Chat Settings adjust the default behavior of the chat page. They affect the default model for new chats, title generation, input and sending behavior, reply metadata, context compaction, file handling, and AI preference memory.
Entry Point#
Open Settings, then choose Chat. The page is organized into Default Model, Input and Sending, Display and Rendering, Context Management, File Handling, and AI Preference Memory.
These settings mainly affect later chat behavior. Existing generated messages are not rewritten when settings change.
Default Model#
New chat default model decides which chat model is preselected when you start a new conversation. Choosing System default lets the platform use the recommended available model for the current account.
If a model is missing from the list, it is usually not enabled, not allowed for the account, not chat-capable, or limited by platform policy.
Auto-generate title creates a title from conversation content. When disabled, a new conversation uses the first 20 characters of the first message as its title. This only affects later conversations.
Input and Sending#
Input and Sending controls the composer and send behavior.
| Setting | Purpose |
|---|---|
| Send shortcut | Choose Enter or Command/Ctrl + Enter. IME composition does not trigger accidental sending. |
| Input height | Set the maximum height the composer can expand to: Compact, Standard, or Loose. |
| Restore draft automatically | Restore text and attachments after a failed send so you can revise and retry. |
| Preserve conversation drafts | Keep unsent drafts when switching conversations. |
| Delete files by default when deleting conversations | Preselect file cleanup in delete dialogs; files referenced by other conversations stay available. |
If you often write long prompts, use a larger input height and preserve drafts. If you frequently clean old conversations, file cleanup can be enabled by default, but still review important files before deletion.
Display and Rendering#
Display and Rendering controls how replies and metadata appear.
| Setting | Purpose |
|---|---|
| Markdown rendering | Render AI replies as formatted Markdown. Turning it off makes replies closer to plain text. |
| Show model | Show the model used for the reply. |
| Show token usage | Show token usage for the request. |
| Show total latency | Show the total call duration. |
| Show total cost | Show call cost when billing mode supports it. Self mode does not show cost. |
For focused reading, hide metadata you do not need. For cost, model, or performance checks, keep model, token, latency, and cost visible.
Context Management#
Auto context compaction summarizes earlier messages when the conversation approaches model context limits, helping long conversations continue.
When disabled, DEEIX does not compact context automatically, and long conversations may hit model limits sooner. Use this only when you want manual control.
File Handling#
File handling mode controls how conversation files are used in later turns.
| Mode | Best for |
|---|---|
| Auto | Let the system choose based on file size, type, context budget, and retrieval capability. Recommended for most cases. |
| Full context | Put file content into context when possible. Useful for small files and precise review. |
| RAG | Retrieve relevant chunks first. Useful for large files, many files, and question-based lookup. |
Use Auto when unsure. Try Full context if a small file detail is missed. Prefer RAG for large or many documents.
AI Preference Memory#
AI Preference Memory stores long-term preferences. Each item has a name and content, for example “response style: concise and direct.”
Up to 20 preference memories can be saved. You can add, edit, or delete them. Saved preferences become long-term references for replies, not direct instructions for a single turn.
Use memory for stable habits. Put one-time requirements in the current prompt, and project-wide requirements in project instructions.
Practical Tips#
Start with system default model and Auto file handling. Adjust shortcut and input height to match your writing style. Decide whether to show tokens, latency, and cost based on whether you track usage. Store long-term habits in AI Preference Memory and one-time requirements in the current chat.