add markdown parsing + latex + tree-sitter highlighting for codeblocks in llm responses

Helpers/Picker.qml

@@ -126,33 +126,29 @@ Item {
 	focus: true
 	opacity: 0
 
-	Behavior on opacity {
-		Anim {
-			duration: 300
-		}
-	}
 	Behavior on rsx {
 		enabled: !selectArea.pressed && root.mode === "select"
 
-		ExAnim {
-		}
+		ExAnim {}
 	}
 	Behavior on rsy {
 		enabled: !selectArea.pressed && root.mode === "select"
 
-		ExAnim {
-		}
+		ExAnim {}
 	}
 	Behavior on sh {
 		enabled: !selectArea.pressed && root.mode === "select"
 
-		ExAnim {
-		}
+		ExAnim {}
 	}
 	Behavior on sw {
 		enabled: !selectArea.pressed && root.mode === "select"
 
-		ExAnim {
+		ExAnim {}
+	}
+	Behavior on opacity {
+		Anim {
+			duration: 300
 		}
 	}
 
@@ -352,8 +348,7 @@ Item {
 		y: selectionRect.y - root.realBorderWidth
 
 		Behavior on border.color {
-			Anim {
-			}
+			Anim {}
 		}
 	}
 
@@ -402,6 +397,5 @@ Item {
 		onUndoRequested: annotations.undo()
 	}
 
-	component ExAnim: Anim {
-	}
+	component ExAnim: Anim {}
 }

Modules/Notifications/Sidebar/Chat/ChatContent.qml

@@ -70,7 +70,7 @@ Item {
 		fadeAmount: 0.05
 		fadeThreshold: Tokens.padding.medium
 		model: root.chatData.messagesModel
-		spacing: Tokens.spacing.medium
+		spacing: 0
 		verticalLayoutDirection: VerticalFadeListView.BottomToTop
 
 		add: Transition {
@@ -127,26 +127,38 @@ Item {
 		visible: !root.chatData
 	}
 
-	IconButton {
-		id: scrollToBottom
-
+	Item {
 		anchors.bottom: input.top
 		anchors.bottomMargin: Tokens.spacing.extraLarge
 		anchors.horizontalCenter: parent.horizontalCenter
-		font.pointSize: Tokens.font.size.large
-		icon: "arrow_downward"
-		isRound: true
-		padding: Tokens.padding.extraSmall
+		implicitHeight: scrollToBottom.implicitHeight
+		implicitWidth: scrollToBottom.implicitWidth
 		scale: list.visibleArea.yPosition + list.visibleArea.heightRatio < 1 && list.contentHeight > list.height ? 1 : 0
-		type: IconButton.Filled
 		visible: scale > 0
 
 		Behavior on scale {
 			Anim {}
 		}
 
-		onClicked: {
-			list.positionViewAtBeginning();
+		Elevation {
+			anchors.fill: parent
+			level: 2
+			radius: scrollToBottom.radius
+		}
+
+		IconButton {
+			id: scrollToBottom
+
+			anchors.centerIn: parent
+			font.pointSize: Tokens.font.size.large
+			icon: "arrow_downward"
+			isRound: true
+			padding: Tokens.padding.extraSmall
+			type: IconButton.Tonal
+
+			onClicked: {
+				list.positionViewAtBeginning();
+			}
 		}
 	}
 

Modules/Notifications/Sidebar/Chat/Content/Actions.qml

@@ -12,10 +12,11 @@ Item {
 	required property ChatGeneration current
 	required property TextEdit edit
 	required property bool hovered
-	readonly property bool isUser: message.role === ChatMessage.Role.User
+	required property bool isUser
 	required property ChatMessage message
+	required property LlmSegment segment
 
-	implicitHeight: root.current.content !== "" ? editButton.implicitHeight : 0
+	implicitHeight: editButton.implicitHeight
 	implicitWidth: generationTools.implicitWidth + editTools.implicitWidth + Tokens.spacing.small
 
 	ButtonRow {
@@ -66,8 +67,7 @@ Item {
 		visible: opacity > 0
 
 		Behavior on opacity {
-			Anim {
-			}
+			Anim {}
 		}
 
 		IconButton {
@@ -78,8 +78,7 @@ Item {
 			visible: !root.isUser
 
 			Behavior on scale {
-				Anim {
-				}
+				Anim {}
 			}
 
 			onClicked: {
@@ -94,8 +93,7 @@ Item {
 			scale: root.edit.readOnly ? 1 : 0
 
 			Behavior on scale {
-				Anim {
-				}
+				Anim {}
 			}
 
 			onClicked: Quickshell.clipboardText = root.current.content
@@ -107,8 +105,11 @@ Item {
 			icon: root.edit.readOnly ? "edit" : "check"
 			inactiveColor: Colors.palette.m3tertiary
 			inactiveOnColor: Colors.palette.m3onTertiary
+			visible: root.isUser
 
-			onClicked: root.edit.readOnly = !root.edit.readOnly
+			onClicked: {
+				root.edit.readOnly = !root.edit.readOnly;
+			}
 		}
 	}
 }

Modules/Notifications/Sidebar/Chat/Content/CodeBlockView.qml

@@ -0,0 +1,157 @@
+import QtQuick
+import QtQuick.Layouts
+import Quickshell
+import ZShell.Config
+import ZShell.Llm
+import qs.Components
+import qs.Services
+
+CustomClippingRect {
+	id: root
+
+	required property string language
+	required property string code
+	property bool copied: false
+	property color codeColor: Colors.palette.m3onSurfaceVariant
+	property color codeBackgroundColor: Colors.palette.m3surfaceContainerLow
+	property color codeHeaderColor: Colors.palette.m3outline
+	property color codeAccentColor: Colors.palette.m3primary
+
+	// Highlighter spans for the current code; refreshed when the code or
+	// its language changes.
+	property var codeSpans: []
+
+	function refresh() {
+		codeSpans = CodeHighlighter.highlight(root.code, root.language);
+	}
+
+	function roleColor(kind) {
+		switch (kind) {
+		case "comment":
+			return Colors.palette.m3outline;
+		case "string":
+			return Colors.palette.m3tertiary;
+		case "string.key":
+			return Colors.palette.m3secondary;
+		case "number":
+		case "constant":
+			return Colors.palette.m3tertiaryFixed;
+		case "keyword":
+			return root.codeAccentColor;
+		case "type":
+			return Colors.palette.m3secondary;
+		case "function":
+		case "method":
+			return Colors.palette.m3onSurface;
+		case "macro":
+		case "preproc":
+			return Colors.palette.m3secondaryContainer;
+		case "operator":
+		case "property":
+			return root.codeColor;
+		case "label":
+		case "attribute":
+			return Colors.palette.m3tertiaryFixedDim;
+		default:
+			return root.codeColor;
+		}
+	}
+
+	function escapeHtml(text) {
+		return text.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;");
+	}
+
+	// Wraps the code in <font color> tags following the highlighter spans.
+	function highlightedHtml(code, spans) {
+		let out;
+		if (!spans.length) {
+			out = escapeHtml(code);
+		} else {
+			out = "";
+			let pos = 0;
+			for (let i = 0; i < spans.length; i++) {
+				const span = spans[i];
+				if (span.start > pos)
+					out += escapeHtml(code.slice(pos, span.start));
+				out += `<font color="${roleColor(span.kind)}">` + escapeHtml(code.slice(span.start, span.start + span.length)) + "</font>";
+				pos = span.start + span.length;
+			}
+			if (pos < code.length)
+				out += escapeHtml(code.slice(pos));
+		}
+		// RichText collapses HTML whitespace: break newlines, and protect
+		// line-leading indentation with nbsp (internal spaces stay normal
+		// so long lines can still wrap).
+		return out
+			.replace(/(^|\n)[ \t]+/g, ws => ws.replace(/[ \t]/g, "&nbsp;"))
+			.replace(/\n/g, "<br>");
+	}
+
+	implicitWidth: headerRow.implicitWidth + headerRow.anchors.leftMargin + headerRow.anchors.rightMargin
+	implicitHeight: headerRow.anchors.topMargin + headerRow.implicitHeight + codeText.implicitHeight + codeText.anchors.margins * 2
+	color: root.codeBackgroundColor
+	radius: Tokens.rounding.medium
+
+	onLanguageChanged: refresh()
+	onCodeChanged: refresh()
+
+	CustomText {
+		id: codeText
+
+		anchors.left: parent.left
+		anchors.top: headerRow.bottom
+		anchors.right: parent.right
+		anchors.margins: Tokens.padding.small
+		color: root.codeColor
+		font.family: Config.appearance.font.family.mono
+		font.pointSize: Tokens.font.size.small
+		textFormat: Text.RichText
+		text: root.highlightedHtml(root.code, root.codeSpans)
+		wrapMode: Text.WrapAtWordBoundaryOrAnywhere
+	}
+
+	RowLayout {
+		id: headerRow
+
+		anchors.top: parent.top
+		anchors.left: parent.left
+		anchors.right: parent.right
+		anchors.topMargin: Tokens.padding.small
+		anchors.leftMargin: Tokens.padding.small
+		anchors.rightMargin: Tokens.padding.small
+		spacing: Tokens.spacing.small
+
+		CustomText {
+			Layout.alignment: Qt.AlignVCenter
+			color: root.codeHeaderColor
+			font.pointSize: Tokens.font.size.small
+			text: root.language
+			visible: root.language.length > 0
+		}
+
+		Item {
+			Layout.fillWidth: true
+		}
+
+		IconButton {
+			icon: root.copied ? "check" : "content_copy"
+			inactiveColor: "transparent"
+			inactiveOnColor: root.codeHeaderColor
+			type: IconButton.Text
+
+			onClicked: {
+				Quickshell.clipboardText = root.code;
+				root.copied = true;
+				copyResetTimer.restart();
+			}
+
+			Timer {
+				id: copyResetTimer
+
+				interval: 1500
+
+				onTriggered: root.copied = false
+			}
+		}
+	}
+}

Modules/Notifications/Sidebar/Chat/Content/ContentBubble.qml

@@ -0,0 +1,120 @@
+import QtQuick
+import QtQuick.Layouts
+import ZShell.Llm
+import ZShell.Config
+import qs.Components
+import qs.Services
+
+Item {
+	id: root
+
+	required property bool isUser
+	required property LlmSegment segment
+	required property bool hovered
+	required property Repeater repeater
+	required property int index
+	required property ChatMessage message
+	required property ChatGeneration current
+
+	// Widest the bubble may grow to; assistant bubbles always use it so
+	// the markdown blocks have a deterministic width to wrap against.
+	readonly property real contentMaxWidth: width - Tokens.spacing.extraSmall - Tokens.spacing.extraLarge
+
+	signal edit(text: string)
+
+	implicitHeight: bubble.implicitHeight + actionsRow.implicitHeight + actionsRow.anchors.topMargin
+
+	CustomRect {
+		id: bubble
+
+		radius: Tokens.rounding.medium
+		color: root.isUser ? Colors.palette.m3primary : Colors.palette.m3surfaceContainer
+		implicitWidth: root.isUser ? Math.min(msgText.implicitWidth + Tokens.padding.medium * 2, root.contentMaxWidth) : root.contentMaxWidth
+		implicitHeight: root.isUser ? msgText.contentHeight + Tokens.padding.medium * 2 : blocks.implicitHeight + Tokens.padding.medium * 2
+		anchors.right: root.isUser ? parent.right : undefined
+
+		// User messages stay a plain editable text field.
+		TextEditBase {
+			id: msgText
+
+			visible: root.isUser
+			anchors.left: parent.left
+			anchors.margins: Tokens.padding.medium
+			anchors.right: parent.right
+			anchors.top: parent.top
+			animateCursor: false
+			color: root.isUser ? Colors.palette.m3onPrimary : Colors.palette.m3onSurface
+			cursor.color: root.isUser ? Colors.palette.m3onPrimary : Colors.palette.m3primary
+			font.pointSize: Tokens.font.size.smaller
+			readOnly: true
+			selectionColor: Qt.alpha((root.isUser ? Colors.palette.m3onPrimary : Colors.palette.m3primary), 0.4)
+			text: root.isUser ? root.segment.text : ""
+			textFormat: CustomText.MarkdownText
+			wrapMode: Text.WordWrap
+
+			Keys.onPressed: event => {
+				if (event.key == Qt.Key_Return) {
+					if (!(event.modifiers & Qt.ShiftModifier)) {
+						readOnly = true;
+						event.accepted = true;
+					}
+				} else if (event.key == Qt.Key_Escape) {
+					text = root.segment.text;
+					readOnly = true;
+					event.accepted = true;
+				}
+			}
+			onEditingFinished: {
+				const old = root.segment.text;
+				if (old !== text)
+					root.edit(text);
+
+				readOnly = true;
+			}
+			onReadOnlyChanged: {
+				if (readOnly) {
+					animateCursor = false;
+					root.forceActiveFocus();
+					textFormat = CustomText.MarkdownText;
+				} else {
+					var raw = root.segment.text;
+					textFormat = CustomText.PlainText;
+					text = raw;
+					forceActiveFocus();
+					cursorPosition = text.length;
+					animateCursor = true;
+				}
+			}
+		}
+
+		// Assistant content: parsed markdown blocks.
+		MarkdownBlocks {
+			id: blocks
+
+			visible: !root.isUser
+			anchors.top: parent.top
+			anchors.left: parent.left
+			anchors.right: parent.right
+			anchors.margins: Tokens.padding.medium
+			blocks: root.segment.markdown
+		}
+	}
+
+	Actions {
+		id: actionsRow
+
+		anchors.left: parent.left
+		anchors.right: (implicitWidth > bubble.implicitWidth) ? undefined : bubble.right
+		anchors.top: bubble.bottom
+		visible: {
+			return (root.segment.type === LlmSegment.Type.Content) && root.segment.status !== LlmSegment.Status.Running && root.repeater.count === (root.index + 1);
+		}
+		anchors.topMargin: Tokens.spacing.medium
+		edit: msgText
+		hovered: root.hovered
+		isUser: root.isUser
+		segment: root.segment
+		current: root.current
+		message: root.message
+	}
+}

Modules/Notifications/Sidebar/Chat/Content/MarkdownBlocks.qml

@@ -0,0 +1,88 @@
+import QtQuick
+import ZShell.Config
+import ZShell.Llm
+import qs.Components
+import qs.Services
+
+Column {
+	id: root
+
+	// Top-level markdown blocks (see MarkdownParser / LlmSegment.markdown).
+	required property var blocks
+
+	spacing: Tokens.spacing.small
+
+	Repeater {
+		id: blockRep
+
+		model: root.blocks
+
+		delegate: DelegateChooser {
+			role: "type"
+
+			DelegateChoice {
+				roleValue: LlmMarkdown.Type.Code
+
+				delegate: CodeBlockView {
+					required property var modelData
+
+					language: modelData.language
+					code: modelData.code
+					anchors.right: parent.right
+					anchors.left: parent.left
+				}
+			}
+
+			DelegateChoice {
+				roleValue: LlmMarkdown.Type.Math
+
+				delegate: MathBlockView {
+					required property var modelData
+
+					latex: modelData.latex
+					anchors.right: parent.right
+					anchors.left: parent.left
+				}
+			}
+
+			DelegateChoice {
+				roleValue: LlmMarkdown.Type.Heading
+
+				delegate: CustomText {
+					required property var modelData
+
+					color: Colors.palette.m3onSurface
+					anchors.right: parent.right
+					anchors.left: parent.left
+					text: modelData.text
+					textFormat: Text.MarkdownText
+					font.bold: true
+					font.pointSize: {
+						if (modelData.level <= 1)
+							return Tokens.font.size.larger;
+						if (modelData.level === 2)
+							return Tokens.font.size.normal;
+						return Tokens.font.size.smaller;
+					}
+					wrapMode: Text.WrapAtWordBoundaryOrAnywhere
+				}
+			}
+
+			DelegateChoice {
+				roleValue: LlmMarkdown.Type.Text
+
+				delegate: CustomText {
+					required property var modelData
+
+					color: Colors.palette.m3onSurface
+					anchors.right: parent.right
+					anchors.left: parent.left
+					text: modelData.text
+					textFormat: Text.MarkdownText
+					font.pointSize: Tokens.font.size.smaller
+					wrapMode: Text.WrapAtWordBoundaryOrAnywhere
+				}
+			}
+		}
+	}
+}

Modules/Notifications/Sidebar/Chat/Content/MathBlockView.qml

@@ -0,0 +1,50 @@
+import QtQuick
+import ZShell.Config
+import ZShell.Llm
+import qs.Components
+import qs.Services
+
+Item {
+	id: root
+
+	required property string latex
+	property color mathColor: Colors.palette.m3onSurface
+	property real fontSize: Tokens.font.size.large
+
+	implicitWidth: Math.max(fallbackText.implicitWidth, root.width)
+	implicitHeight: Math.max(equationImage.height, fallbackText.contentHeight) + Tokens.padding.small * 2
+
+	LlmMathText {
+		id: math
+
+		latex: root.latex
+		color: root.mathColor
+		fontPointSize: root.fontSize
+		devicePixelRatio: Screen.devicePixelRatio
+	}
+
+	Image {
+		id: equationImage
+
+		anchors.centerIn: parent
+		source: math.imageUrl
+		asynchronous: true
+		visible: math.ok
+		fillMode: Image.PreserveAspectFit
+		smooth: true
+		width: Math.min(math.width, root.width)
+		height: math.width > 0 ? width * (math.height / math.width) : 0
+	}
+
+	CustomText {
+		id: fallbackText
+
+		anchors.centerIn: parent
+		color: root.mathColor
+		font.family: Config.appearance.font.family.mono
+		font.pointSize: Tokens.font.size.small
+		text: root.latex
+		wrapMode: Text.WrapAtWordBoundaryOrAnywhere
+		visible: !math.ok
+	}
+}

Modules/Notifications/Sidebar/Chat/Content/MessageDelegate.qml

@@ -1,6 +1,8 @@
 pragma ComponentBehavior: Bound
 
 import QtQuick
+import QtQuick.Layouts
+import Quickshell
 import ZShell.Config
 import ZShell.Llm
 import qs.Components
@@ -14,6 +16,7 @@ MouseArea {
 	readonly property bool isUser: modelData.role === ChatMessage.Role.User
 	required property ChatMessage modelData
 	property bool reasoningExpanded: false
+	readonly property var blocks: blockify(current.segments)
 	property real savedContentY: -1
 
 	function handleReasoningToggle(expanded: bool): void {
@@ -34,14 +37,36 @@ MouseArea {
 		}
 	}
 
-	anchors.left: parent.left
-	anchors.leftMargin: root.isUser ? 0 : Tokens.spacing.extraSmall
-	anchors.right: parent.right
-	anchors.rightMargin: root.isUser ? Tokens.spacing.extraSmall : 0
+	function blockify(segs) {
+		const blocks = [];
+		for (let i = 0; i < segs.length; i++) {
+			const seg = segs[i];
+			if (seg.type === LlmSegment.Type.Content) {
+				blocks.push({
+					kind: "content",
+					id: i,
+					segments: [seg]
+				});
+			} else {
+				const last = blocks.length ? blocks[blocks.length - 1] : null;
+				if (last && last.kind === "process") {
+					last.segments.push(seg);
+				} else {
+					blocks.push({
+						kind: "process",
+						id: i,
+						segments: [seg]
+					});
+				}
+			}
+		}
+		return blocks;
+	}
+
+	implicitWidth: ListView.view.width
 	hoverEnabled: true
-	// Explicit, synchronous height — sum of children, no Layout polish involved.
-	implicitHeight: reasoningLoader.implicitHeight + (reasoningLoader.active ? Tokens.spacing.medium : 0) + bubble.implicitHeight + actionsRow.implicitHeight + actionsRow.anchors.topMargin
 	preventStealing: true
+	implicitHeight: layout.implicitHeight + Tokens.padding.medium
 
