Skip to main content
AI Search is the main entry point for querying models in TouchAI. It appears as a borderless, always-on-top overlay window that floats above your other apps, accepts text and attachments, streams the model response live, and auto-hides when you move focus away.

​
Triggering the overlay

Press AltSpace (default) to show or raise the window. The shortcut is configurable in General Settings. When the window appears, focus is immediately placed in the search bar so you can start typing.
If the shortcut is already registered by another application, TouchAI sends a system notification and the shortcut will not activate. Change it in Settings → General.

​
SearchBar

The SearchBar is the input area at the bottom of the overlay. It is a rich-text editor built on Tiptap/ProseMirror and supports multi-line text, inline model tags, and attachment chips.

​
Text input

The editor expands up to three lines before becoming scrollable. When the cursor is on the first line of a single-line input, TouchAI treats that as the “cursor at start” position and enables certain backspace shortcuts (see Keyboard shortcuts below).

​
Model selection

Click the model logo on the left side of the search bar, or press @, to open the model picker dropdown. Selecting a model inserts a model tag inline in the editor and sets a per-query model override — that query will use the chosen model regardless of your default. The override stores both the model ID and the provider ID: 
interface SearchModelOverride {
  modelId: string | null;
  providerId: number | null;
}
To clear the override, press Backspace when the cursor is at the start of the input, or press Escape when the input is empty and a model tag is present.

​
Attachments

TouchAI supports two attachment types:
TypeValue
imageInline image, sent as base64 to multimodal models
fileAny file, sent as text content to models that support files
Attachments are added by pasting from the clipboard (CtrlV / CmdV). The clipboard reader accepts both image data and file paths. Each attachment is checked against the active model’s declared capabilities (supportsImages, supportsFiles). Attachments that the model does not support are marked as unsupported and cannot be submitted — a notification explains why.
If you attach a file and then switch to a model that does not support files, the attachment is marked unsupported. Remove it or switch to a compatible model before submitting.

​
Streaming response

When a query is submitted, the AI service streams chunks back to the UI. While the request is in flight, the overlay border animates as a flowing gradient cycling through blue → violet → pink → violet → blue over a 1.5 s loop. 
/* Active during streaming */
.search-view-container.loading {
  border: 2px solid transparent;
  background-image:
    linear-gradient(var(--color-background-primary), var(--color-background-primary)),
    linear-gradient(
      90deg,
      var(--color-blue-500),
      var(--color-violet-500),
      var(--color-pink-500),
      var(--color-violet-500),
      var(--color-blue-500)
    );
  background-origin: border-box;
  background-clip: padding-box, border-box;
  animation: border-flow 1.5s linear infinite;
}
Once streaming finishes, the border returns to its default static style.

​
Window behavior

Transparent & borderless

The window has no native title bar. The container uses backdrop-blur-xl and a subtle border to float above the desktop.

Always on top

The overlay stays above other windows so you can query without switching away from your work.

Auto-hide on blur

When the app loses focus and no conversation is pinned, the window hides automatically. If hidden for 5 minutes or longer, the conversation is cleared on next open.

Pin to keep open

While a conversation is active, click the pin button in the toolbar to prevent the window from hiding on blur.

​
Keyboard shortcuts

KeyAction
EnterSubmit the query
ShiftEnterInsert a newline
@Open the model picker dropdown
ArrowUpSubmit (when input is single-line)
ArrowDownOpen Quick Search panel
EscapeCancel streaming → clear model override → clear conversation → hide window (in priority order)
TabMove focus to the conversation panel (when a conversation is open)
Backspace (cursor at start)Remove the model override tag
Ctrl/CmdVPaste text or import clipboard images/files as attachments
When the model dropdown is open, ArrowUp, ArrowDown, and Enter navigate and select within the dropdown instead of the normal search bar behaviour.

Build docs developers (and LLMs) love

Get started for freeTalk to us