your account, or not a chat model.
Configure title and label generation#
Under Conversation basics:
- Turn Auto-generate titles on or off. When off, a new chat uses the first 16 characters of the first message as its title.
- Turn Auto-generate labels on or off. When off, conversation labels are no longer generated automatically, which reduces auxiliary model calls and the tokens they use.
Both options apply to new chats only.
Adjust input and sending#
Under Input and sending:
- Choose a Send shortcut: Enter or the platform modifier plus Enter.
- Choose an Input area height: Compact, Standard, or Loose. This caps how tall the composer can grow when it expands automatically.
- Turn Restore draft on failure on to bring back the text and attachments of a failed send.
- Turn Preserve conversation drafts on to keep unsent content per conversation when you switch chats.
- Turn Reuse model configuration on to auto-fill the model parameters you used last time when a new chat uses the same model.
- Turn Delete files by default when deleting chats on to preselect file deletion in the delete dialog. Files still referenced by other conversations are kept.
The send shortcut line shows the actual key for your platform, for example Enter or Ctrl+Enter / Command+Enter.
Choose what appears under replies#
Under Display and rendering:
- Turn Markdown rendering on to format replies as Markdown, or off for plain text.
- Turn Auto-expand deep thinking on to open the active thinking step while a reply is generated.
- Turn Auto-expand tool calls on to open the active tool call while it runs.
- Turn Show called model, Show token usage, Show total latency, and Show total cost on or off.
Show total cost is disabled and hidden in self-use mode. The billing mode is set by the platform; see Subscription.
Set the chat area width and font#
Under Display and rendering, below the metadata switches:
- Pick a Chat area width: Compact, Standard, or Wide.
- Pick a Chat font: Default, Serif, Sans, or Mono.
- Pick a Chat font weight: Regular, Medium, Semibold, or Bold.
Manage context#
Under Context management:
- Turn Auto-compact context on to summarize older turns and keep the most recent turns in full when a conversation exceeds the turn or token threshold. This switch appears only when an administrator has enabled Context compression in the admin console.
- Turn Pass back reasoning context on so supported Chat Completions reasoning models include prior reasoning content in later turns.
With Auto-compact context off, long conversations reach the model context limit sooner.
Choose how files are processed#
Under File processing:
- Open File processing mode.
- Select one of the modes below.
| Mode | Behavior |
|---|---|
| Auto | The system chooses full-context injection or RAG based on file size, type, context budget, and retrieval capability. |
| Full context | Later turns try to include conversation file content directly. Best for precise analysis of small files. |
| Vector retrieval | Later turns retrieve relevant file chunks first, then fall back to full-context injection when retrieval is unavailable. |
The line under the selector states the current mode, for example Current: Auto.
Manage AI preference memory#
AI preference memory stores long-term preferences that guide replies.
- Click Add preference.
- Enter a Name, for example
Reply style. - Enter the Preference content, for example
concise and direct, avoid filler. - Click Add.
The counter above the list shows how many entries you have, up to 20. When it is full, the button reads Full.
To change an entry, click the edit icon, update the content, and click Save. To remove an entry, click the delete icon and confirm Delete this preference?.
Saved preferences apply to future conversations. One-off requirements belong in the current message.
Option reference#
| Section | Option | Values |
|---|---|---|
| Conversation basics | Default model for new chats | System default or a chat-capable model |
| Conversation basics | Auto-generate titles | On / off |
| Conversation basics | Auto-generate labels | On / off |
| Input and sending | Send shortcut | Enter or modifier + Enter |
| Input and sending | Input area height | Compact, Standard, Loose |
| Input and sending | Restore draft on failure | On / off |
| Input and sending | Preserve conversation drafts | On / off |
| Input and sending | Reuse model configuration | On / off |
| Input and sending | Delete files by default when deleting chats | On / off |
| Display and rendering | Markdown rendering | On / off |
| Display and rendering | Auto-expand deep thinking | On / off |
| Display and rendering | Auto-expand tool calls | On / off |
| Display and rendering | Show called model | On / off |
| Display and rendering | Show token usage | On / off |
| Display and rendering | Show total latency | On / off |
| Display and rendering | Show total cost | On / off (disabled in self-use mode) |
| Display and rendering | Chat area width | Compact, Standard, Wide |
| Display and rendering | Chat font | Default, Serif, Sans, Mono |
| Display and rendering | Chat font weight | Regular, Medium, Semibold, Bold |
| Context management | Auto-compact context | On / off (needs admin Context compression) |
| Context management | Pass back reasoning context | On / off |
| File processing | File processing mode | Auto, Full context, Vector retrieval |
Common problems#
| Problem | What to check |
|---|---|
| A model is missing from the default model list. | The model is not enabled, not chat-capable, or not available to your account. |
| Show total cost cannot be turned on. | The platform is in self-use mode, so cost is hidden. |
| Auto-compact context does not appear. | An administrator has not enabled Context compression in the admin console. |
| Add preference is disabled. | You reached the limit of 20 entries. Delete one first. |
| A change does not affect an existing chat. | Chat settings apply to new behavior. Existing messages are not rewritten. |