 	Binding {
 		property: "contentY"
@@ -59,106 +84,61 @@ MouseArea {
 		type: Anim.DefaultEffects
 	}
 
-	Loader {
-		id: reasoningLoader
+	Column {
+		id: layout
 
-		readonly property bool shouldBeActive: root.current.reasoningActive || root.current.content === ""
-
-		active: !root.isUser
+		anchors.top: parent.top
 		anchors.left: parent.left
 		anchors.right: parent.right
-		anchors.rightMargin: Tokens.spacing.extraLarge
-		anchors.top: parent.top
 
-		sourceComponent: Reasoning {
-			current: root.current
+		Repeater {
+			id: segmentRep
 
-			onExpandedChanged: {
-				root.handleReasoningToggle(expanded);
+			model: ScriptModel {
+				values: root.blocks
+				objectProp: "id"
 			}
-		}
-	}
+			delegate: DelegateChooser {
+				role: "kind"
 
-	CustomRect {
-		id: bubble
+				DelegateChoice {
+					roleValue: "process"
 
-		anchors.left: root.isUser ? undefined : parent.left
-		anchors.right: root.isUser ? parent.right : undefined
-		anchors.top: reasoningLoader.active ? reasoningLoader.bottom : parent.top
-		anchors.topMargin: reasoningLoader.active ? Tokens.spacing.medium : 0
-		color: root.isUser ? Colors.palette.m3primary : Colors.palette.m3surfaceContainer
-		implicitHeight: root.current.content !== "" ? msgText.contentHeight + Tokens.padding.medium * 2 : 0
-		implicitWidth: Math.min(msgText.implicitWidth + Tokens.padding.medium * 2, root.width - Tokens.spacing.extraSmall - Tokens.spacing.extraLarge)
-		radius: Tokens.rounding.medium
-		visible: !root.current.reasoningActive
+					delegate: ProcessBlock {
+						required property int index
+						required property var modelData
 
-		TextEditBase {
-			id: msgText
+						segments: modelData.segments
+						isActive: index === root.blocks.length - 1
+						width: root.width
 
-			anchors.left: parent.left
-			anchors.margins: Tokens.padding.medium
-			anchors.right: parent.right
-			anchors.top: parent.top
-			animateCursor: false
-			color: root.isUser ? Colors.palette.m3onPrimary : Colors.palette.m3onSurface
-			cursor.color: root.isUser ? Colors.palette.m3onPrimary : Colors.palette.m3primary
-			font.pointSize: Tokens.font.size.smaller
-			readOnly: true
-			selectionColor: Qt.alpha((root.isUser ? Colors.palette.m3onPrimary : Colors.palette.m3primary), 0.4)
-			text: root.current.content
-			textFormat: CustomText.MarkdownText
-			wrapMode: Text.WordWrap
-
-			Keys.onPressed: event => {
-				if (event.key == Qt.Key_Return) {
-					if (!(event.modifiers & Qt.ShiftModifier)) {
-						readOnly = true;
-						event.accepted = true;
+						onExpandedChanged: root.handleReasoningToggle(expanded)
 					}
-				} else if (event.key == Qt.Key_Escape) {
-					text = root.current.content;
-					readOnly = true;
-					event.accepted = true;
-				}
-			}
-			onEditingFinished: {
-				const old = root.current.content;
-				if (old !== text) {
-					root.modelData.edit(text);
-
-					if (root.isUser)
-						root.modelData.generate();
 				}
 
-				readOnly = true;
-			}
-			onReadOnlyChanged: {
-				if (readOnly) {
-					animateCursor = false;
-					root.forceActiveFocus();
-					textFormat = CustomText.MarkdownText;
-				} else {
-					var raw = root.current.content;
-					textFormat = CustomText.PlainText;
-					text = raw;
-					forceActiveFocus();
-					cursorPosition = text.length;
-					animateCursor = true;
+				DelegateChoice {
+					roleValue: "content"
+
+					delegate: ContentBubble {
+						required property var modelData
+
+						isUser: root.isUser
+						segment: modelData.segments[0]
+						width: root.width
+						repeater: segmentRep
+						message: root.modelData
+						current: root.current
+						hovered: root.containsMouse
+
+						onEdit: text => {
+							root.modelData.edit(text);
+
+							if (root.isUser)
+								root.modelData.generate();
+						}
+					}
 				}
 			}
 		}
 	}
-
-	Actions {
-		id: actionsRow
-
-		anchors.left: parent.left
-		anchors.right: (implicitWidth > bubble.implicitWidth) ? undefined : bubble.right
-		anchors.top: bubble.bottom
-		anchors.topMargin: Tokens.spacing.medium
-		current: root.current
-		edit: msgText
-		hovered: root.containsMouse
-		message: root.modelData
-	}
 }

Modules/Notifications/Sidebar/Chat/Content/ProcessBlock.qml

@@ -0,0 +1,149 @@
+import QtQuick
+import QtQuick.Layouts
+import ZShell.Config
+import ZShell.Llm
+import qs.Components
+import qs.Services
+
+Item {
+	id: root
+
+	required property var segments
+	required property bool isActive
+	property bool expanded: false
+	readonly property LlmSegment lastSegment: root.segments[root.segments.length - 1]
+	readonly property real totalElapsedMs: root.segments.reduce((sum, s) => sum + (s.elapsedMs ?? 0), 0)
+
+	implicitHeight: expandedRect.implicitHeight + collapsedText.implicitHeight + expandedRect.anchors.topMargin
+
+	LoadingIndicator {
+		id: spinnerReasoning
+
+		anchors.centerIn: expandBtn
+		implicitSize: collapsedText.implicitHeight
+		opacity: root.isActive ? 1 : 0
+		visible: opacity > 0
+
+		Behavior on opacity {
+			Anim {}
+		}
+	}
+
+	IconButton {
+		id: expandBtn
+
+		anchors.left: parent.left
+		anchors.verticalCenter: collapsedText.verticalCenter
+		font.pointSize: Tokens.font.size.large
+		anchors.topMargin: 0
+		icon: "keyboard_arrow_down"
+		inactiveOnColor: hovered ? Colors.palette.m3onSurface : Colors.palette.m3outline
+		opacity: root.isActive ? 0 : 1
+		rotation: root.expanded ? 180 : 0
+		type: IconButton.Text
+
+		Behavior on rotation {
+			Anim {}
+		}
+
+		onClicked: root.expanded = !root.expanded
+	}
+
+	CustomText {
+		id: collapsedText
+
+		anchors.left: expandBtn.right
+		anchors.margins: Tokens.padding.medium
+		anchors.right: parent.right
+		anchors.top: parent.top
+		anchors.topMargin: 0
+		color: Colors.palette.m3outline
+		font.pointSize: Tokens.font.size.small
+		text: {
+			if (root.isActive) {
+				if (root.lastSegment.type === LlmSegment.Type.ToolCall)
+					return qsTr("Using %1...").arg(root.lastSegment.name);
+				return qsTr("Thinking...");
+			}
+			return qsTr("Worked for %1s").arg((root.totalElapsedMs / 1000).toFixed(1));
+		}
+
+		Behavior on opacity {
+			Anim {}
+		}
+	}
+
+	CustomClippingRect {
+		id: expandedRect
+
+		anchors.left: parent.left
+		anchors.right: parent.right
+		anchors.top: collapsedText.bottom
+		anchors.topMargin: Tokens.spacing.medium
+		color: Colors.palette.m3surfaceContainerLow
+		implicitHeight: root.expanded ? expandedContent.contentHeight + expandedContent.anchors.margins * 2 : 0
+		opacity: root.expanded ? 1 : 0
+		radius: Tokens.rounding.medium
+		visible: opacity > 0
+
+		Behavior on implicitHeight {
+			Anim {
+				type: Anim.DefaultEffects
+			}
+		}
+		Behavior on opacity {
+			Anim {}
+		}
+
+		ListView {
+			id: expandedContent
+
+			anchors.fill: parent
+			anchors.margins: Tokens.padding.medium
+			model: root.segments
+			spacing: Tokens.spacing.medium
+
+			delegate: Item {
+				id: segment
+
+				required property LlmSegment modelData
+				required property int index
+
+				implicitWidth: ListView.view.width
+				implicitHeight: header.implicitHeight + content.implicitHeight + content.anchors.topMargin
+
+				RowLayout {
+					id: header
+
+					anchors.left: parent.left
+					anchors.right: parent.right
+					spacing: Tokens.spacing.small
+
+					MaterialIcon {
+						text: segment.modelData.type === LlmSegment.Type.Reasoning ? "cognition" : "build"
+						fill: 1
+					}
+
+					CustomText {
+						Layout.fillWidth: true
+						text: segment.modelData.type === LlmSegment.Type.Reasoning ? qsTr("Thought for %1s").arg((segment.modelData.elapsedMs / 1000).toFixed(1)) : qsTr("Used %1 for %2s").arg(segment.modelData.name).arg((segment.modelData.elapsedMs / 1000).toFixed(1))
+					}
+				}
+
+				CustomText {
+					id: content
+
+					anchors.left: parent.left
+					anchors.leftMargin: Tokens.padding.small
+					anchors.top: header.bottom
+					anchors.topMargin: Tokens.spacing.small
+					anchors.right: parent.right
+					color: Colors.palette.m3outline
+					font.pointSize: Tokens.font.size.small
+					wrapMode: CustomText.WrapAtWordBoundaryOrAnywhere
+					text: segment.modelData.type === LlmSegment.Type.Reasoning ? segment.modelData.text : qsTr("Fetched %1").arg(JSON.parse(segment.modelData.arguments)?.url ?? "website")
+				}
+			}
+		}
+	}
+}

Modules/Notifications/Sidebar/Chat/Content/Reasoning.qml

@@ -1,107 +0,0 @@
-import QtQuick
-import ZShell.Config
-import ZShell.Llm
-import qs.Components
-import qs.Services
-
-Item {
-	id: root
-
-	required property ChatGeneration current
-	property bool expanded: false
-	readonly property bool isReasoning: root.current.reasoningActive || root.current.content === ""
-
-	implicitHeight: expandedRect.implicitHeight + collapsedText.implicitHeight + expandedRect.anchors.topMargin
-
-	LoadingIndicator {
-		id: spinnerReasoning
-
-		anchors.left: parent.left
-		anchors.margins: Tokens.padding.medium
-		anchors.verticalCenter: parent.verticalCenter
-		implicitSize: collapsedText.implicitHeight
-		opacity: root.isReasoning ? 1 : 0
-		visible: opacity > 0
-
-		Behavior on opacity {
-			Anim {
-			}
-		}
-	}
-
-	IconButton {
-		id: expandBtn
-
-		anchors.left: parent.left
-		anchors.margins: Tokens.padding.medium
-		anchors.verticalCenter: collapsedText.verticalCenter
-		font.pointSize: Tokens.font.size.large
-		icon: "keyboard_arrow_down"
-		inactiveOnColor: hovered ? Colors.palette.m3onSurface : Colors.palette.m3outline
-		opacity: root.isReasoning ? 0 : 1
-		rotation: root.expanded ? 180 : 0
-		type: IconButton.Text
-
-		Behavior on rotation {
-			Anim {
-			}
-		}
-
-		onClicked: root.expanded = !root.expanded
-	}
-
-	CustomText {
-		id: collapsedText
-
-		anchors.left: root.current.reasoningActive ? spinnerReasoning.right : expandBtn.right
-		anchors.margins: Tokens.padding.medium
-		anchors.right: parent.right
-		anchors.top: parent.top
-		anchors.topMargin: 0
-		color: Colors.palette.m3outline
-		font.pointSize: Tokens.font.size.small
-		text: root.current.reasoningActive ? qsTr("Thinking...") : qsTr("Thought for %1s").arg((root.current.reasoningElapsedMs / 1000).toFixed(1))
-
-		Behavior on opacity {
-			Anim {
-			}
-		}
-	}
-
-	CustomClippingRect {
-		id: expandedRect
-
-		anchors.left: parent.left
-		anchors.right: parent.right
-		anchors.top: collapsedText.bottom
-		anchors.topMargin: Tokens.spacing.medium
-		color: Colors.palette.m3surfaceContainerLow
-		implicitHeight: root.expanded ? expandedText.implicitHeight + expandedText.anchors.margins * 2 : -anchors.topMargin
-		implicitWidth: expandedText.implicitWidth + expandedText.anchors.margins * 2
-		opacity: root.expanded ? 1 : 0
-		radius: Tokens.rounding.medium
-		visible: opacity > 0
-
-		Behavior on implicitHeight {
-			Anim {
-				type: Anim.DefaultEffects
-			}
-		}
-		Behavior on opacity {
-			Anim {
-			}
-		}
-
-		CustomText {
-			id: expandedText
-
-			anchors.fill: parent
-			anchors.margins: Tokens.padding.medium
-			color: Colors.palette.m3onSurface
-			font.pointSize: Tokens.font.size.small
-			text: root.current.reasoning
-			textFormat: CustomText.MarkdownText
-			wrapMode: Text.WordWrap
-		}
-	}
-}

Plugins/ZShell/Config/llm.hpp

@@ -11,6 +11,9 @@ class Llm : public ConfigObject {
 	CFG_PROPERTY(QString, endpoint, "http://localhost:8080")
 	CFG_PROPERTY(QString, model, "")
 	CFG_PROPERTY(double, temperature, 0.7)
+	// Whether tools are offered to the model. Models without a tool-calling
+	// template should run with this off.
+	CFG_PROPERTY(bool, tools, true)
 
 	public:
 	explicit Llm(QObject* parent = nullptr) : ConfigObject(parent) {}

Plugins/ZShell/Llm/CMakeLists.txt

@@ -1,17 +1,76 @@
+# cmark-gfm: markdown -> block structure (pkg-config; no CMake config
+# package is installed).
+pkg_check_modules(CMARK_GFM REQUIRED IMPORTED_TARGET libcmark-gfm)
+
+# tree-sitter runtime for code block highlighting (grammars are
+# dlopen()'d at runtime and optional).
+pkg_check_modules(TREE_SITTER REQUIRED IMPORTED_TARGET tree-sitter)
+
+# JKQTMathText (JKQtPlotter) for LaTeX rendering. The config files live
+# in a shared JKQTPlotter6 directory, not one named after the package.
+find_path(JKQTPlotter6_CMAKE_DIR
+	NAMES JKQTMathText6Config.cmake
+	HINTS /usr/lib/cmake/JKQTPlotter6 /usr/local/lib/cmake/JKQTPlotter6
+	DOC "Directory containing the JKQtPlotter cmake package files")
+find_package(JKQTMathText6 REQUIRED PATHS "${JKQTPlotter6_CMAKE_DIR}")
+
+# Embed the vendored highlight queries as C++ string literals.
+set(HIGHLIGHT_QUERY_LANGS c cpp python javascript typescript bash json rust go yaml toml sql)
+set(HIGHLIGHT_QUERIES_HPP "${CMAKE_CURRENT_BINARY_DIR}/highlight-queries.hpp")
+set(_hl_header "#pragma once\n\n// Vendored tree-sitter highlight queries (MIT; see the per-file\n// source headers in the highlight-queries/ directory).\nnamespace ZShell::llm::hq {\n")
+foreach(_hl_lang IN LISTS HIGHLIGHT_QUERY_LANGS)
+	file(READ "${CMAKE_CURRENT_SOURCE_DIR}/highlight-queries/${_hl_lang}.scm" _hl_src)
+	string(APPEND _hl_header "inline constexpr const char* ${_hl_lang} = R\"ZSQUERY(${_hl_src})ZSQUERY\";\n")
+endforeach()
+string(APPEND _hl_header "}\n")
+file(WRITE "${HIGHLIGHT_QUERIES_HPP}" "${_hl_header}")
+
+# Embed the vendored Latin Modern fonts (GUST Font License; provenance
+# in fonts/latinmodern/GUST-FONT-LICENSE.txt) as byte arrays.
+set(LM_FONTS
+	lmroman10-regular
+	lmroman10-italic
+	lmroman10-bold
+	lmroman10-bolditalic
+	latinmodern-math)
+set(LM_FONTS_HPP "${CMAKE_CURRENT_BINARY_DIR}/latinmodern-fonts.hpp")
+set(_lm_header "#pragma once\n\n// Vendored Latin Modern fonts (GUST Font License; see\n// fonts/latinmodern/GUST-FONT-LICENSE.txt).\nnamespace ZShell::llm::lmfont {\n")
+foreach(_lm_font IN LISTS LM_FONTS)
+	string(REPLACE "-" "_" _lm_sym "${_lm_font}")
+	file(READ "${CMAKE_CURRENT_SOURCE_DIR}/fonts/latinmodern/${_lm_font}.otf" _lm_hex HEX)
+	string(REGEX REPLACE "([0-9a-f][0-9a-f])" "0x\\1," _lm_bytes "${_lm_hex}")
+	string(APPEND _lm_header "inline const unsigned char ${_lm_sym}[] = { ${_lm_bytes} };\n")
+endforeach()
+string(APPEND _lm_header "}\n")
+file(WRITE "${LM_FONTS_HPP}" "${_lm_header}")
+
 qml_module(ZShell-llm
 	URI ZShell.Llm
 	SOURCES
 		chat.hpp chat.cpp
 		chatstore.hpp chatstore.cpp
+		codehighlighter.hpp codehighlighter.cpp
 		generation.hpp generation.cpp
 		llmclient.hpp llmclient.cpp
+		markdownblock.hpp
+		markdownparser.hpp markdownparser.cpp
+		mathtext.hpp mathtext.cpp
 		message.hpp message.cpp
 		messagemodel.hpp messagemodel.cpp
+		segment.hpp segment.cpp
 		session.hpp session.cpp
+		tool.hpp tool.cpp
+		webfetchtool.hpp webfetchtool.cpp
 	LIBRARIES
 		Qt::Network
 		Qt::Sql
+		Qt::Gui
+		Qt::Widgets
 		ZShell-config
+		JKQTPlotter::JKQTMathText
+		PkgConfig::CMARK_GFM
+		PkgConfig::TREE_SITTER
 )
 
+target_include_directories(ZShell-llm PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
 target_include_directories(ZShell-llm PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../Config)

Plugins/ZShell/Llm/chat.cpp

@@ -3,6 +3,7 @@
 #include "config.hpp"
 #include "llm.hpp"
 #include "llmclient.hpp"
+#include "webfetchtool.hpp"
 
 #include <QDebug>
 
@@ -14,11 +15,13 @@ Chat::Chat(QObject* parent)
 		new config::Config();
 
 	m_store->setLlmClient(m_client);
+	m_client->tools()->registerTool(new WebFetchTool(m_client->tools()));
 
 	const auto* llm = config::Config::instance()->llm();
 	m_client->setEndpoint(llm->endpoint());
 	m_client->setModel(llm->model());
 	m_client->setTemperature(llm->temperature());
+	m_client->setToolsEnabled(llm->tools());
 
 	connect(llm, &config::Llm::endpointChanged, this, [this, llm]() {
 		m_client->setEndpoint(llm->endpoint());
@@ -29,6 +32,9 @@ Chat::Chat(QObject* parent)
 	connect(llm, &config::Llm::temperatureChanged, this, [this, llm]() {
 		m_client->setTemperature(llm->temperature());
 	});
+	connect(llm, &config::Llm::toolsChanged, this, [this, llm]() {
+		m_client->setToolsEnabled(llm->tools());
+	});
 
 	connect(m_client, &LlmClient::busyChanged, this, [this]() {
 		// A fresh run supersedes the previous error.
@@ -57,6 +63,11 @@ Chat::Chat(QObject* parent)
 		&LlmClient::streamingChatIdChanged,
 		this,
 		&Chat::streamingChatIdChanged);
+	connect(
+		m_client,
+		&LlmClient::toolsEnabledChanged,
+		this,
+		&Chat::toolsEnabledChanged);
 	connect(
 		m_client,
 		&LlmClient::errorOccurred,
@@ -113,6 +124,16 @@ int Chat::contextSize() const {
 	return m_client->contextSize();
 }
 
