Chat and Code
Two sides, each with its own history, model and settings.
Ctrl+Shift+X switches between them.
Nothing set on one side reaches the other. Code works inside one folder and shows every change as a diff.
| Chat | Code | |
|---|---|---|
| Conversations | Chat history | Sessions, one list per project |
| Files | What it wrote for you | The project folder |
| Model | Its own | Its own, or the same as Chat |
| Tools | Web, browser, documents | Files, commands, code, git, plans |
| Voice | Talk | Not here |
The composer#
| Control | Does |
|---|---|
| Off / Low / Medium / High / Max | How hard the model thinks. Off turns reasoning off rather than asking for less. |
| On / Off | Whether it may use the web. |
| Model name | The model answering. Ctrl+Shift+I opens the model menu. |
| Context wheel | How full the window is. Click for the breakdown. |
Type / for slash commands: /new, /model, /compact, /settings, /web, /think, /voice, /files, /image, or any enabled skill. /permissions, /memory and /init work in Code only.
Images pasted or attached to the composer show preview cards with quick removal before sending.
Interactive questions#
When the model needs choices or clarification, it displays interactive question cards in the chat. Questions support single choice, multiple choice, or free-form text input.
A question session bar tracks progress through multi-part questionnaires with next, submit, and skip controls.
Queueing messages#
While a reply is generating, type a follow-up and press Enter to steer: the note goes into the running reply. Press Ctrl+Enter to queue it for the next turn, or Ctrl+Shift+Enter to interrupt and send now. Queued messages are saved and persist if you restart the app.
Exporting chats#
The sidebar session menu lets you export a conversation to Markdown (.md) or standalone HTML (.html). The export includes reasoning steps, tool activity, code diffs and generation metrics.
When a conversation gets long#
When a reply reaches the context limit and the model has room, Draggy automatically expands the window to the next bucket and continues generating. If the conversation reaches 60% of what the model can hold, Draggy folds the older part of the conversation into a summary, and the conversation carries on. For paid cloud models, that point is capped at 64,000 tokens unless you set a limit.
/compact does it now. /compact-limit 20k moves the point, and auto puts it back.