+bool Chat::toolsEnabled() const {
+	return m_client->toolsEnabled();
+}
+
+void Chat::setToolsEnabled(bool value) {
+	m_client->setToolsEnabled(value);
+	if (auto* config = config::Config::instance())
+		config->llm()->set_tools(value);
+}
+
 QString Chat::streamingChatId() const {
 	return m_client->streamingChatId();
 }

Plugins/ZShell/Llm/chat.hpp

@@ -27,6 +27,7 @@ class Chat : public QObject {
 	Q_PROPERTY(QString model READ model NOTIFY modelChanged)
 	Q_PROPERTY(QStringList availableModels READ availableModels NOTIFY availableModelsChanged)
 	Q_PROPERTY(int contextSize READ contextSize NOTIFY contextSizeChanged)
+	Q_PROPERTY(bool toolsEnabled READ toolsEnabled WRITE setToolsEnabled NOTIFY toolsEnabledChanged)
 	Q_PROPERTY(QString lastError READ lastError NOTIFY lastErrorChanged)
 	Q_PROPERTY(ZShell::llm::ChatStore* chats READ chats CONSTANT)
 	Q_PROPERTY(QString streamingChatId READ streamingChatId NOTIFY streamingChatIdChanged)
@@ -39,6 +40,8 @@ class Chat : public QObject {
 	[[nodiscard]] QString model() const;
 	[[nodiscard]] QStringList availableModels() const;
 	[[nodiscard]] int contextSize() const;
+	[[nodiscard]] bool toolsEnabled() const;
+	void setToolsEnabled(bool value);
 	[[nodiscard]] QString lastError() const { return m_lastError; }
 	[[nodiscard]] ChatStore* chats() const { return m_store; }
 	[[nodiscard]] QString streamingChatId() const;
@@ -56,6 +59,7 @@ class Chat : public QObject {
 	void modelChanged();
 	void availableModelsChanged();
 	void contextSizeChanged();
+	void toolsEnabledChanged();
 	void errorOccurred(const QString& message);
 	void lastErrorChanged();
 	void streamingChatIdChanged();

Plugins/ZShell/Llm/chatstore.cpp

@@ -2,6 +2,7 @@
 
 #include "llmclient.hpp"
 #include "message.hpp"
+#include "segment.hpp"
 
 #include <QDateTime>
 #include <QDebug>
@@ -17,6 +18,38 @@
 
 namespace ZShell::llm {
 
+namespace {
+
+QString segmentTypeName(LlmSegment::Type type) {
+	switch (type) {
+	case LlmSegment::Type::Reasoning:
+		return QStringLiteral("reasoning");
+	case LlmSegment::Type::ToolCall:
+		return QStringLiteral("tool_call");
+	case LlmSegment::Type::Content:
+		return QStringLiteral("content");
+	}
+	return QStringLiteral("reasoning");
+}
+
+LlmSegment::Type segmentTypeFromName(const QString& name) {
+	if (name == QLatin1String("tool_call"))
+		return LlmSegment::Type::ToolCall;
+	if (name == QLatin1String("content"))
+		return LlmSegment::Type::Content;
+	return LlmSegment::Type::Reasoning;
+}
+
+// A null QString binds as SQL NULL, which violates the NOT NULL columns;
+// DEFAULT only applies to omitted columns, not explicit NULLs.
+QString sqlText(const QString& value) {
+	if (value.isNull())
+		return QStringLiteral("");
+	return value;
+}
+
+} // namespace
+
 ChatStore::ChatStore(QObject* parent)
 : QObject(parent), m_connectionName(QUuid::createUuid().toString()) {
 	openDb();
@@ -86,14 +119,29 @@ void ChatStore::openDb() {
 				"  id INTEGER PRIMARY KEY AUTOINCREMENT,\n"
 				"  message_id INTEGER NOT NULL REFERENCES messages "
 				"(id) ON DELETE CASCADE,\n"
-				"  content TEXT,\n"
-				"  reasoning TEXT,\n"
 				"  timestamp INTEGER NOT NULL,\n"
-				"  reasoning_elapsed_ms INTEGER NOT NULL DEFAULT 0,\n"
-				"  content_elapsed_ms INTEGER NOT NULL DEFAULT 0,\n"
 				"  is_active INTEGER NOT NULL DEFAULT 1\n"
 				")"));
 	}
+	{
+		QSqlQuery query(db);
+		query.exec(
+			QStringLiteral(
+				"CREATE TABLE IF NOT EXISTS segments (\n"
+				"  id INTEGER PRIMARY KEY AUTOINCREMENT,\n"
+				"  generation_id INTEGER NOT NULL REFERENCES generations "
+				"(id) ON DELETE CASCADE,\n"
+				"  type TEXT NOT NULL,\n"
+				"  text TEXT NOT NULL DEFAULT '',\n"
+				"  name TEXT NOT NULL DEFAULT '',\n"
+				"  tool_call_id TEXT NOT NULL DEFAULT '',\n"
+				"  arguments TEXT NOT NULL DEFAULT '',\n"
+				"  result TEXT NOT NULL DEFAULT '',\n"
+				"  status INTEGER NOT NULL DEFAULT 0,\n"
+				"  elapsed_ms INTEGER NOT NULL DEFAULT 0,\n"
+				"  timestamp INTEGER NOT NULL\n"
+				")"));
+	}
 	{
 		QSqlQuery query(db);
 		query.exec(QStringLiteral(
@@ -102,6 +150,9 @@ void ChatStore::openDb() {
 		query.exec(QStringLiteral(
 			"CREATE INDEX IF NOT EXISTS idx_generations_message "
 			"ON generations (message_id)"));
+		query.exec(QStringLiteral(
+			"CREATE INDEX IF NOT EXISTS idx_segments_generation "
+			"ON segments (generation_id)"));
 	}
 }
 
@@ -212,8 +263,8 @@ void ChatStore::saveMeta(ChatSession* session) {
 	QSqlQuery query(db());
 	query.prepare("UPDATE sessions SET title = :title, icon = :icon "
 	              "WHERE id = :id");
-	query.bindValue(":title", session->title());
-	query.bindValue(":icon", session->icon());
+	query.bindValue(":title", sqlText(session->title()));
+	query.bindValue(":icon", sqlText(session->icon()));
 	query.bindValue(":id", session->id());
 	if (!query.exec())
 		qWarning() << "ChatStore: failed to save meta for" << session->id()
@@ -234,7 +285,7 @@ bool ChatStore::saveSession(ChatSession* session) {
 		query.prepare(
 			"UPDATE sessions SET title = :title, updated_at = :updated_at "
 			"WHERE id = :id");
-		query.bindValue(":title", session->title());
+		query.bindValue(":title", sqlText(session->title()));
 		query.bindValue(":updated_at", session->updatedAtMs());
 		query.bindValue(":id", session->id());
 		ok = query.exec();
@@ -252,10 +303,15 @@ bool ChatStore::saveSession(ChatSession* session) {
 			"VALUES (:id, :role, :timestamp)");
 		QSqlQuery generationInsert(handle);
 		ok = ok && generationInsert.prepare(
-			"INSERT INTO generations (message_id, content, reasoning, "
-			"timestamp, reasoning_elapsed_ms, content_elapsed_ms, "
-			"is_active) VALUES (:mid, :content, :reasoning, :timestamp, "
-			":reasoning_elapsed_ms, :content_elapsed_ms, :is_active)");
+			"INSERT INTO generations (message_id, timestamp, is_active) "
+			"VALUES (:mid, :timestamp, :is_active)");
+		QSqlQuery segmentInsert(handle);
+		ok = ok && segmentInsert.prepare(
+			"INSERT INTO segments (generation_id, type, text, name, "
+			"tool_call_id, arguments, result, status, elapsed_ms, "
+			"timestamp) VALUES (:gid, :type, :text, :name, "
+			":tool_call_id, :arguments, :result, :status, :elapsed_ms, "
+			":timestamp)");
 		// The model holds messages most recent first; the database keeps
 		// natural rowid order, so iterate from the oldest row up.
 		const auto* model = session->messagesModel();
@@ -279,14 +335,8 @@ bool ChatStore::saveSession(ChatSession* session) {
 			for (int i = 0; ok && i < message->generationCount(); ++i) {
 				const auto* generation = message->generation(i);
 				generationInsert.bindValue(":mid", messageId);
-				generationInsert.bindValue(":content", generation->content());
 				generationInsert.bindValue(
-					":reasoning", generation->reasoning());
-				generationInsert.bindValue(":timestamp", generation->timestamp());
-				generationInsert.bindValue(
-					":reasoning_elapsed_ms", generation->reasoningElapsedMs());
-				generationInsert.bindValue(
-					":content_elapsed_ms", generation->contentElapsedMs());
+					":timestamp", generation->timestamp());
 				generationInsert.bindValue(
 					":is_active",
 					i == message->activeGenerationIndex() ? 1 : 0);
@@ -297,6 +347,33 @@ bool ChatStore::saveSession(ChatSession* session) {
 					           << generationInsert.lastError().text();
 					break;
 				}
+				const int generationId =
+					generationInsert.lastInsertId().toInt();
+				for (const auto* segment : generation->segments()) {
+					segmentInsert.bindValue(":gid", generationId);
+					segmentInsert.bindValue(
+						":type", segmentTypeName(segment->type()));
+					segmentInsert.bindValue(":text", sqlText(segment->text()));
+					segmentInsert.bindValue(":name", sqlText(segment->name()));
+					segmentInsert.bindValue(
+						":tool_call_id", sqlText(segment->toolCallId()));
+					segmentInsert.bindValue(
+						":arguments", sqlText(segment->arguments()));
+					segmentInsert.bindValue(":result", sqlText(segment->result()));
+					segmentInsert.bindValue(
+						":status", static_cast<int>(segment->status()));
+					segmentInsert.bindValue(
+						":elapsed_ms", segment->elapsedMs());
+					segmentInsert.bindValue(
+						":timestamp", segment->timestamp());
+					if (!segmentInsert.exec()) {
+						ok = false;
+						qWarning() << "ChatStore: saveSession" << id
+						           << "segment insert failed:"
+						           << segmentInsert.lastError().text();
+						break;
+					}
+				}
 			}
 		}
 	}
@@ -333,21 +410,54 @@ void ChatStore::loadMessagesInto(ChatSession* session) {
 			query.value(2).toLongLong());
 		QSqlQuery generationQuery(db());
 		generationQuery.prepare(
-			"SELECT content, reasoning, timestamp, reasoning_elapsed_ms, "
-			"content_elapsed_ms, is_active FROM generations "
+			"SELECT id, timestamp, is_active FROM generations "
 			"WHERE message_id = :mid ORDER BY rowid");
 		generationQuery.bindValue(":mid", messageId);
 		int activeIndex = 0;
 		if (generationQuery.exec()) {
 			int index = 0;
 			while (generationQuery.next()) {
-				message->addGeneration(
-					generationQuery.value(2).toLongLong(),
-					generationQuery.value(0).toString(),
-					generationQuery.value(1).toString(),
-					generationQuery.value(3).toLongLong(),
-					generationQuery.value(4).toLongLong());
-				if (generationQuery.value(5).toInt() != 0)
+				auto* generation = message->addGeneration(
+					generationQuery.value(1).toLongLong());
+				QSqlQuery segmentQuery(db());
+				segmentQuery.prepare(
+					"SELECT type, text, name, tool_call_id, arguments, "
+					"result, status, elapsed_ms, timestamp FROM segments "
+					"WHERE generation_id = :gid ORDER BY rowid");
+				segmentQuery.bindValue(
+					":gid", generationQuery.value(0).toInt());
+				if (segmentQuery.exec()) {
+					while (segmentQuery.next()) {
+						auto* segment = new LlmSegment(
+							segmentTypeFromName(
+								segmentQuery.value(0).toString()),
+							segmentQuery.value(8).toLongLong(),
+							generation);
+						segment->setText(
+							segmentQuery.value(1).toString());
+						segment->setName(
+							segmentQuery.value(2).toString());
+						segment->setToolCallId(
+							segmentQuery.value(3).toString());
+						segment->appendArguments(
+							segmentQuery.value(4).toString());
+						segment->setResult(
+							segmentQuery.value(5).toString());
+						segment->setStatus(
+							static_cast<LlmSegment::Status>(
+								segmentQuery.value(6).toInt()));
+						segment->restore(
+							segmentQuery.value(7).toLongLong());
+						generation->addSegment(segment);
+					}
+				} else {
+					qWarning() << "ChatStore: failed to load segments for "
+					           << "generation"
+					           << generationQuery.value(0).toInt()
+					           << ":"
+					           << segmentQuery.lastError().text();
+				}
+				if (generationQuery.value(2).toInt() != 0)
 					activeIndex = index;
 				++index;
 			}

Plugins/ZShell/Llm/codehighlighter.cpp

@@ -0,0 +1,384 @@
+#include "codehighlighter.hpp"
+
+#include "highlight-queries.hpp"
+
+#include <tree_sitter/api.h>
+
+#include <QHash>
+#include <QMap>
+#include <QVariantMap>
+
+#include <dlfcn.h>
+
+namespace ZShell::llm {
+
+namespace hl {
+
+// Role ids; 0 means "no color".
+enum Role : uint8_t {
+	None = 0,
+	Comment,
+	String,
+	StringKey,
+	Number,
+	Constant,
+	Keyword,
+	Type,
+	Function,
+	Method,
+	Macro,
+	Preproc,
+	Operator,
+	Property,
+	Label,
+	Attribute,
+};
+
+const char* roleName(Role role) {
+	switch (role) {
+	case Comment: return "comment";
+	case String: return "string";
+	case StringKey: return "string.key";
+	case Number: return "number";
+	case Constant: return "constant";
+	case Keyword: return "keyword";
+	case Type: return "type";
+	case Function: return "function";
+	case Method: return "method";
+	case Macro: return "macro";
+	case Preproc: return "preproc";
+	case Operator: return "operator";
+	case Property: return "property";
+	case Label: return "label";
+	case Attribute: return "attribute";
+	default: return "";
+	}
+}
+
+using LanguageFn = const TSLanguage* (*)();
+
+// Query sources, indexed by Grammar::queries.
+struct QuerySources {
+	std::vector<std::string> sources;
+	int c, cpp, python, javascript, typescriptExtended, typescript, bash, json, rust, go, yaml, toml, sql, cppExtended;
+};
+
+const QuerySources& querySources() {
+	static const QuerySources sources = [] {
+		QuerySources qs;
+		// javascript + typescript concatenated: the TS grammar reuses the
+		// JS node names, so the JS query usually compiles against it and
+		// gives full coverage; the TS-only query is the fallback.
+		const std::string tsExtended =
+			std::string(hq::javascript) + "\n" + std::string(hq::typescript);
+		// Same for cpp: the C++ grammar is a superset of C, and the C++
+		// query only covers the C++-specific delta. Base C coverage
+		// (keywords, types, calls, strings, comments) comes from the C
+		// query.
+		const std::string cppExtended =
+			std::string(hq::c) + "\n" + std::string(hq::cpp);
+		qs.sources.reserve(14);
+		qs.sources.push_back(hq::c);
+		qs.sources.push_back(hq::cpp);
+		qs.sources.push_back(hq::python);
+		qs.sources.push_back(hq::javascript);
+		qs.sources.push_back(tsExtended);
+		qs.sources.push_back(hq::typescript);
+		qs.sources.push_back(hq::bash);
+		qs.sources.push_back(hq::json);
+		qs.sources.push_back(hq::rust);
+		qs.sources.push_back(hq::go);
+		qs.sources.push_back(hq::yaml);
+		qs.sources.push_back(hq::toml);
+		qs.sources.push_back(hq::sql);
+		qs.sources.push_back(cppExtended);
+		qs.c = 0;
+		qs.cpp = 1;
+		qs.python = 2;
+		qs.javascript = 3;
+		qs.typescriptExtended = 4;
+		qs.typescript = 5;
+		qs.bash = 6;
+		qs.json = 7;
+		qs.rust = 8;
+		qs.go = 9;
+		qs.yaml = 10;
+		qs.toml = 11;
+		qs.sql = 12;
+		qs.cppExtended = 13;
+		return qs;
+	}();
+	return sources;
+}
+
+const QHash<QString, CodeHighlighter::Grammar>& grammars() {
+	static const QHash<QString, CodeHighlighter::Grammar> grammars = [] {
+		const auto& qs = querySources();
+		QHash<QString, CodeHighlighter::Grammar> map;
+		map.insert("c", {"libtree-sitter-c.so", "tree_sitter_c", {qs.c}});
+		map.insert("cpp", {"libtree-sitter-cpp.so", "tree_sitter_cpp", {qs.cppExtended, qs.cpp}});
+		map.insert("python", {"libtree-sitter-python.so", "tree_sitter_python", {qs.python}});
+		map.insert("javascript", {"libtree-sitter-javascript.so", "tree_sitter_javascript", {qs.javascript}});
+		map.insert("typescript", {"libtree-sitter-typescript.so", "tree_sitter_typescript", {qs.typescriptExtended, qs.typescript}});
+		map.insert("tsx", {"libtree-sitter-tsx.so", "tree_sitter_tsx", {qs.typescriptExtended, qs.typescript}});
+		map.insert("bash", {"libtree-sitter-bash.so", "tree_sitter_bash", {qs.bash}});
+		map.insert("json", {"libtree-sitter-json.so", "tree_sitter_json", {qs.json}});
+		map.insert("rust", {"libtree-sitter-rust.so", "tree_sitter_rust", {qs.rust}});
+		map.insert("go", {"libtree-sitter-go.so", "tree_sitter_go", {qs.go}});
+		map.insert("yaml", {"libtree-sitter-yaml.so", "tree_sitter_yaml", {qs.yaml}});
+		map.insert("toml", {"libtree-sitter-toml.so", "tree_sitter_toml", {qs.toml}});
+		map.insert("sql", {"libtree-sitter-sql.so", "tree_sitter_sql", {qs.sql}});
+		return map;
+	}();
+	return grammars;
+}
+
+} // namespace hl
+
+CodeHighlighter* CodeHighlighter::s_instance = nullptr;
+
+const QHash<QString, QString>& CodeHighlighter::aliases() {
+	// Language tags as written in code fences (and common variants) to
+	// grammar id.
+	static const QHash<QString, QString> aliases = [] {
+		QHash<QString, QString> map;
+		map.insert("c", "c");
+		map.insert("h", "c");
+		map.insert("cpp", "cpp");
+		map.insert("c++", "cpp");
+		map.insert("cc", "cpp");
+		map.insert("cxx", "cpp");
+		map.insert("h++", "cpp");
+		map.insert("hpp", "cpp");
+		map.insert("hh", "cpp");
+		map.insert("python", "python");
+		map.insert("py", "python");
+		map.insert("javascript", "javascript");
+		map.insert("js", "javascript");
+		map.insert("jsx", "javascript");
+		map.insert("mjs", "javascript");
+		map.insert("cjs", "javascript");
+		map.insert("typescript", "typescript");
+		map.insert("ts", "typescript");
+		map.insert("mts", "typescript");
+		map.insert("cts", "typescript");
+		map.insert("tsx", "tsx");
+		map.insert("bash", "bash");
+		map.insert("sh", "bash");
+		map.insert("shell", "bash");
+		map.insert("shellscript", "bash");
+		map.insert("shell-session", "bash");
+		map.insert("zsh", "bash");
+		map.insert("console", "bash");
+		map.insert("json", "json");
+		map.insert("jsonc", "json");
+		map.insert("rust", "rust");
+		map.insert("rs", "rust");
+		map.insert("go", "go");
+		map.insert("golang", "go");
+		map.insert("yaml", "yaml");
+		map.insert("yml", "yaml");
+		map.insert("toml", "toml");
+		map.insert("sql", "sql");
+		map.insert("mysql", "sql");
+		map.insert("postgres", "sql");
+		map.insert("postgresql", "sql");
+		map.insert("sqlite", "sql");
+		map.insert("sqlite3", "sql");
+		return map;
+	}();
+	return aliases;
+}
+
+const std::vector<std::string>& CodeHighlighter::querySources() const {
+	return hl::querySources().sources;
+}
+
+uint8_t CodeHighlighter::roleFor(const char* name, uint32_t length) {
+	const QString n = QString::fromUtf8(name, length);
+	if (n == "comment")
+		return hl::Role::Comment;
+	if (n.startsWith("string"))
+		return n == "string.special.key" ? hl::Role::StringKey : hl::Role::String;
+	if (n == "escape" || n == "regexp")
+		return hl::Role::String;
+	if (n.startsWith("number"))
+		return hl::Role::Number;
+	if (n.startsWith("constant"))
+		return hl::Role::Constant;
+	if (n.startsWith("keyword"))
+		return hl::Role::Keyword;
+	if (n == "type" || n.startsWith("type."))
+		return hl::Role::Type;
+	if (n == "namespace" || n == "module")
+		return hl::Role::Type;
+	if (n.startsWith("function") || n == "constructor")
+		return hl::Role::Function;
+	if (n == "method" || n == "method.builtin")
+		return hl::Role::Method;
+	if (n.startsWith("macro"))
+		return hl::Role::Macro;
+	if (n.startsWith("preproc"))
+		return hl::Role::Preproc;
+	if (n == "operator" || n == "punctuation.operator" || n.startsWith("operator."))
+		return hl::Role::Operator;
+	if (n == "property" || n == "field" || n.startsWith("property."))
+		return hl::Role::Property;
+	if (n == "label")
+		return hl::Role::Label;
+	if (n.startsWith("attribute") || n == "annotation")
+		return hl::Role::Attribute;
+	return hl::Role::None;
+}
+
+const char* CodeHighlighter::roleName(uint8_t role) {
+	return hl::roleName(static_cast<hl::Role>(role));
+}
+
+QVariantList CodeHighlighter::highlight(const QString& code, const QString& language) const {
+	QVariantList spans;
+	if (code.isEmpty())
+		return spans;
+
+	const QString id = aliases().value(language.trimmed().toLower());
+	if (id.isEmpty())
+		return spans;
+	const Grammar& grammar = hl::grammars().value(id);
+
+	// Guard against pathological blocks; highlighting is best-effort.
+	static constexpr size_t kMaxBytes = 512 * 1024;
+	const QByteArray utf8 = code.toUtf8();
+	if (static_cast<size_t>(utf8.size()) > kMaxBytes)
+		return spans;
+
+	State& state = m_states[id];
+	if (state.bad)
+		return spans;
+
+	if (!state.lang) {
+		// Missing library is retriable (it may be installed while the
+		// shell runs); ABI/query failures below are not.
+		state.lib = dlopen(grammar.lib.toUtf8().constData(), RTLD_NOW | RTLD_LOCAL);
+		if (!state.lib)
+			return spans;
+		auto* symbol = reinterpret_cast<hl::LanguageFn>(
+			dlsym(state.lib, grammar.symbol.toUtf8().constData()));
+		if (!symbol) {
+			dlclose(state.lib);
+			state.lib = nullptr;
+			return spans;
+		}
+		const TSLanguage* lang = symbol();
+		const uint32_t version = ts_language_abi_version(lang);
+		if (version < TREE_SITTER_MIN_COMPATIBLE_LANGUAGE_VERSION ||
+		    version > TREE_SITTER_LANGUAGE_VERSION) {
+			state.bad = true;
+			return spans;
+		}
+		state.lang = lang;
+	}
+
+	if (!state.query) {
+		// Candidates in priority order; first that compiles wins.
+		for (const int candidate : grammar.queries) {
+			const std::string& source =
+				querySources()[static_cast<size_t>(candidate)];
+			TSQueryError errorType = TSQueryErrorNone;
+			uint32_t errorOffset = 0;
+			TSQuery* query = ts_query_new(
+				static_cast<const TSLanguage*>(state.lang),
+				source.data(),
+				static_cast<uint32_t>(source.size()),
+				&errorOffset,
+				&errorType);
+			if (!query)
+				continue;
+			state.query = query;
+			break;
+		}
+		if (!state.query) {
+			state.bad = true;
+			return spans;
+		}
+	}
+
+	TSParser* parser = ts_parser_new();
+	ts_parser_set_language(parser, static_cast<const TSLanguage*>(state.lang));
+	TSTree* tree = ts_parser_parse_string(
+		parser, nullptr, utf8.constData(), static_cast<uint32_t>(utf8.size()));
+	if (!tree) {
+		ts_parser_delete(parser);
+		return spans;
+	}
+
+	TSQuery* query = static_cast<TSQuery*>(state.query);
+	TSQueryCursor* cursor = ts_query_cursor_new();
+	ts_query_cursor_exec(cursor, query, ts_tree_root_node(tree));
+
+	// Per-byte winner table: captures arrive in document order and later
+	// captures overwrite earlier ones (tree-sitter highlight convention).
+	const uint32_t size = static_cast<uint32_t>(utf8.size());
+	std::vector<uint8_t> kinds(size, 0);
+
+	// QML slices the code by UTF-16 code unit, so spans must be in code
+	// units, not bytes. cu[b] = code units before byte b.
+	std::vector<uint32_t> cu(size + 1, 0);
+	for (uint32_t b = 0; b < size; ++b) {
+		cu[b + 1] = cu[b];
+		const unsigned char c = static_cast<unsigned char>(utf8[b]);
+		if (c < 0x80)
+			cu[b + 1] += 1;
+		else if (c < 0xC0)
+			; // continuation byte
+		else if (c < 0xF0)
+			cu[b + 1] += 1; // 2/3-byte lead -> BMP -> one unit
+		else
+			cu[b + 1] += 2; // 4-byte lead -> surrogate pair
+	}
+
+	TSQueryMatch match;
+	uint32_t captureIndex = 0;
+	while (ts_query_cursor_next_capture(cursor, &match, &captureIndex)) {
+		const TSQueryCapture& capture = match.captures[captureIndex];
+		uint32_t nameLength = 0;
+		const char* name = ts_query_capture_name_for_id(query, capture.index, &nameLength);
+		const uint8_t role = roleFor(name, nameLength);
+		if (role == 0)
+			continue;
+		const uint32_t start = ts_node_start_byte(capture.node);
+		const uint32_t end = ts_node_end_byte(capture.node);
+		if (end <= start || end > size)
+			continue;
+		std::fill(kinds.begin() + start, kinds.begin() + end, role);
+	}
+
+	uint32_t position = 0;
+	while (position < size) {
+		if (kinds[position] == 0) {
+			++position;
+			continue;
+		}
+		const uint8_t role = kinds[position];
+		const uint32_t start = position;
+		while (position < size && kinds[position] == role)
+			++position;
+		QVariantMap span;
+		span.insert("start", static_cast<int>(cu[start]));
+		span.insert("length", static_cast<int>(cu[position] - cu[start]));
+		span.insert("kind", roleName(role));
+		spans.append(span);
+	}
+
+	ts_query_cursor_delete(cursor);
+	ts_tree_delete(tree);
+	ts_parser_delete(parser);
+	return spans;
+}
+
+CodeHighlighter* CodeHighlighter::create(QQmlEngine*, QJSEngine*) {
+	if (!s_instance)
+		s_instance = new CodeHighlighter();
+	return s_instance;
+}
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/codehighlighter.hpp

@@ -0,0 +1,69 @@
+#pragma once
+
+#include <QObject>
+#include <QString>
+#include <QVariantList>
+#include <QtQml>
+
+#include <cstdint>
+#include <vector>
+
+class QQmlEngine;
+class QJSEngine;
+
+namespace ZShell::llm {
+
+// Syntax highlighting for LLM code blocks via tree-sitter.
+//
+// The tree-sitter runtime is linked. Grammar libraries
+// (libtree-sitter-<lang>.so) are dlopen()'d lazily, so a missing
+// grammar package degrades that language to plain text instead of
+// breaking the build or the app. Highlight queries are embedded at
+// build time (highlight-queries/*.scm, vendored from the grammar
+// repos, MIT).
+//
+// The fence language the LLM wrote (```cpp, ```python, ...) is mapped
+// to a grammar through an alias table.
+//
+// highlight() returns a list of span maps:
+//   { "start": int, "length": int, "kind": QString }
+// where kind is a semantic role (keyword, string, comment, number,
+// function, type, ...) that QML maps to theme colors. An empty list
+// means "no highlighting" (unknown language or grammar not installed).
+class CodeHighlighter : public QObject {
+	Q_OBJECT
+	QML_ELEMENT
+	QML_SINGLETON
+
+	public:
+	Q_INVOKABLE QVariantList highlight(const QString& code, const QString& language) const;
+
+	static CodeHighlighter* create(QQmlEngine*, QJSEngine*);
+
+	struct Grammar {
+		QString lib; // soname to dlopen
+		QString symbol; // tree_sitter_<lang>() entry point
+		// Candidate query sources, first wins. Lets typescript reuse
+		// the javascript query when it compiles against the grammar.
+		std::vector<int> queries; // indices into querySources()
+	};
+
+	private:
+	struct State {
+		bool bad = false; // permanent failure, do not retry
+		void* lib = nullptr;
+		const void* lang = nullptr; // const TSLanguage*
+		void* query = nullptr; // TSQuery*
+	};
+
+	[[nodiscard]] static const QHash<QString, QString>& aliases();
+	[[nodiscard]] const std::vector<std::string>& querySources() const;
+	// Maps a tree-sitter capture name to a role index (0 = unstyled).
+	[[nodiscard]] static uint8_t roleFor(const char* name, uint32_t length);
+	[[nodiscard]] static const char* roleName(uint8_t role);
+
+	mutable QHash<QString, State> m_states;
+	static CodeHighlighter* s_instance;
+};
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/fonts/latinmodern/GUST-FONT-LICENSE.txt

@@ -0,0 +1,40 @@
+% This is a preliminary version (2006-09-30), barring acceptance from
+% the LaTeX Project Team and other feedback, of the GUST Font License.
+% (GUST is the Polish TeX Users Group, http://www.gust.org.pl)
+%
+% For the most recent version of this license see
+% http://www.gust.org.pl/fonts/licenses/GUST-FONT-LICENSE.txt
+% or
+% http://tug.org/fonts/licenses/GUST-FONT-LICENSE.txt
+%
+% This work may be distributed and/or modified under the conditions
+% of the LaTeX Project Public License, either version 1.3c of this
+% license or (at your option) any later version.
+% 
+% Please also observe the following clause:
+% 1) it is requested, but not legally required, that derived works be
+%    distributed only after changing the names of the fonts comprising this
+%    work and given in an accompanying "manifest", and that the
+%    files comprising the Work, as listed in the manifest, also be given
+%    new names. Any exceptions to this request are also given in the
+%    manifest.
+%    
+%    We recommend the manifest be given in a separate file named
+%    MANIFEST-<fontid>.txt, where <fontid> is some unique identification
+%    of the font family. If a separate "readme" file accompanies the Work, 
+%    we recommend a name of the form README-<fontid>.txt.
+%
+% The latest version of the LaTeX Project Public License is in
+% http://www.latex-project.org/lppl.txt and version 1.3c or later
+% is part of all distributions of LaTeX version 2006/05/20 or later.
+
+
+---
+
+Provenance:
+  lmroman10-*.otf: Latin Modern v2.007 (GUST, 31-03-2026)
+    https://www.gust.org.pl/projects/e-foundry/latin-modern/download
+    (Latin_Modern-otf-2_007-31_03_2026.zip)
+  latinmodern-math.otf: Latin Modern Math v1.959 (GUST)
+    https://www.gust.org.pl/projects/e-foundry/lm-math/download
+    (latinmodern-math-1959.zip; same release as CTAN fonts/lm-math)

Plugins/ZShell/Llm/generation.cpp

@@ -10,34 +10,92 @@ ChatGeneration::ChatGeneration(qint64 timestamp, QObject* parent)
 	m_timer.setInterval(500);
 	m_timer.setTimerType(Qt::CoarseTimer);
 	connect(&m_timer, &QTimer::timeout, this, [this]() {
-		if (!reasoningInFlight() && !contentInFlight()) {
-			m_timer.stop();
-			return;
+		bool anyRunning = false;
+		for (auto* segment : m_segments) {
+			if (!segment->running())
+				continue;
+			anyRunning = true;
+			segment->elapsedMsChanged();
 		}
-		Q_EMIT elapsedMsChanged();
+		if (anyRunning)
+			Q_EMIT elapsedMsChanged();
+		if (!anyRunning && !m_streaming)
+			m_timer.stop();
 	});
 }
 
+QString ChatGeneration::content() const {
+	QStringList parts;
+	for (const auto* segment : m_segments) {
+		if (segment->type() != LlmSegment::Type::Content ||
+		    segment->text().isEmpty())
+			continue;
+		parts.append(segment->text());
+	}
+	return parts.join(QStringLiteral("\n\n"));
+}
+
+QString ChatGeneration::reasoning() const {
+	QStringList parts;
+	for (const auto* segment : m_segments) {
+		if (segment->type() != LlmSegment::Type::Reasoning ||
+		    segment->text().isEmpty())
+			continue;
+		parts.append(segment->text());
+	}
+	return parts.join(QStringLiteral("\n\n"));
+}
+
+bool ChatGeneration::reasoningActive() const {
+	if (!m_streaming)
+		return false;
+	if (!content().isEmpty())
+		return false;
+	return !hasRunningTool();
+}
+
 qint64 ChatGeneration::reasoningElapsedMs() const {
-	if (m_reasoningStartedAt <= 0)
-		return 0;
-	const qint64 end = m_reasoningEndedAt > 0
-		? m_reasoningEndedAt
-		: QDateTime::currentMSecsSinceEpoch();
-	return end - m_reasoningStartedAt;
+	qint64 total = 0;
+	for (const auto* segment : m_segments)
+		if (segment->type() == LlmSegment::Type::Reasoning)
+			total += segment->elapsedMs();
+	return total;
 }
 
 qint64 ChatGeneration::contentElapsedMs() const {
-	if (m_contentStartedAt <= 0)
-		return 0;
-	const qint64 end = m_contentEndedAt > 0
-		? m_contentEndedAt
-		: QDateTime::currentMSecsSinceEpoch();
-	return end - m_contentStartedAt;
+	qint64 total = 0;
+	for (const auto* segment : m_segments)
+		if (segment->type() == LlmSegment::Type::Content)
+			total += segment->elapsedMs();
+	return total;
+}
+
+qint64 ChatGeneration::toolsElapsedMs() const {
+	qint64 total = 0;
+	for (const auto* segment : m_segments)
+		if (segment->type() == LlmSegment::Type::ToolCall)
+			total += segment->elapsedMs();
+	return total;
+}
+
+int ChatGeneration::toolCallCount() const {
+	int count = 0;
+	for (const auto* segment : m_segments)
+		if (segment->type() == LlmSegment::Type::ToolCall)
+			++count;
+	return count;
+}
+
+bool ChatGeneration::hasRunningTool() const {
+	for (const auto* segment : m_segments)
+		if (segment->type() == LlmSegment::Type::ToolCall &&
+		    segment->running())
+			return true;
+	return false;
 }
 
 void ChatGeneration::updateReasoningActive() {
-	const bool active = m_streaming && m_content.isEmpty();
+	const bool active = reasoningActive();
 	if (m_reasoningActive == active)
 		return;
 	m_reasoningActive = active;
@@ -45,60 +103,51 @@ void ChatGeneration::updateReasoningActive() {
 }
 
 void ChatGeneration::setContent(const QString& value) {
-	if (m_content == value)
-		return;
-	m_content = value;
-	Q_EMIT contentChanged();
-}
-
-void ChatGeneration::setReasoning(const QString& value) {
-	if (m_reasoning == value)
-		return;
-	m_reasoning = value;
-	Q_EMIT reasoningChanged();
+	// Replaces the entire answer: the first content segment takes the
+	// new text, any later content bursts are cleared.
+	LlmSegment* first = nullptr;
+	for (auto* segment : m_segments) {
+		if (segment->type() != LlmSegment::Type::Content)
+			continue;
+		if (!first)
+			first = segment;
+		else
+			segment->setText(QString());
+	}
+	if (!first) {
+		// Do not materialize an empty content segment (e.g. the assistant
+		// placeholder created before the stream starts).
+		if (value.isEmpty())
+			return;
+		first = new LlmSegment(
+			LlmSegment::Type::Content,
+			QDateTime::currentMSecsSinceEpoch(),
+			this);
+		addSegment(first);
+	}
+	first->setText(value);
 }
 
 void ChatGeneration::appendContent(const QString& piece) {
 	if (piece.isEmpty())
 		return;
-	if (m_content.isEmpty()) {
-		const qint64 now = QDateTime::currentMSecsSinceEpoch();
-		if (reasoningInFlight())
-			m_reasoningEndedAt = now;
-		m_contentStartedAt = now;
-		if (!m_timer.isActive())
-			m_timer.start();
+	for (auto* segment : m_segments) {
+		if (segment->type() == LlmSegment::Type::Reasoning &&
+		    segment->running())
+			segment->close();
 	}
-	m_content += piece;
-	Q_EMIT contentChanged();
-	updateReasoningActive();
-	Q_EMIT elapsedMsChanged();
+	openContentSegment()->appendText(piece);
 }
 
 void ChatGeneration::appendReasoning(const QString& piece) {
 	if (piece.isEmpty())
 		return;
-	if (m_reasoning.isEmpty()) {
-		if (m_reasoningStartedAt <= 0)
-			m_reasoningStartedAt = QDateTime::currentMSecsSinceEpoch();
-		if (!m_timer.isActive())
-			m_timer.start();
-	}
-	m_reasoning += piece;
-	Q_EMIT reasoningChanged();
-	updateReasoningActive();
-	Q_EMIT elapsedMsChanged();
-}
-
-void ChatGeneration::setElapsedMs(qint64 reasoningMs, qint64 contentMs) {
-	if (reasoningMs > 0) {
-		m_reasoningStartedAt = m_timestamp;
-		m_reasoningEndedAt = m_timestamp + reasoningMs;
-	}
-	if (contentMs > 0) {
-		m_contentStartedAt = m_timestamp;
-		m_contentEndedAt = m_timestamp + contentMs;
+	for (auto* segment : m_segments) {
+		if (segment->type() == LlmSegment::Type::Content &&
+		    segment->running())
+			segment->close();
 	}
+	openReasoningSegment()->appendText(piece);
 }
 
 void ChatGeneration::setStreaming(bool value) {
@@ -107,19 +156,94 @@ void ChatGeneration::setStreaming(bool value) {
 	m_streaming = value;
 	Q_EMIT streamingChanged();
 	if (value) {
-		if (m_reasoningStartedAt <= 0)
-			m_reasoningStartedAt = QDateTime::currentMSecsSinceEpoch();
 		if (!m_timer.isActive())
 			m_timer.start();
 	} else {
-		const qint64 now = QDateTime::currentMSecsSinceEpoch();
-		if (reasoningInFlight())
-			m_reasoningEndedAt = now;
-		if (contentInFlight())
-			m_contentEndedAt = now;
+		closeOpenSegments();
 		m_timer.stop();
-		Q_EMIT elapsedMsChanged();
 	}
+	Q_EMIT elapsedMsChanged();
+	updateReasoningActive();
+}
+
+LlmSegment* ChatGeneration::openContentSegment() {
+	for (auto* segment : m_segments)
+		if (segment->type() == LlmSegment::Type::Content &&
+		    segment->running())
+			return segment;
+	auto* segment = new LlmSegment(
+		LlmSegment::Type::Content,
+		QDateTime::currentMSecsSinceEpoch(),
+		this);
+	segment->begin();
+	addSegment(segment);
+	return segment;
+}
+
+LlmSegment* ChatGeneration::openReasoningSegment() {
+	for (auto* segment : m_segments)
+		if (segment->type() == LlmSegment::Type::Reasoning &&
+		    segment->running())
+			return segment;
+	auto* segment = new LlmSegment(
+		LlmSegment::Type::Reasoning,
+		QDateTime::currentMSecsSinceEpoch(),
+		this);
+	segment->begin();
+	addSegment(segment);
+	return segment;
+}
+
+LlmSegment* ChatGeneration::beginToolCall(
+	const QString& name, const QString& toolCallId) {
+	closeOpenSegments();
+	auto* segment = new LlmSegment(
+		LlmSegment::Type::ToolCall,
+		QDateTime::currentMSecsSinceEpoch(),
+		this);
+	segment->setName(name);
+	segment->setToolCallId(toolCallId);
+	segment->setStatus(LlmSegment::Status::Running);
+	segment->begin();
+	addSegment(segment);
+	Q_EMIT toolStateChanged();
+	return segment;
+}
+
+void ChatGeneration::addSegment(LlmSegment* segment) {
+	if (!segment || m_segments.contains(segment))
+		return;
+	segment->setParent(this);
+	connect(
+		segment, &LlmSegment::textChanged, this, [this, segment]() {
+			if (segment->type() == LlmSegment::Type::Reasoning)
+				Q_EMIT reasoningChanged();
+			else if (segment->type() == LlmSegment::Type::Content)
+				Q_EMIT contentChanged();
+			updateReasoningActive();
+		});
+	connect(
+		segment, &LlmSegment::statusChanged, this, [this]() {
+			Q_EMIT toolStateChanged();
+		});
+	connect(
+		segment, &LlmSegment::resultChanged, this, [this]() {
+			Q_EMIT toolStateChanged();
+		});
+	connect(
+		segment, &LlmSegment::runningChanged, this, [this]() {
+			Q_EMIT elapsedMsChanged();
+			Q_EMIT toolStateChanged();
+			updateReasoningActive();
+		});
+	m_segments.append(segment);
+	Q_EMIT segmentsChanged();
+}
+
+void ChatGeneration::closeOpenSegments() {
+	for (auto* segment : m_segments)
+		if (segment->running())
+			segment->close();
 	updateReasoningActive();
 }
 

Plugins/ZShell/Llm/generation.hpp

@@ -1,5 +1,8 @@
 #pragma once
 
+#include "segment.hpp"
+
+#include <QList>
 #include <QObject>
 #include <QString>
 #include <QTimer>
@@ -7,6 +10,11 @@
 
 namespace ZShell::llm {
 
+// One attempt at answering a message: a chronologically ordered list
+// of segments. Content bursts, reasoning bursts and tool calls all
+// appear in the order the model produced them; a new content (or
+// reasoning) segment starts whenever the model switches between them.
+// Together with the attempt's aggregate state.
 class ChatGeneration : public QObject {
 	Q_OBJECT
 	QML_ELEMENT
@@ -16,55 +24,71 @@ class ChatGeneration : public QObject {
 	Q_PROPERTY(QString reasoning READ reasoning NOTIFY reasoningChanged)
 	Q_PROPERTY(qint64 reasoningElapsedMs READ reasoningElapsedMs NOTIFY elapsedMsChanged)
 	Q_PROPERTY(qint64 contentElapsedMs READ contentElapsedMs NOTIFY elapsedMsChanged)
+	Q_PROPERTY(qint64 toolsElapsedMs READ toolsElapsedMs NOTIFY elapsedMsChanged)
 	Q_PROPERTY(bool streaming READ streaming NOTIFY streamingChanged)
 	Q_PROPERTY(bool reasoningActive READ reasoningActive NOTIFY reasoningActiveChanged)
+	Q_PROPERTY(bool hasRunningTool READ hasRunningTool NOTIFY toolStateChanged)
 	Q_PROPERTY(qint64 timestamp READ timestamp CONSTANT)
+	Q_PROPERTY(
+		QList<ZShell::llm::LlmSegment*> segments READ segments
+			NOTIFY segmentsChanged)
+	Q_PROPERTY(int toolCallCount READ toolCallCount NOTIFY segmentsChanged)
 
 	public:
 	explicit ChatGeneration(qint64 timestamp, QObject* parent = nullptr);
 
-	[[nodiscard]] QString content() const { return m_content; }
-	[[nodiscard]] QString reasoning() const { return m_reasoning; }
-	[[nodiscard]] bool reasoningActive() const { return m_reasoningActive; }
+	// All content bursts, joined (the model's full answer).
+	[[nodiscard]] QString content() const;
+	// Every reasoning burst, joined.
+	[[nodiscard]] QString reasoning() const;
+	// True while the model is thinking: streaming, no content yet, and
+	// no tool call in flight.
+	[[nodiscard]] bool reasoningActive() const;
 	[[nodiscard]] qint64 reasoningElapsedMs() const;
 	[[nodiscard]] qint64 contentElapsedMs() const;
+	[[nodiscard]] qint64 toolsElapsedMs() const;
 	[[nodiscard]] bool streaming() const { return m_streaming; }
 	[[nodiscard]] qint64 timestamp() const { return m_timestamp; }
+	[[nodiscard]] QList<LlmSegment*> segments() const { return m_segments; }
+	[[nodiscard]] int toolCallCount() const;
+	[[nodiscard]] bool hasRunningTool() const;
 
 	void setContent(const QString& value);
-	void setReasoning(const QString& value);
 	void appendContent(const QString& piece);
 	void appendReasoning(const QString& piece);
-	void setElapsedMs(qint64 reasoningMs, qint64 contentMs);
 	void setStreaming(bool value);
 
+	// The in-flight content segment, or a fresh one. A new content
+	// segment starts whenever the model resumes writing after reasoning
+	// or a tool call.
+	[[nodiscard]] LlmSegment* openContentSegment();
+	// The in-flight reasoning segment, or a fresh one.
+	[[nodiscard]] LlmSegment* openReasoningSegment();
+	// Creates and appends a running tool-call segment.
+	[[nodiscard]] LlmSegment* beginToolCall(
+		const QString& name, const QString& toolCallId);
+	// Appends a segment created by the persistence layer.
+	void addSegment(LlmSegment* segment);
+	// Stops the clocks of every in-flight segment.
+	void closeOpenSegments();
+
 	Q_SIGNALS:
 	void contentChanged();
 	void reasoningChanged();
 	void reasoningActiveChanged();
 	void elapsedMsChanged();
 	void streamingChanged();
+	void toolStateChanged();
+	void segmentsChanged();
 
 	private:
 	void updateReasoningActive();
 
-	[[nodiscard]] bool reasoningInFlight() const {
-		return m_reasoningStartedAt > 0 && m_reasoningEndedAt <= 0;
-	}
-	[[nodiscard]] bool contentInFlight() const {
-		return m_contentStartedAt > 0 && m_contentEndedAt <= 0;
-	}
-
 	QTimer m_timer;
-	QString m_content;
-	QString m_reasoning;
+	QList<LlmSegment*> m_segments;
 	bool m_reasoningActive = false;
 	bool m_streaming = false;
 	qint64 m_timestamp;
-	qint64 m_reasoningStartedAt = 0;
-	qint64 m_reasoningEndedAt = 0;
-	qint64 m_contentStartedAt = 0;
-	qint64 m_contentEndedAt = 0;
 };
 
 } // namespace ZShell::llm

Plugins/ZShell/Llm/highlight-queries/bash.scm

@@ -0,0 +1,59 @@
+; Vendored from bash (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-bash
+
+[
+  (string)
+  (raw_string)
+  (heredoc_body)
+  (heredoc_start)
+] @string
+
+(command_name) @function
+
+(variable_name) @property
+
+[
+  "case"
+  "do"
+  "done"
+  "elif"
+  "else"
+  "esac"
+  "export"
+  "fi"
+  "for"
+  "function"
+  "if"
+  "in"
+  "select"
+  "then"
+  "unset"
+  "until"
+  "while"
+] @keyword
+
+(comment) @comment
+
+(function_definition name: (word) @function)
+
+(file_descriptor) @number
+
+[
+  (command_substitution)
+  (process_substitution)
+  (expansion)
+]@embedded
+
+[
+  "$"
+  "&&"
+  ">"
+  ">>"
+  "<"
+  "|"
+] @operator
+
+(
+  (command (_) @constant)
+  (#match? @constant "^-")
+)

Plugins/ZShell/Llm/highlight-queries/c.scm

@@ -0,0 +1,84 @@
+; Vendored from c (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-c
+
+(identifier) @variable
+
+((identifier) @constant
+ (#match? @constant "^[A-Z][A-Z\\d_]*$"))
+
+"break" @keyword
+"case" @keyword
+"const" @keyword
+"continue" @keyword
+"default" @keyword
+"do" @keyword
+"else" @keyword
+"enum" @keyword
+"extern" @keyword
+"for" @keyword
+"if" @keyword
+"inline" @keyword
+"return" @keyword
+"sizeof" @keyword
+"static" @keyword
+"struct" @keyword
+"switch" @keyword
+"typedef" @keyword
+"union" @keyword
+"volatile" @keyword
+"while" @keyword
+
+"#define" @keyword
+"#elif" @keyword
+"#else" @keyword
+"#endif" @keyword
+"#if" @keyword
+"#ifdef" @keyword
+"#ifndef" @keyword
+"#include" @keyword
+(preproc_directive) @keyword
+
+"--" @operator
+"-" @operator
+"-=" @operator
+"->" @operator
+"=" @operator
+"!=" @operator
+"*" @operator
+"&" @operator
+"&&" @operator
+"+" @operator
+"++" @operator
+"+=" @operator
+"<" @operator
+"==" @operator
+">" @operator
+"||" @operator
+
+"." @delimiter
+";" @delimiter
+
+(string_literal) @string
+(system_lib_string) @string
+
+(null) @constant
+(number_literal) @number
+(char_literal) @number
+
+(field_identifier) @property
+(statement_identifier) @label
+(type_identifier) @type
+(primitive_type) @type
+(sized_type_specifier) @type
+
+(call_expression
+  function: (identifier) @function)
+(call_expression
+  function: (field_expression
+    field: (field_identifier) @function))
+(function_declarator
+  declarator: (identifier) @function)
+(preproc_function_def
+  name: (identifier) @function.special)
+
+(comment) @comment

Plugins/ZShell/Llm/highlight-queries/cpp.scm

@@ -0,0 +1,73 @@
+; Vendored from cpp (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-cpp (tag v0.23.4, matching the distro grammar)
+
+; Functions
+
+(call_expression
+  function: (qualified_identifier
+    name: (identifier) @function))
+
+(template_function
+  name: (identifier) @function)
+
+(template_method
+  name: (field_identifier) @function)
+
+(template_function
+  name: (identifier) @function)
+
+(function_declarator
+  declarator: (qualified_identifier
+    name: (identifier) @function))
+
+(function_declarator
+  declarator: (field_identifier) @function)
+
+; Types
+
+((namespace_identifier) @type
+ (#match? @type "^[A-Z]"))
+
+(auto) @type
+
+; Constants
+
+(this) @variable.builtin
+(null "nullptr" @constant)
+
+; Keywords
+
+[
+ "catch"
+ "class"
+ "co_await"
+ "co_return"
+ "co_yield"
+ "constexpr"
+ "constinit"
+ "consteval"
+ "delete"
+ "explicit"
+ "final"
+ "friend"
+ "mutable"
+ "namespace"
+ "noexcept"
+ "new"
+ "override"
+ "private"
+ "protected"
+ "public"
+ "template"
+ "throw"
+ "try"
+ "typename"
+ "using"
+ "concept"
+ "requires"
+ "virtual"
+] @keyword
+
+; Strings
+
+(raw_string_literal) @string

Plugins/ZShell/Llm/highlight-queries/go.scm

@@ -0,0 +1,126 @@
+; Vendored from go (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-go
+
+; Function calls
+
+(call_expression
+  function: (identifier) @function)
+
+(call_expression
+  function: (identifier) @function.builtin
+  (#match? @function.builtin "^(append|cap|close|complex|copy|delete|imag|len|make|new|panic|print|println|real|recover)$"))
+
+(call_expression
+  function: (selector_expression
+    field: (field_identifier) @function.method))
+
+; Function definitions
+
+(function_declaration
+  name: (identifier) @function)
+
+(method_declaration
+  name: (field_identifier) @function.method)
+
+; Identifiers
+
+(type_identifier) @type
+(field_identifier) @property
+(identifier) @variable
+
+; Operators
+
+[
+  "--"
+  "-"
+  "-="
+  ":="
+  "!"
+  "!="
+  "..."
+  "*"
+  "*"
+  "*="
+  "/"
+  "/="
+  "&"
+  "&&"
+  "&="
+  "%"
+  "%="
+  "^"
+  "^="
+  "+"
+  "++"
+  "+="
+  "<-"
+  "<"
+  "<<"
+  "<<="
+  "<="
+  "="
+  "=="
+  ">"
+  ">="
+  ">>"
+  ">>="
+  "|"
+  "|="
+  "||"
+  "~"
+] @operator
+
+; Keywords
+
+[
+  "break"
+  "case"
+  "chan"
+  "const"
+  "continue"
+  "default"
+  "defer"
+  "else"
+  "fallthrough"
+  "for"
+  "func"
+  "go"
+  "goto"
+  "if"
+  "import"
+  "interface"
+  "map"
+  "package"
+  "range"
+  "return"
+  "select"
+  "struct"
+  "switch"
+  "type"
+  "var"
+] @keyword
+
+; Literals
+
+[
+  (interpreted_string_literal)
+  (raw_string_literal)
+  (rune_literal)
+] @string
+
+(escape_sequence) @escape
+
+[
+  (int_literal)
+  (float_literal)
+  (imaginary_literal)
+] @number
+
+[
+  (true)
+  (false)
+  (nil)
+  (iota)
+] @constant.builtin
+
+(comment) @comment

Plugins/ZShell/Llm/highlight-queries/javascript.scm

@@ -0,0 +1,207 @@
+; Vendored from javascript (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-javascript
+
+; Variables
+;----------
+
+(identifier) @variable
+
+; Properties
+;-----------
+
+(property_identifier) @property
+
+; Function and method definitions
+;--------------------------------
+
+(function_expression
+  name: (identifier) @function)
+(function_declaration
+  name: (identifier) @function)
+(method_definition
+  name: (property_identifier) @function.method)
+
+(pair
+  key: (property_identifier) @function.method
+  value: [(function_expression) (arrow_function)])
+
+(assignment_expression
+  left: (member_expression
+    property: (property_identifier) @function.method)
+  right: [(function_expression) (arrow_function)])
+
+(variable_declarator
+  name: (identifier) @function
+  value: [(function_expression) (arrow_function)])
+
+(assignment_expression
+  left: (identifier) @function
+  right: [(function_expression) (arrow_function)])
+
+; Function and method calls
+;--------------------------
+
+(call_expression
+  function: (identifier) @function)
+
+(call_expression
+  function: (member_expression
+    property: (property_identifier) @function.method))
+
+; Special identifiers
+;--------------------
+
+((identifier) @constructor
+ (#match? @constructor "^[A-Z]"))
+
+([
+    (identifier)
+    (shorthand_property_identifier)
+    (shorthand_property_identifier_pattern)
+ ] @constant
+ (#match? @constant "^[A-Z_][A-Z\\d_]+$"))
+
+((identifier) @variable.builtin
+ (#match? @variable.builtin "^(arguments|module|console|window|document)$")
+ (#is-not? local))
+
+((identifier) @function.builtin
+ (#eq? @function.builtin "require")
+ (#is-not? local))
+
+; Literals
+;---------
+
+(this) @variable.builtin
+(super) @variable.builtin
+
+[
+  (true)
+  (false)
+  (null)
+  (undefined)
+] @constant.builtin
+
+(comment) @comment
+
+[
+  (string)
+  (template_string)
+] @string
+
+(regex) @string.special
+(number) @number
+
+; Tokens
+;-------
+
+[
+  ";"
+  (optional_chain)
+  "."
+  ","
+] @punctuation.delimiter
+
+[
+  "-"
+  "--"
+  "-="
+  "+"
+  "++"
+  "+="
+  "*"
+  "*="
+  "**"
+  "**="
+  "/"
+  "/="
+  "%"
+  "%="
+  "<"
+  "<="
+  "<<"
+  "<<="
+  "="
+  "=="
+  "==="
+  "!"
+  "!="
+  "!=="
+  "=>"
+  ">"
+  ">="
+  ">>"
+  ">>="
+  ">>>"
+  ">>>="
+  "~"
+  "^"
+  "&"
+  "|"
+  "^="
+  "&="
+  "|="
+  "&&"
+  "||"
+  "??"
+  "&&="
+  "||="
+  "??="
+] @operator
+
+[
+  "("
+  ")"
+  "["
+  "]"
+  "{"
+  "}"
+]  @punctuation.bracket
+
+(template_substitution
+  "${" @punctuation.special
+  "}" @punctuation.special) @embedded
+
+[
+  "as"
+  "async"
+  "await"
+  "break"
+  "case"
+  "catch"
+  "class"
+  "const"
+  "continue"
+  "debugger"
+  "default"
+  "delete"
+  "do"
+  "else"
+  "export"
+  "extends"
+  "finally"
+  "for"
+  "from"
+  "function"
+  "get"
+  "if"
+  "import"
+  "in"
+  "instanceof"
+  "let"
+  "new"
+  "of"
+  "return"
+  "set"
+  "static"
+  "switch"
+  "target"
+  "throw"
+  "try"
+  "typeof"
+  "var"
+  "void"
+  "while"
+  "with"
+  "yield"
+] @keyword

Plugins/ZShell/Llm/highlight-queries/json.scm

@@ -0,0 +1,19 @@
+; Vendored from json (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-json
+
+(pair
+  key: (_) @string.special.key)
+
+(string) @string
+
+(number) @number
+
+[
+  (null)
+  (true)
+  (false)
+] @constant.builtin
+
+(escape_sequence) @escape
+
+(comment) @comment

Plugins/ZShell/Llm/highlight-queries/python.scm

@@ -0,0 +1,140 @@
+; Vendored from python (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-python
+
+; Identifier naming conventions
+
+(identifier) @variable
+
+((identifier) @constructor
+ (#match? @constructor "^[A-Z]"))
+
+((identifier) @constant
+ (#match? @constant "^[A-Z][A-Z_]*$"))
+
+; Function calls
+
+(decorator) @function
+(decorator
+  (identifier) @function)
+
+(call
+  function: (attribute attribute: (identifier) @function.method))
+(call
+  function: (identifier) @function)
+
+; Builtin functions
+
+((call
+  function: (identifier) @function.builtin)
+ (#match?
+   @function.builtin
+   "^(abs|all|any|ascii|bin|bool|breakpoint|bytearray|bytes|callable|chr|classmethod|compile|complex|delattr|dict|dir|divmod|enumerate|eval|exec|filter|float|format|frozenset|getattr|globals|hasattr|hash|help|hex|id|input|int|isinstance|issubclass|iter|len|list|locals|map|max|memoryview|min|next|object|oct|open|ord|pow|print|property|range|repr|reversed|round|set|setattr|slice|sorted|staticmethod|str|sum|super|tuple|type|vars|zip|__import__)$"))
+
+; Function definitions
+
+(function_definition
+  name: (identifier) @function)
+
+(attribute attribute: (identifier) @property)
+(type (identifier) @type)
+
+; Literals
+
+[
+  (none)
+  (true)
+  (false)
+] @constant.builtin
+
+[
+  (integer)
+  (float)
+] @number
+
+(comment) @comment
+(string) @string
+(escape_sequence) @escape
+
+(interpolation
+  "{" @punctuation.special
+  "}" @punctuation.special) @embedded
+
+[
+  "-"
+  "-="
+  "!="
+  "*"
+  "**"
+  "**="
+  "*="
+  "/"
+  "//"
+  "//="
+  "/="
+  "&"
+  "&="
+  "%"
+  "%="
+  "^"
+  "^="
+  "+"
+  "->"
+  "+="
+  "<"
+  "<<"
+  "<<="
+  "<="
+  "<>"
+  "="
+  ":="
+  "=="
+  ">"
+  ">="
+  ">>"
+  ">>="
+  "|"
+  "|="
+  "~"
+  "@="
+  "and"
+  "in"
+  "is"
+  "not"
+  "or"
+  "is not"
+  "not in"
+] @operator
+
+[
+  "as"
+  "assert"
+  "async"
+  "await"
+  "break"
+  "class"
+  "continue"
+  "def"
+  "del"
+  "elif"
+  "else"
+  "except"
+  "exec"
+  "finally"
+  "for"
+  "from"
+  "global"
+  "if"
+  "import"
+  "lambda"
+  "nonlocal"
+  "pass"
+  "print"
+  "raise"
+  "return"
+  "try"
+  "while"
+  "with"
+  "yield"
+  "match"
+  "case"
+] @keyword

Plugins/ZShell/Llm/highlight-queries/rust.scm

@@ -0,0 +1,164 @@
+; Vendored from rust (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-rust
+
+; Identifiers
+
+(type_identifier) @type
+(primitive_type) @type.builtin
+(field_identifier) @property
+
+; Identifier conventions
+
+; Assume all-caps names are constants
+((identifier) @constant
+ (#match? @constant "^[A-Z][A-Z\\d_]+$'"))
+
+; Assume uppercase names are enum constructors
+((identifier) @constructor
+ (#match? @constructor "^[A-Z]"))
+
+; Assume that uppercase names in paths are types
+((scoped_identifier
+  path: (identifier) @type)
+ (#match? @type "^[A-Z]"))
+((scoped_identifier
+  path: (scoped_identifier
+    name: (identifier) @type))
+ (#match? @type "^[A-Z]"))
+((scoped_type_identifier
+  path: (identifier) @type)
+ (#match? @type "^[A-Z]"))
+((scoped_type_identifier
+  path: (scoped_identifier
+    name: (identifier) @type))
+ (#match? @type "^[A-Z]"))
+
+; Assume all qualified names in struct patterns are enum constructors. (They're
+; either that, or struct names; highlighting both as constructors seems to be
+; the less glaring choice of error, visually.)
+(struct_pattern
+  type: (scoped_type_identifier
+    name: (type_identifier) @constructor))
+
+; Function calls
+
+(call_expression
+  function: (identifier) @function)
+(call_expression
+  function: (field_expression
+    field: (field_identifier) @function.method))
+(call_expression
+  function: (scoped_identifier
+    "::"
+    name: (identifier) @function))
+
+(generic_function
+  function: (identifier) @function)
+(generic_function
+  function: (scoped_identifier
+    name: (identifier) @function))
+(generic_function
+  function: (field_expression
+    field: (field_identifier) @function.method))
+
+(macro_invocation
+  macro: (identifier) @function.macro
+  "!" @function.macro)
+
+; Function definitions
+
+(function_item (identifier) @function)
+(function_signature_item (identifier) @function)
+
+(line_comment) @comment
+(block_comment) @comment
+
+(line_comment (doc_comment)) @comment.documentation
+(block_comment (doc_comment)) @comment.documentation
+
+"(" @punctuation.bracket
+")" @punctuation.bracket
+"[" @punctuation.bracket
+"]" @punctuation.bracket
+"{" @punctuation.bracket
+"}" @punctuation.bracket
+
+(type_arguments
+  "<" @punctuation.bracket
+  ">" @punctuation.bracket)
+(type_parameters
+  "<" @punctuation.bracket
+  ">" @punctuation.bracket)
+
+"::" @punctuation.delimiter
+":" @punctuation.delimiter
+"." @punctuation.delimiter
+"," @punctuation.delimiter
+";" @punctuation.delimiter
+
+(parameter (identifier) @variable.parameter)
+
+(lifetime (identifier) @label)
+
+"as" @keyword
+"async" @keyword
+"await" @keyword
+"break" @keyword
+"const" @keyword
+"continue" @keyword
+"default" @keyword
+"dyn" @keyword
+"else" @keyword
+"enum" @keyword
+"extern" @keyword
+"fn" @keyword
+"for" @keyword
+"gen" @keyword
+"if" @keyword
+"impl" @keyword
+"in" @keyword
+"let" @keyword
+"loop" @keyword
+"macro_rules!" @keyword
+"match" @keyword
+"mod" @keyword
+"move" @keyword
+"pub" @keyword
+"raw" @keyword
+"ref" @keyword
+"return" @keyword
+"static" @keyword
+"struct" @keyword
+"trait" @keyword
+"type" @keyword
+"union" @keyword
+"unsafe" @keyword
+"use" @keyword
+"where" @keyword
+"while" @keyword
+"yield" @keyword
+(crate) @keyword
+(mutable_specifier) @keyword
+(use_list (self) @keyword)
+(scoped_use_list (self) @keyword)
+(scoped_identifier (self) @keyword)
+(super) @keyword
+
+(self) @variable.builtin
+
+(char_literal) @string
+(string_literal) @string
+(raw_string_literal) @string
+
+(boolean_literal) @constant.builtin
+(integer_literal) @constant.builtin
+(float_literal) @constant.builtin
+
+(escape_sequence) @escape
+
+(attribute_item) @attribute
+(inner_attribute_item) @attribute
+
+"*" @operator
+"&" @operator
+"'" @operator

Plugins/ZShell/Llm/highlight-queries/sql.scm

@@ -0,0 +1,463 @@
+; Vendored from sql (MIT License)
+; Source: https://github.com/DerekStride/tree-sitter-sql
+
+(object_reference
+  name: (identifier) @type)
+
+(invocation
+  (object_reference
+    name: (identifier) @function.call))
+
+[
+  (keyword_gist)
+  (keyword_btree)
+  (keyword_hash)
+  (keyword_spgist)
+  (keyword_gin)
+  (keyword_brin)
+  (keyword_array)
+  (keyword_object_id)
+] @function.call
+
+(relation
+  alias: (identifier) @variable)
+
+(field
+  name: (identifier) @field)
+
+(term
+  alias: (identifier) @variable)
+
+((term
+   value: (cast
+    name: (keyword_cast) @function.call
+    parameter: [(literal)]?)))
+
+(literal) @string
+(comment) @comment @spell
+(marginalia) @comment
+
+((literal) @number
+   (#match? @number "^[-+]?%d+$"))
+
+((literal) @float
+  (#match? @float "^[-+]?%d*\.%d*$"))
+
+(parameter) @parameter
+
+[
+ (keyword_true)
+ (keyword_false)
+] @boolean
+
+[
+ (keyword_asc)
+ (keyword_desc)
+ (keyword_terminated)
+ (keyword_escaped)
+ (keyword_unsigned)
+ (keyword_nulls)
+ (keyword_last)
+ (keyword_delimited)
+ (keyword_replication)
+ (keyword_auto_increment)
+ (keyword_default)
+ (keyword_collate)
+ (keyword_concurrently)
+ (keyword_engine)
+ (keyword_always)
+ (keyword_generated)
+ (keyword_preceding)
+ (keyword_following)
+ (keyword_first)
+ (keyword_current_timestamp)
+ (keyword_immutable)
+ (keyword_atomic)
+ (keyword_parallel)
+ (keyword_leakproof)
+ (keyword_safe)
+ (keyword_cost)
+ (keyword_strict)
+] @attribute
+
+[
+ (keyword_materialized)
+ (keyword_recursive)
+ (keyword_temp)
+ (keyword_temporary)
+ (keyword_unlogged)
+ (keyword_external)
+ (keyword_parquet)
+ (keyword_csv)
+ (keyword_rcfile)
+ (keyword_textfile)
+ (keyword_orc)
+ (keyword_avro)
+ (keyword_jsonfile)
+ (keyword_sequencefile)
+ (keyword_volatile)
+] @storageclass
+
+[
+ (keyword_case)
+ (keyword_when)
+ (keyword_then)
+ (keyword_else)
+] @conditional
+
+[
+  (keyword_select)
+  (keyword_from)
+  (keyword_where)
+  (keyword_index)
+  (keyword_join)
+  (keyword_primary)
+  (keyword_delete)
+  (keyword_create)
+  (keyword_show)
+  (keyword_unload)
+  (keyword_insert)
+  (keyword_merge)
+  (keyword_distinct)
+  (keyword_replace)
+  (keyword_update)
+  (keyword_into)
+  (keyword_overwrite)
+  (keyword_matched)
+  (keyword_values)
+  (keyword_value)
+  (keyword_attribute)
+  (keyword_set)
+  (keyword_left)
+  (keyword_right)
+  (keyword_outer)
+  (keyword_inner)
+  (keyword_full)
+  (keyword_order)
+  (keyword_partition)
+  (keyword_group)
+  (keyword_with)
+  (keyword_without)
+  (keyword_as)
+  (keyword_having)
+  (keyword_limit)
+  (keyword_offset)
+  (keyword_table)
+  (keyword_tables)
+  (keyword_key)
+  (keyword_references)
+  (keyword_foreign)
+  (keyword_constraint)
+  (keyword_force)
+  (keyword_use)
+  (keyword_include)
+  (keyword_for)
+  (keyword_if)
+  (keyword_exists)
+  (keyword_column)
+  (keyword_columns)
+  (keyword_cross)
+  (keyword_lateral)
+  (keyword_natural)
+  (keyword_alter)
+  (keyword_drop)
+  (keyword_add)
+  (keyword_view)
+  (keyword_end)
+  (keyword_is)
+  (keyword_using)
+  (keyword_between)
+  (keyword_window)
+  (keyword_no)
+  (keyword_data)
+  (keyword_type)
+  (keyword_rename)
+  (keyword_refresh)
+  (keyword_to)
+  (keyword_schema)
+  (keyword_owner)
+  (keyword_authorization)
+  (keyword_all)
+  (keyword_any)
+  (keyword_some)
+  (keyword_returning)
+  (keyword_begin)
+  (keyword_commit)
+  (keyword_rollback)
+  (keyword_transaction)
+  (keyword_only)
+  (keyword_like)
+  (keyword_rlike)
+  (keyword_similar)
+  (keyword_over)
+  (keyword_change)
+  (keyword_modify)
+  (keyword_after)
+  (keyword_before)
+  (keyword_range)
+  (keyword_rows)
+  (keyword_groups)
+  (keyword_exclude)
+  (keyword_current)
+  (keyword_ties)
+  (keyword_others)
+  (keyword_zerofill)
+  (keyword_format)
+  (keyword_fields)
+  (keyword_row)
+  (keyword_sort)
+  (keyword_compute)
+  (keyword_comment)
+  (keyword_location)
+  (keyword_cached)
+  (keyword_uncached)
+  (keyword_lines)
+  (keyword_stored)
+  (keyword_virtual)
+  (keyword_partitioned)
+  (keyword_analyze)
+  (keyword_explain)
+  (keyword_verbose)
+  (keyword_truncate)
+  (keyword_rewrite)
+  (keyword_optimize)
+  (keyword_vacuum)
+  (keyword_cache)
+  (keyword_language)
+  (keyword_called)
+  (keyword_conflict)
+  (keyword_declare)
+  (keyword_filter)
+  (keyword_function)
+  (keyword_input)
+  (keyword_name)
+  (keyword_oid)
+  (keyword_oids)
+  (keyword_precision)
+  (keyword_regclass)
+  (keyword_regnamespace)
+  (keyword_regproc)
+  (keyword_regtype)
+  (keyword_restricted)
+  (keyword_return)
+  (keyword_returns)
+  (keyword_separator)
+  (keyword_setof)
+  (keyword_stable)
+  (keyword_support)
+  (keyword_tblproperties)
+  (keyword_trigger)
+  (keyword_unsafe)
+  (keyword_admin)
+  (keyword_connection)
+  (keyword_cycle)
+  (keyword_database)
+  (keyword_encrypted)
+  (keyword_increment)
+  (keyword_logged)
+  (keyword_none)
+  (keyword_owned)
+  (keyword_password)
+  (keyword_reset)
+  (keyword_role)
+  (keyword_current_role)
+  (keyword_sequence)
+  (keyword_start)
+  (keyword_restart)
+  (keyword_tablespace)
+  (keyword_split)
+  (keyword_tablets)
+  (keyword_until)
+  (keyword_user)
+  (keyword_current_user)
+  (keyword_session_user)
+  (keyword_valid)
+  (keyword_action)
+  (keyword_definer)
+  (keyword_invoker)
+  (keyword_enable)
+  (keyword_disable)
+  (keyword_security)
+  (keyword_policy)
+  (keyword_permissive)
+  (keyword_restrictive)
+  (keyword_public)
+  (keyword_extension)
+  (keyword_version)
+  (keyword_out)
+  (keyword_inout)
+  (keyword_variadic)
+  (keyword_ordinality)
+  (keyword_session)
+  (keyword_isolation)
+  (keyword_level)
+  (keyword_serializable)
+  (keyword_repeatable)
+  (keyword_read)
+  (keyword_write)
+  (keyword_committed)
+  (keyword_uncommitted)
+  (keyword_deferrable)
+  (keyword_names)
+  (keyword_zone)
+  (keyword_immediate)
+  (keyword_deferred)
+  (keyword_constraints)
+  (keyword_snapshot)
+  (keyword_characteristics)
+  (keyword_off)
+  (keyword_follows)
+  (keyword_precedes)
+  (keyword_each)
+  (keyword_instead)
+  (keyword_of)
+  (keyword_initially)
+  (keyword_old)
+  (keyword_new)
+  (keyword_referencing)
+  (keyword_statement)
+  (keyword_execute)
+  (keyword_procedure)
+  (keyword_copy)
+  (keyword_delimiter)
+  (keyword_encoding)
+  (keyword_escape)
+  (keyword_force_not_null)
+  (keyword_force_null)
+  (keyword_force_quote)
+  (keyword_freeze)
+  (keyword_header)
+  (keyword_match)
+  (keyword_program)
+  (keyword_quote)
+  (keyword_stdin)
+  (keyword_extended)
+  (keyword_main)
+  (keyword_plain)
+  (keyword_storage)
+  (keyword_compression)
+  (keyword_duplicate)
+  (keyword_while)
+] @keyword
+
+[
+ (keyword_restrict)
+ (keyword_unbounded)
+ (keyword_unique)
+ (keyword_cascade)
+ (keyword_delayed)
+ (keyword_high_priority)
+ (keyword_low_priority)
+ (keyword_ignore)
+ (keyword_nothing)
+ (keyword_check)
+ (keyword_option)
+ (keyword_local)
+ (keyword_cascaded)
+ (keyword_wait)
+ (keyword_nowait)
+ (keyword_metadata)
+ (keyword_incremental)
+ (keyword_bin_pack)
+ (keyword_noscan)
+ (keyword_stats)
+ (keyword_statistics)
+ (keyword_maxvalue)
+ (keyword_minvalue)
+] @type.qualifier
+
+[
+  (keyword_int)
+  (keyword_null)
+  (keyword_boolean)
+  (keyword_binary)
+  (keyword_varbinary)
+  (keyword_image)
+  (keyword_bit)
+  (keyword_inet)
+  (keyword_character)
+  (keyword_smallserial)
+  (keyword_serial)
+  (keyword_bigserial)
+  (keyword_smallint)
+  (keyword_mediumint)
+  (keyword_bigint)
+  (keyword_tinyint)
+  (keyword_decimal)
+  (keyword_float)
+  (keyword_double)
+  (keyword_numeric)
+  (keyword_real)
+  (double)
+  (keyword_money)
+  (keyword_smallmoney)
+  (keyword_char)
+  (keyword_nchar)
+  (keyword_varchar)
+  (keyword_nvarchar)
+  (keyword_varying)
+  (keyword_text)
+  (keyword_string)
+  (keyword_uuid)
+  (keyword_json)
+  (keyword_jsonb)
+  (keyword_xml)
+  (keyword_bytea)
+  (keyword_enum)
+  (keyword_date)
+  (keyword_datetime)
+  (keyword_time)
+  (keyword_datetime2)
+  (keyword_datetimeoffset)
+  (keyword_smalldatetime)
+  (keyword_timestamp)
+  (keyword_timestamptz)
+  (keyword_geometry)
+  (keyword_geography)
+  (keyword_box2d)
+  (keyword_box3d)
+  (keyword_interval)
+] @type.builtin
+
+[
+  (keyword_in)
+  (keyword_and)
+  (keyword_or)
+  (keyword_not)
+  (keyword_by)
+  (keyword_on)
+  (keyword_do)
+  (keyword_union)
+  (keyword_except)
+  (keyword_intersect)
+] @keyword.operator
+
+[
+  "+"
+  "-"
+  "*"
+  "/"
+  "%"
+  "^"
+  ":="
+  "="
+  "<"
+  "<="
+  "!="
+  ">="
+  ">"
+  "<>"
+  (op_other)
+  (op_unary_other)
+] @operator
+
+[
+  "("
+  ")"
+] @punctuation.bracket
+
+[
+  ";"
+  ","
+  "."
+] @punctuation.delimiter

Plugins/ZShell/Llm/highlight-queries/toml.scm

@@ -0,0 +1,36 @@
+; Vendored from toml (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-toml
+
+; Properties
+;-----------
+
+(bare_key) @property
+(quoted_key) @string
+
+; Literals
+;---------
+
+(boolean) @constant.builtin
+(comment) @comment
+(string) @string
+(integer) @number
+(float) @number
+(offset_date_time) @string.special
+(local_date_time) @string.special
+(local_date) @string.special
+(local_time) @string.special
+
+; Punctuation
+;------------
+
+"." @punctuation.delimiter
+"," @punctuation.delimiter
+
+"=" @operator
+
+"[" @punctuation.bracket
+"]" @punctuation.bracket
+"[[" @punctuation.bracket
+"]]" @punctuation.bracket
+"{" @punctuation.bracket
+"}" @punctuation.bracket

Plugins/ZShell/Llm/highlight-queries/typescript.scm

@@ -0,0 +1,38 @@
+; Vendored from typescript (MIT License)
+; Source: https://github.com/tree-sitter/tree-sitter-typescript
+
+; Types
+
+(type_identifier) @type
+(predefined_type) @type.builtin
+
+((identifier) @type
+ (#match? @type "^[A-Z]"))
+
+(type_arguments
+  "<" @punctuation.bracket
+  ">" @punctuation.bracket)
+
+; Variables
+
+(required_parameter (identifier) @variable.parameter)
+(optional_parameter (identifier) @variable.parameter)
+
+; Keywords
+
+[ "abstract"
+  "declare"
+  "enum"
+  "export"
+  "implements"
+  "interface"
+  "keyof"
+  "namespace"
+  "private"
+  "protected"
+  "public"
+  "type"
+  "readonly"
+  "override"
+  "satisfies"
+] @keyword

Plugins/ZShell/Llm/highlight-queries/yaml.scm

@@ -0,0 +1,82 @@
+; Vendored from yaml (MIT License)
+; Source: https://github.com/tree-sitter-grammars/tree-sitter-yaml
+
+(boolean_scalar) @boolean
+
+(null_scalar) @constant.builtin
+
+[
+  (double_quote_scalar)
+  (single_quote_scalar)
+  (block_scalar)
+  (string_scalar)
+] @string
+
+[
+  (integer_scalar)
+  (float_scalar)
+] @number
+
+(comment) @comment
+
+[
+  (anchor_name)
+  (alias_name)
+] @label
+
+(tag) @type
+
+[
+  (yaml_directive)
+  (tag_directive)
+  (reserved_directive)
+] @attribute
+
+(block_mapping_pair
+  key: (flow_node
+    [
+      (double_quote_scalar)
+      (single_quote_scalar)
+    ] @property))
+
+(block_mapping_pair
+  key: (flow_node
+    (plain_scalar
+      (string_scalar) @property)))
+
+(flow_mapping
+  (_
+    key: (flow_node
+      [
+        (double_quote_scalar)
+        (single_quote_scalar)
+      ] @property)))
+
+(flow_mapping
+  (_
+    key: (flow_node
+      (plain_scalar
+        (string_scalar) @property))))
+
+[
+  ","
+  "-"
+  ":"
+  ">"
+  "?"
+  "|"
+] @punctuation.delimiter
+
+[
+  "["
+  "]"
+  "{"
+  "}"
+] @punctuation.bracket
+
+[
+  "*"
+  "&"
+  "---"
+  "..."
+] @punctuation.special

Plugins/ZShell/Llm/llmclient.cpp

@@ -1,7 +1,9 @@
 #include "llmclient.hpp"
 
+#include "generation.hpp"
 #include "message.hpp"
 #include "messagemodel.hpp"
+#include "segment.hpp"
 #include "session.hpp"
 
 #include <QDebug>
@@ -62,12 +64,19 @@ void LlmClient::setStreamingChatId(const QString& id) {
 }
 
 LlmClient::LlmClient(QObject* parent) : QObject(parent) {
+	m_tools = new ToolRegistry(this);
+	connect(
+		m_tools,
+		&ToolRegistry::enabledChanged,
+		this,
+		&LlmClient::toolsEnabledChanged);
 	probeContextSize();
 }
 
 LlmClient::~LlmClient() {
 	if (m_reply)
 		m_reply->abort();
+	m_tools->cancelAll();
 	endStream();
 }
 
@@ -111,10 +120,6 @@ void LlmClient::startGeneration(
 		return;
 	}
 
-	QNetworkRequest request(url);
-	request.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
-	request.setRawHeader("Accept", "text/event-stream");
-
 	const auto* model = session->messagesModel();
 	const int targetRow =
 		model->rowOf(qobject_cast<ChatMessage*>(target->parent()));
@@ -124,25 +129,50 @@ void LlmClient::startGeneration(
 		return;
 	}
 
-	// Context: every message older than the target, oldest first.
-	QJsonArray messages;
-	for (int row = model->rowCount() - 1; row > targetRow; --row) {
-		const auto* message = model->at(row);
-		const auto* generation = message->activeGeneration();
-		if (!generation)
-			continue;
-		QJsonObject messageObj;
-		messageObj[QStringLiteral("role")] =
-			message->role() == ChatMessage::Role::User
-				? QStringLiteral("user")
-				: QStringLiteral("assistant");
-		messageObj[QStringLiteral("content")] = generation->content();
-		if (!generation->reasoning().isEmpty())
-			messageObj[QStringLiteral("reasoning_content")] =
-				generation->reasoning();
-		messages.append(messageObj);
+	m_transcript = QJsonArray();
+	m_callBuilders.clear();
+	m_callResults.clear();
+	m_finishReason.clear();
+	m_round = 0;
+	m_contentMark = 0;
+	m_reasoningMark = 0;
+
+	sendRound();
+}
+
+void LlmClient::sendRound() {
+	if (!m_active || !m_streaming)
+		return;
+	m_finishReason.clear();
+	m_callBuilders.clear();
+	m_callResults.clear();
+	m_roundDone = false;
+	m_toolPhase = false;
+	m_pendingCalls = 0;
+	m_buffer.clear();
+
+	const QUrl url =
+		QUrl::fromUserInput(completionsPath(m_endpoint, "/chat/completions"));
+	if (!url.isValid() || url.host().isEmpty()) {
+		fail(QStringLiteral("Invalid LLM endpoint: %1").arg(m_endpoint));
+		return;
 	}
 
+	const auto* model = m_active->messagesModel();
+	const int targetRow =
+		model->rowOf(qobject_cast<ChatMessage*>(m_streaming->parent()));
+	if (targetRow < 0) {
+		fail(QStringLiteral("Internal error: generation target is not in the "
+		                    "session"));
+		return;
+	}
+
+	// Context: every message older than the target, oldest first, plus
+	// the tool exchanges of the current turn so far.
+	QJsonArray messages = buildContextMessages(m_active, targetRow);
+	for (const QJsonValue& value : m_transcript)
+		messages.append(value);
+
 	QJsonObject body;
 	QJsonObject streamOptions;
 	streamOptions[QStringLiteral("include_usage")] = true;
@@ -152,8 +182,14 @@ void LlmClient::startGeneration(
 	body[QStringLiteral("temperature")] = m_temperature;
 	if (!m_model.isEmpty())
 		body[QStringLiteral("model")] = m_model;
+	const QJsonArray toolSpecs = m_tools->specifications();
+	if (!toolSpecs.isEmpty())
+		body[QStringLiteral("tools")] = toolSpecs;
+
+	QNetworkRequest request(url);
+	request.setHeader(QNetworkRequest::ContentTypeHeader, "application/json");
+	request.setRawHeader("Accept", "text/event-stream");
 
-	m_buffer.clear();
 	m_reply = m_manager.post(request, QJsonDocument(body).toJson());
 
 	connect(m_reply, &QNetworkReply::readyRead, this, [this]() {
@@ -178,14 +214,493 @@ void LlmClient::startGeneration(
 		if (!m_streaming)
 			return;
 
-		if (error == QNetworkReply::NoError ||
-			error == QNetworkReply::OperationCanceledError)
-			finalize();
+		if (error == QNetworkReply::NoError)
+			roundFinished();
+		else if (error == QNetworkReply::OperationCanceledError)
+			finishTurn();
 		else
 			fail(serverErrorMessage(responseBody, errorString));
 	});
 }
 
+QJsonArray LlmClient::buildContextMessages(
+	ChatSession* session, int stopBeforeRow) const {
+	const auto* model = session->messagesModel();
+	QJsonArray messages;
+	for (int row = model->rowCount() - 1; row > stopBeforeRow; --row) {
+		const auto* message = model->at(row);
+		const auto* generation = message->activeGeneration();
+		if (!generation)
+			continue;
+
+		if (message->role() == ChatMessage::Role::User) {
+			QJsonObject user;
+			user[QStringLiteral("role")] = QStringLiteral("user");
+			user[QStringLiteral("content")] = generation->content();
+			messages.append(user);
+			continue;
+		}
+
+		// Assistant message: replay its tool calls (and their results)
+		// so the model keeps the full history of the turn.
+		QList<const LlmSegment*> toolSegments;
+		for (const auto* segment : generation->segments())
+			if (segment->type() == LlmSegment::Type::ToolCall)
+				toolSegments.append(segment);
+
+		QJsonObject assistant;
+		assistant[QStringLiteral("role")] = QStringLiteral("assistant");
+		if (toolSegments.isEmpty())
+			assistant[QStringLiteral("content")] = generation->content();
+		else if (!generation->content().isEmpty())
+			assistant[QStringLiteral("content")] = generation->content();
+		if (!generation->reasoning().isEmpty())
+			assistant[QStringLiteral("reasoning_content")] =
+				generation->reasoning();
+
+		if (toolSegments.isEmpty()) {
+			messages.append(assistant);
+			continue;
+		}
+		QJsonArray calls;
+		for (const auto* segment : toolSegments) {
+			QJsonObject function;
+			function[QStringLiteral("name")] = segment->name();
+			function[QStringLiteral("arguments")] = segment->arguments();
+			QJsonObject call;
+			call[QStringLiteral("id")] = segment->toolCallId();
+			call[QStringLiteral("type")] = QStringLiteral("function");
+			call[QStringLiteral("function")] = function;
+			calls.append(call);
+		}
+		assistant[QStringLiteral("tool_calls")] = calls;
+		messages.append(assistant);
+		for (const auto* segment : toolSegments) {
+			QJsonObject toolMessage;
+			toolMessage[QStringLiteral("role")] = QStringLiteral("tool");
+			toolMessage[QStringLiteral("tool_call_id")] =
+				segment->toolCallId();
+			toolMessage[QStringLiteral("content")] = segment->result();
+			messages.append(toolMessage);
+		}
+	}
+	return messages;
+}
+
+void LlmClient::applyToolCallDelta(const QJsonObject& call) {
+	if (!m_streaming)
+		return;
+	const int index = call[QStringLiteral("index")].toInt(-1);
+	if (index < 0)
+		return;
+	while (m_callBuilders.size() <= index)
+		m_callBuilders.append(ToolCallBuilder{});
+	ToolCallBuilder& builder = m_callBuilders[index];
+	builder.seen = true;
+	const QString id = call[QStringLiteral("id")].toString();
+	if (!id.isEmpty())
+		builder.id = id;
+	const QJsonObject function = call[QStringLiteral("function")].toObject();
+	const QString name = function[QStringLiteral("name")].toString();
+	if (!name.isEmpty())
+		builder.name = name;
+	const QString arguments =
+		function[QStringLiteral("arguments")].toString();
+	if (!arguments.isEmpty())
+		builder.arguments += arguments;
+
+	if (!builder.segment) {
+		// A new call: close the in-flight text segments and open a
+		// running tool-call segment so the UI can track it live.
+		m_streaming->closeOpenSegments();
+		builder.segment =
+			m_streaming->beginToolCall(builder.name, builder.id);
+	}
+	builder.segment->setName(builder.name);
+	builder.segment->setToolCallId(builder.id);
+	if (!arguments.isEmpty())
+		builder.segment->appendArguments(arguments);
+}
+
+void LlmClient::roundFinished() {
+	// [DONE] and the reply's finished signal both funnel here; only the
+	// first may act.
+	if (m_roundDone || !m_streaming)
+		return;
+	m_roundDone = true;
+
+	bool hasCalls = false;
+	for (const auto& builder : m_callBuilders)
+		if (builder.seen)
+			hasCalls = true;
+	if (m_finishReason != QLatin1String("tool_calls") && !hasCalls) {
+		finishTurn();
+		return;
+	}
+	if (m_round >= kMaxToolRounds) {
+		qWarning() << "LlmClient: tool round limit reached, ending turn";
+		finishTurn();
+		return;
+	}
+
+	m_streaming->closeOpenSegments();
+
+	// Record the assistant's tool-call message in the transcript so the
+	// next round (and the model) can see it.
+	const QString content = m_streaming->content();
+	const QString reasoning = m_streaming->reasoning();
+	QJsonObject assistant;
+	assistant[QStringLiteral("role")] = QStringLiteral("assistant");
+	if (content.size() > m_contentMark) {
+		assistant[QStringLiteral("content")] = content.mid(m_contentMark);
+		m_contentMark = content.size();
+	}
+	if (reasoning.size() > m_reasoningMark) {
+		assistant[QStringLiteral("reasoning_content")] =
+			reasoning.mid(m_reasoningMark);
+		m_reasoningMark = reasoning.size();
+	}
+	QJsonArray calls;
+	for (const auto& builder : m_callBuilders) {
+		if (!builder.seen)
+			continue;
+		QJsonObject function;
+		function[QStringLiteral("name")] = builder.name;
+		function[QStringLiteral("arguments")] = builder.arguments;
+		QJsonObject call;
+		call[QStringLiteral("id")] = builder.id;
+		call[QStringLiteral("type")] = QStringLiteral("function");
+		call[QStringLiteral("function")] = function;
+		calls.append(call);
+	}
+	assistant[QStringLiteral("tool_calls")] = calls;
+	m_transcript.append(assistant);
+	m_round++;
+
+	executeAllCalls();
+}
+
+void LlmClient::executeAllCalls() {
+	m_toolPhase = true;
+	m_pendingCalls = 0;
+	m_callResults = QList<ToolCallResult>(m_callBuilders.size());
+
+	for (int i = 0; i < m_callBuilders.size(); ++i) {
+		const auto& call = m_callBuilders.at(i);
+		if (!call.seen)
+			continue;
+
+		LlmTool* tool = m_tools->tool(call.name);
+		QJsonObject args;
+		QString errorText;
+		if (!tool) {
+			errorText = QStringLiteral("Error: unknown tool '%1'")
+				.arg(call.name);
+		} else if (!call.arguments.isEmpty()) {
+			const QJsonDocument doc =
+				QJsonDocument::fromJson(call.arguments.toUtf8());
+			if (!doc.isObject()) {
+				errorText =
+					QStringLiteral("Error: tool arguments are not valid "
+					               "JSON: %1")
+						.arg(call.arguments);
+			} else {
+				args = doc.object();
+			}
+		}
+		if (!errorText.isEmpty()) {
+			m_callResults[i] = { errorText, false };
+			if (LlmSegment* segment = call.segment)
+				segment->finishTool(errorText, false);
+			continue;
+		}
+
+		++m_pendingCalls;
+		tool->execute(
+			args,
+			[this, i, call](const QJsonObject& result) {
+				if (!m_streaming)
+					return;
+				const bool success =
+					result.contains(QStringLiteral("output"));
+				const QString content = success
+					? result[QStringLiteral("output")].toString()
+					: QStringLiteral("Error: ") +
+							result[QStringLiteral("error")].toString();
+				m_callResults[i] = { content, success };
+				if (LlmSegment* segment = call.segment)
+					segment->finishTool(content, success);
+				if (--m_pendingCalls == 0)
+					flushCallResults();
+			});
+	}
+	if (m_pendingCalls == 0)
+		flushCallResults();
+}
+
+void LlmClient::flushCallResults() {
+	if (!m_toolPhase)
+		return;
+	m_toolPhase = false;
+	if (!m_streaming)
+		return;
+	for (int i = 0; i < m_callBuilders.size(); ++i) {
+		const auto& call = m_callBuilders.at(i);
+		if (!call.seen)
+			continue;
+		QJsonObject toolMessage;
+		toolMessage[QStringLiteral("role")] = QStringLiteral("tool");
+		toolMessage[QStringLiteral("tool_call_id")] = call.id;
+		toolMessage[QStringLiteral("content")] =
+			m_callResults.at(i).content;
+		m_transcript.append(toolMessage);
+	}
+	sendRound();
+}
+
+void LlmClient::stop() {
+	if (!m_busy)
+		return;
+	if (m_toolPhase) {
+		m_tools->cancelAll();
+		if (m_streaming) {
+			for (auto* segment : m_streaming->segments()) {
+				if (segment->type() == LlmSegment::Type::ToolCall &&
+				    segment->running())
+					segment->finishTool(
+						QStringLiteral("Cancelled"), false);
+			}
+		}
+		finishTurn();
+		return;
+	}
+	if (m_reply)
+		m_reply->abort();
+}
+
+void LlmClient::endStream() {
+	if (!m_streaming)
+		return;
+	auto* generation = m_streaming;
+	auto* session = m_active;
+	m_streaming = nullptr;
+	m_active = nullptr;
+	generation->setStreaming(false);
+	if (generation->content().isEmpty() &&
+	    generation->reasoning().isEmpty() &&
+	    generation->toolCallCount() == 0) {
+		if (auto* message = qobject_cast<ChatMessage*>(generation->parent())) {
+			if (message->generationCount() <= 1) {
+				if (session)
+					session->removeMessage(message);
+			} else {
+				message->removeGeneration(generation);
+			}
+		}
+	}
+	setBusy(false);
+	setStreamingChatId(QString());
+}
+
+void LlmClient::finishTurn() {
+	if (!m_streaming)
+		return;
+	ChatSession* session = m_active;
+	endStream();
+	if (session && m_pendingClear == session) {
+		session->clearMessages();
+		m_pendingClear.clear();
+	}
+	if (session)
+		session->persist();
+}
+
+void LlmClient::clearOnFinish(ChatSession* session) {
+	m_pendingClear = session;
+}
+
+void LlmClient::sessionRemoved(ChatSession* session) {
+	if (m_pendingClear == session)
+		m_pendingClear.clear();
+	if (m_active == session) {
+		stop();
+		endStream();
+	}
+}
+
+void LlmClient::fail(const QString& message) {
+	qWarning() << "LlmClient:" << message;
+	finishTurn();
+	Q_EMIT errorOccurred(message);
+}
+
+void LlmClient::drainBuffer() {
+	while (true) {
+		const qsizetype newline = m_buffer.indexOf('\n');
+		if (newline < 0)
+			break;
+		const QByteArray line = m_buffer.left(newline).trimmed();
+		m_buffer.remove(0, newline + 1);
+		handleLine(line);
+	}
+}
+
+void LlmClient::handleLine(const QByteArray& line) {
+	if (!m_streaming || line.isEmpty() || !line.startsWith("data:"))
+		return;
+
+	const QByteArray data = line.mid(5).trimmed();
+	if (data == "[DONE]") {
+		roundFinished();
+		return;
+	}
+
+	const QJsonDocument doc = QJsonDocument::fromJson(data);
+	if (!doc.isObject())
+		return;
+	const QJsonObject obj = doc.object();
+	updateTokenUsage(obj);
+
+	if (obj.contains("error")) {
+		const QJsonObject error = obj["error"].toObject();
+		const QString message = error["message"].toString();
+		fail(message.isEmpty()
+			 ? QStringLiteral("LLM server returned an error")
+			 : message);
+		return;
+	}
+
+	for (const QJsonValue& choiceValue : obj["choices"].toArray()) {
+		if (!m_streaming)
+			continue;
+		const QJsonObject choice = choiceValue.toObject();
+		const QJsonObject delta = choice["delta"].toObject();
+
+		const QString finishReason =
+			choice[QStringLiteral("finish_reason")].toString();
+		if (!finishReason.isEmpty())
+			m_finishReason = finishReason;
+
+		m_streaming->appendContent(delta["content"].toString());
+		QString reasoning =
+			delta["reasoning_content"].toString();
+		if (reasoning.isEmpty())
+			reasoning = delta["reasoning"].toString();
+		m_streaming->appendReasoning(reasoning);
+
+		for (const QJsonValue& callValue :
+		     delta["tool_calls"].toArray()) {
+			if (!m_streaming)
+				break;
+			applyToolCallDelta(callValue.toObject());
+		}
+	}
+}
+
+void LlmClient::refreshModels() {
+	const QUrl url =
+		QUrl::fromUserInput(completionsPath(m_endpoint, "/models"));
+	if (!url.isValid() || url.host().isEmpty())
+		return;
+
+	auto* reply = m_manager.get(QNetworkRequest(url));
+	connect(reply, &QNetworkReply::finished, this, [this, reply]() {
+		const QNetworkReply::NetworkError error = reply->error();
+		const QByteArray data = reply->readAll();
+		reply->deleteLater();
+
+		if (error != QNetworkReply::NoError) {
+			qWarning() << "LlmClient: failed to fetch models:" << error;
+			return;
+		}
+		const QJsonArray dataArr =
+			QJsonDocument::fromJson(data).object()["data"].toArray();
+
+		QStringList models;
+		QSet<QString> seen;
+		for (const QJsonValue& value : dataArr) {
+			const QString id = value.toObject()["id"].toString();
+			if (!id.isEmpty() && !seen.contains(id)) {
+				seen.insert(id);
+				models.append(id);
+			}
+		}
+		if (models.isEmpty())
+			return;
+
+		m_availableModels = models;
+		Q_EMIT availableModelsChanged();
+
+		if (m_model.isEmpty()) {
+			m_model = models.first();
+			Q_EMIT modelChanged();
+		}
+	});
+}
+
+void LlmClient::setContextSize(int size) {
+	if (size <= 0 || m_contextSize == size)
+		return;
+	m_contextSize = size;
+	Q_EMIT contextSizeChanged();
+}
+
+void LlmClient::probeContextSize() {
+	// llama.cpp-specific endpoint; other servers fall back to 4096.
+	QString base = m_endpoint.trimmed();
+	while (base.endsWith('/'))
+		base.chop(1);
+	const QUrl url = QUrl::fromUserInput(base + "/props");
+	if (!url.isValid() || url.host().isEmpty())
+		return;
+
+	auto* reply = m_manager.get(QNetworkRequest(url));
+	connect(
+		reply,
+		&QNetworkReply::finished,
+		this,
+		[this, reply]() {
+			const QNetworkReply::NetworkError error = reply->error();
+			const QByteArray data = reply->readAll();
+			reply->deleteLater();
+
+			int size = 0;
+			if (error == QNetworkReply::NoError) {
+				const QJsonDocument doc = QJsonDocument::fromJson(data);
+				if (doc.isArray()) {
+					for (const auto& value : doc.array()) {
+						const QJsonObject slot = value.toObject();
+						if (slot.contains("n_ctx")) {
+							size = slot["n_ctx"].toInt(0);
+							if (size > 0)
+								break;
+						}
+					}
+				} else if (doc.isObject()) {
+					const QJsonObject obj = doc.object();
+					size = obj["n_ctx"].toInt(0);
+					if (size <= 0)
+						size = obj["default_generation_settings"].toObject()
+						       ["n_ctx"].toInt(0);
+				}
+			}
+			setContextSize(size > 0 ? size : 4096);
+		});
+}
+
+void LlmClient::updateTokenUsage(const QJsonObject& data) {
+	if (!m_active || m_contextSize <= 0)
+		return;
+	const QJsonObject usage = data["usage"].toObject();
+	if (usage.isEmpty())
+		return;
+	const double used =
+		usage.value("prompt_tokens").toDouble() +
+		usage.value("completion_tokens").toDouble();
+	if (used > 0)
+		m_active->setLastTokenCount(static_cast<int>(used));
+}
+
 void LlmClient::shortRequest(
 	const QString& tag,
 	const QString& systemPrompt,
@@ -340,229 +855,4 @@ void LlmClient::requestIcon(ChatSession* session, const QString& userText) {
 		});
 }
 
-void LlmClient::stop() {
-	if (!m_busy)
-		return;
-	if (m_reply)
-		m_reply->abort();
-}
-
-void LlmClient::endStream() {
-	if (!m_streaming)
-		return;
-	auto* generation = m_streaming;
-	auto* session = m_active;
-	m_streaming = nullptr;
-	m_active = nullptr;
-	generation->setStreaming(false);
-	if (generation->content().isEmpty() &&
-	    generation->reasoning().isEmpty()) {
-		if (auto* message = qobject_cast<ChatMessage*>(generation->parent())) {
-			if (message->generationCount() <= 1) {
-				if (session)
-					session->removeMessage(message);
-			} else {
-				message->removeGeneration(generation);
-			}
-		}
-	}
-	setBusy(false);
-	setStreamingChatId(QString());
-}
-
-void LlmClient::finalize() {
-	if (!m_streaming)
-		return;
-	ChatSession* session = m_active;
-	endStream();
-	if (session && m_pendingClear == session) {
-		session->clearMessages();
-		m_pendingClear.clear();
-	}
-	if (session)
-		session->persist();
-}
-
-void LlmClient::clearOnFinish(ChatSession* session) {
-	m_pendingClear = session;
-}
-
-void LlmClient::sessionRemoved(ChatSession* session) {
-	if (m_pendingClear == session)
-		m_pendingClear.clear();
-	if (m_active == session) {
-		stop();
-		endStream();
-	}
-}
-
-void LlmClient::fail(const QString& message) {
-	qWarning() << "LlmClient:" << message;
-
-	if (m_streaming) {
-		ChatSession* session = m_active;
-		endStream();
-		if (session && m_pendingClear == session) {
-			session->clearMessages();
-			m_pendingClear.clear();
-		}
-		if (session)
-			session->persist();
-	}
-	Q_EMIT errorOccurred(message);
-}
-
-void LlmClient::drainBuffer() {
-	while (true) {
-		const qsizetype newline = m_buffer.indexOf('\n');
-		if (newline < 0)
-			break;
-		const QByteArray line = m_buffer.left(newline).trimmed();
-		m_buffer.remove(0, newline + 1);
-		handleLine(line);
-	}
-}
-
-void LlmClient::handleLine(const QByteArray& line) {
-	if (!m_streaming || line.isEmpty() || !line.startsWith("data:"))
-		return;
-
-	const QByteArray data = line.mid(5).trimmed();
-	if (data == "[DONE]") {
-		finalize();
-		return;
-	}
-
-	const QJsonDocument doc = QJsonDocument::fromJson(data);
-	if (!doc.isObject())
-		return;
-	const QJsonObject obj = doc.object();
-	updateTokenUsage(obj);
-
-	if (obj.contains("error")) {
-		const QJsonObject error = obj["error"].toObject();
-		const QString message = error["message"].toString();
-		fail(message.isEmpty()
-			 ? QStringLiteral("LLM server returned an error")
-			 : message);
-		return;
-	}
-
-	for (const QJsonValue& choiceValue : obj["choices"].toArray()) {
-		if (!m_streaming)
-			continue;
-		const QJsonObject delta =
-			choiceValue.toObject()["delta"].toObject();
-		m_streaming->appendContent(delta["content"].toString());
-		QString reasoning = delta["reasoning_content"].toString();
-		if (reasoning.isEmpty())
-			reasoning = delta["reasoning"].toString();
-		m_streaming->appendReasoning(reasoning);
-	}
-}
-
-void LlmClient::refreshModels() {
-	const QUrl url =
-		QUrl::fromUserInput(completionsPath(m_endpoint, "/models"));
-	if (!url.isValid() || url.host().isEmpty())
-		return;
-
-	auto* reply = m_manager.get(QNetworkRequest(url));
-	connect(reply, &QNetworkReply::finished, this, [this, reply]() {
-		const QNetworkReply::NetworkError error = reply->error();
-		const QByteArray data = reply->readAll();
-		reply->deleteLater();
-
-		if (error != QNetworkReply::NoError) {
-			qWarning() << "LlmClient: failed to fetch models:" << error;
-			return;
-		}
-		const QJsonArray dataArr =
-			QJsonDocument::fromJson(data).object()["data"].toArray();
-
-		QStringList models;
-		QSet<QString> seen;
-		for (const QJsonValue& value : dataArr) {
-			const QString id = value.toObject()["id"].toString();
-			if (!id.isEmpty() && !seen.contains(id)) {
-				seen.insert(id);
-				models.append(id);
-			}
-		}
-		if (models.isEmpty())
-			return;
-
-		m_availableModels = models;
-		Q_EMIT availableModelsChanged();
-
-		if (m_model.isEmpty()) {
-			m_model = models.first();
-			Q_EMIT modelChanged();
-		}
-	});
-}
-
-void LlmClient::setContextSize(int size) {
-	if (size <= 0 || m_contextSize == size)
-		return;
-	m_contextSize = size;
-	Q_EMIT contextSizeChanged();
-}
-
-void LlmClient::probeContextSize() {
-	// llama.cpp-specific endpoint; other servers fall back to 4096.
-	QString base = m_endpoint.trimmed();
-	while (base.endsWith('/'))
-		base.chop(1);
-	const QUrl url = QUrl::fromUserInput(base + "/props");
-	if (!url.isValid() || url.host().isEmpty())
-		return;
-
-	auto* reply = m_manager.get(QNetworkRequest(url));
-	connect(
-		reply,
-		&QNetworkReply::finished,
-		this,
-		[this, reply]() {
-			const QNetworkReply::NetworkError error = reply->error();
-			const QByteArray data = reply->readAll();
-			reply->deleteLater();
-
-			int size = 0;
-			if (error == QNetworkReply::NoError) {
-				const QJsonDocument doc = QJsonDocument::fromJson(data);
-				if (doc.isArray()) {
-					for (const auto& value : doc.array()) {
-						const QJsonObject slot = value.toObject();
-						if (slot.contains("n_ctx")) {
-							size = slot["n_ctx"].toInt(0);
-							if (size > 0)
-								break;
-						}
-					}
-				} else if (doc.isObject()) {
-					const QJsonObject obj = doc.object();
-					size = obj["n_ctx"].toInt(0);
-					if (size <= 0)
-						size = obj["default_generation_settings"].toObject()
-						       ["n_ctx"].toInt(0);
-				}
-			}
-			setContextSize(size > 0 ? size : 4096);
-		});
-}
-
-void LlmClient::updateTokenUsage(const QJsonObject& data) {
-	if (!m_active || m_contextSize <= 0)
-		return;
-	const QJsonObject usage = data["usage"].toObject();
-	if (usage.isEmpty())
-		return;
-	const double used =
-		usage.value("prompt_tokens").toDouble() +
-		usage.value("completion_tokens").toDouble();
-	if (used > 0)
-		m_active->setLastTokenCount(static_cast<int>(used));
-}
-
 } // namespace ZShell::llm

Plugins/ZShell/Llm/llmclient.hpp

@@ -1,11 +1,16 @@
 #pragma once
 
+#include "segment.hpp"
+#include "tool.hpp"
+
 #include <QByteArray>
+#include <QJsonArray>
 #include <QNetworkAccessManager>
 #include <QObject>
 #include <QPointer>
 #include <QString>
 #include <QStringList>
+
 #include <functional>
 
 class QJsonObject;
@@ -15,9 +20,12 @@ namespace ZShell::llm {
 
 class ChatGeneration;
 class ChatSession;
+class LlmSegment;
+class LlmTool;
 
 // The only component that talks to the LLM server: owns the network
-// manager, the in-flight streaming state and the SSE parsing.
+// manager, the in-flight streaming state, the SSE parsing and the
+// tool-calling loop.
 class LlmClient : public QObject {
 	Q_OBJECT
 
@@ -32,6 +40,7 @@ class LlmClient : public QObject {
 		return m_availableModels;
 	}
 	[[nodiscard]] int contextSize() const { return m_contextSize; }
+	[[nodiscard]] ToolRegistry* tools() const { return m_tools; }
 	void setEndpoint(const QString& value);
 	void setModel(const QString& value);
 	void setTemperature(double value);
@@ -39,12 +48,16 @@ class LlmClient : public QObject {
 	void probeContextSize();
 
 	[[nodiscard]] bool busy() const { return m_busy; }
+	[[nodiscard]] bool toolsEnabled() const { return m_tools->enabled(); }
+	void setToolsEnabled(bool value) { m_tools->setEnabled(value); }
 	[[nodiscard]] QString streamingChatId() const { return m_streamingChatId; }
 	[[nodiscard]] ChatSession* streamingSession() const { return m_active; }
 
-	// Streams a new assistant reply into `target` (the active generation of
-	// a message of `session`). The context sent to the model is every
+	// Streams a new assistant reply into `target` (the active generation
+	// of a message of `session`). The context sent to the model is every
 	// message of the session that is older than the target message.
+	// Tool calls made by the model are executed and fed back
+	// transparently until the model produces its final answer.
 	void startGeneration(ChatSession* session, ChatGeneration* target);
 	void stop();
 	void endStream();
@@ -63,13 +76,39 @@ class LlmClient : public QObject {
 	void modelChanged();
 	void availableModelsChanged();
 	void contextSizeChanged();
+	void toolsEnabledChanged();
 	void streamingChatIdChanged();
 	void errorOccurred(const QString& message);
 	void titleSuggested(ZShell::llm::ChatSession* session, const QString& title);
 	void iconSuggested(ZShell::llm::ChatSession* session, const QString& icon);
 
 	private:
-	void finalize();
+	struct ToolCallBuilder {
+		QString id;
+		QString name;
+		QString arguments;
+		QPointer<LlmSegment> segment;
+		bool seen = false;
+	};
+
+	// Sends one streaming round: context + transcript so far.
+	void sendRound();
+	// The session context for a round, oldest first, ending just before
+	// `stopBeforeRow` (the message of the generation being streamed).
+	QJsonArray buildContextMessages(
+		ChatSession* session, int stopBeforeRow) const;
+	void applyToolCallDelta(const QJsonObject& call);
+	// One round's stream ended; either ends the turn or executes the
+	// requested tool calls and sends the next round. Runs at most once
+	// per round ([DONE] and the reply's finished signal both reach it).
+	void roundFinished();
+	// Dispatches every call of the round; tools run concurrently.
+	void executeAllCalls();
+	// All results in: appends the tool messages (in call order) and
+	// sends the next round.
+	void flushCallResults();
+	// Ends the current turn gracefully and persists the session.
+	void finishTurn();
 	void fail(const QString& message);
 	void handleLine(const QByteArray& line);
 	void drainBuffer();
@@ -86,6 +125,7 @@ class LlmClient : public QObject {
 		const QByteArray& body, const QString& fallback);
 
 	QNetworkAccessManager m_manager;
+	ToolRegistry* m_tools = nullptr;
 	QNetworkReply* m_reply = nullptr;
 	QByteArray m_buffer;
 	ChatSession* m_active = nullptr;
@@ -98,6 +138,23 @@ class LlmClient : public QObject {
 	QStringList m_availableModels;
 	double m_temperature = 0.7;
 	int m_contextSize = 0;
+
+	// State of the multi-round tool loop of the current turn.
+	QJsonArray m_transcript;
+	QList<ToolCallBuilder> m_callBuilders;
+	struct ToolCallResult {
+		QString content;
+		bool success = false;
+	};
+	QList<ToolCallResult> m_callResults;
+	QString m_finishReason;
+	int m_round = 0;
+	qsizetype m_contentMark = 0;
+	qsizetype m_reasoningMark = 0;
+	bool m_roundDone = false;
+	bool m_toolPhase = false;
+	int m_pendingCalls = 0;
+	static constexpr int kMaxToolRounds = 12;
 };
 
 } // namespace ZShell::llm

Plugins/ZShell/Llm/markdownblock.hpp

@@ -0,0 +1,28 @@
+#pragma once
+
+#include <QObject>
+#include <QtQml>
+
+namespace ZShell::llm {
+
+// Marker class exposing the LlmMarkdown block type enum to QML
+// (LlmMarkdown.Type.*). Blocks themselves are value maps produced by
+// MarkdownParser and stored on LlmSegment.
+class LlmMarkdown : public QObject {
+	Q_OBJECT
+	QML_ELEMENT
+	QML_UNCREATABLE("Blocks are produced by LlmSegment")
+
+	public:
+	enum class Type : int {
+		Text = 0, // Paragraph, list, quote, table; "text" is markdown source
+		Heading, // "level" + "text" (markdown source of the content)
+		Code, // "language" + "code"
+		Math // "latex" (display math, without the $$ delimiters)
+	};
+	Q_ENUM(Type)
+
+	explicit LlmMarkdown(QObject* parent = nullptr) : QObject(parent) {}
+};
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/markdownparser.cpp

@@ -0,0 +1,144 @@
+#include "markdownparser.hpp"
+
+#include "markdownblock.hpp"
+
+#include <cmark-gfm-extension_api.h>
+#include <cmark-gfm.h>
+
+#include <QRegularExpression>
+#include <QStringList>
+#include <QVariantMap>
+
+namespace ZShell::llm {
+
+QVariantList MarkdownParser::parse(const QString& source) {
+	QVariantList blocks;
+	if (source.trimmed().isEmpty())
+		return blocks;
+
+	const QStringList lines = source.split('\n');
+
+	// cmark-gfm line/column numbers are 1-based and inclusive.
+	auto sliceSource = [&](int startLine, int endLine) -> QString {
+		if (startLine < 1 || endLine < startLine)
+			return QString();
+		const int from = startLine;
+		const int to = qMin(endLine, static_cast<int>(lines.size()));
+		return lines.mid(from - 1, to - from + 1).join('\n').trimmed();
+	};
+
+	const QByteArray utf8 = source.toUtf8();
+	cmark_node* doc = cmark_parse_document(
+		utf8.constData(), static_cast<size_t>(utf8.size()),
+		CMARK_OPT_DEFAULT | CMARK_OPT_SOURCEPOS);
+	if (!doc)
+		return blocks;
+
+	// cmark-gfm has no math extension, so $$...$$ parses as an ordinary
+	// paragraph. Re-detect display math (a $$...$$ pair) here and split it
+	// out as its own block. Single-$ inline math is intentionally left
+	// untouched (rendered raw) for now.
+	const QRegularExpression mathRe(
+		QStringLiteral("\\$\\$(.+?)\\$\\$"),
+		QRegularExpression::DotMatchesEverythingOption);
+
+	auto makeBlock = [](LlmMarkdown::Type type) {
+		QVariantMap block;
+		block.insert("type", static_cast<int>(type));
+		return block;
+	};
+
+	auto appendText = [&](const QString& text) {
+		if (text.trimmed().isEmpty())
+			return;
+		QVariantMap block = makeBlock(LlmMarkdown::Type::Text);
+		block.insert("text", text);
+		blocks.append(block);
+	};
+
+	auto appendMath = [&](const QString& latex) {
+		if (latex.trimmed().isEmpty())
+			return;
+		QVariantMap block = makeBlock(LlmMarkdown::Type::Math);
+		block.insert("latex", latex);
+		blocks.append(block);
+	};
+
+	auto appendCode = [&](const QString& language, const QString& code) {
+		QVariantMap block = makeBlock(LlmMarkdown::Type::Code);
+		block.insert("language", language);
+		block.insert("code", code);
+		blocks.append(block);
+	};
+
+	auto appendHeading = [&](int level, const QString& text) {
+		if (text.trimmed().isEmpty())
+			return;
+		QVariantMap block = makeBlock(LlmMarkdown::Type::Heading);
+		block.insert("level", level);
+		block.insert("text", text);
+		blocks.append(block);
+	};
+
+	for (cmark_node* node = cmark_node_first_child(doc); node;
+	     node = cmark_node_next(node)) {
+		const cmark_node_type type = cmark_node_get_type(node);
+		const int startLine = cmark_node_get_start_line(node);
+		const int endLine = cmark_node_get_end_line(node);
+
+		if (type == CMARK_NODE_CODE_BLOCK) {
+			const char* literal = cmark_node_get_literal(node);
+			QString code = literal ? QString::fromUtf8(literal) : QString();
+			// Fenced block literals carry a trailing newline; drop one.
+			if (code.endsWith('\n'))
+				code.chop(1);
+
+			QString language;
+			if (const char* info = cmark_node_get_fence_info(node); info)
+				language = QString::fromUtf8(info).section(' ', 0, 0).trimmed().toLower();
+
+			appendCode(language, code);
+			continue;
+		}
+
+		if (type == CMARK_NODE_HEADING) {
+			// Inline content only (no leading #), so QML can style by
+			// level.
+			const char* content = cmark_node_get_string_content(node);
+			appendHeading(
+				cmark_node_get_heading_level(node),
+				content ? QString::fromUtf8(content).trimmed() : QString());
+			continue;
+		}
+
+		if (type == CMARK_NODE_PARAGRAPH) {
+			const QString text = sliceSource(startLine, endLine);
+			int cursor = 0;
+			bool anyMath = false;
+			for (auto it = mathRe.globalMatch(text, cursor); it.hasNext();
+			     it = mathRe.globalMatch(text, cursor)) {
+				const QRegularExpressionMatch m = it.next();
+				anyMath = true;
+				appendText(
+					text.mid(cursor, static_cast<int>(m.capturedStart() - cursor)));
+				appendMath(m.captured(1).trimmed());
+				cursor = static_cast<int>(m.capturedEnd());
+			}
+			if (!anyMath)
+				appendText(text);
+			else
+				appendText(text.mid(cursor));
+			continue;
+		}
+
+		// Lists, block quotes, tables, horizontal rules, custom blocks:
+		// hand the raw markdown source to QML (rendered via
+		// Text.MarkdownText).
+		appendText(sliceSource(startLine, endLine));
+	}
+
+	cmark_node_free(doc);
+	return blocks;
+}
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/markdownparser.hpp

@@ -0,0 +1,25 @@
+#pragma once
+
+#include <QString>
+#include <QVariantList>
+
+namespace ZShell::llm {
+
+// Splits a markdown document into top-level blocks for QML rendering
+// (cmark-gfm AST walk). Inline structure is not flattened; Text and
+// Heading blocks carry markdown source that QML renders with
+// Text.MarkdownText.
+//
+// Block map keys:
+//   "type"     int (LlmMarkdown::Type)
+//   "level"    int (Heading)
+//   "language" QString (Code, lowercased, empty when unknown)
+//   "code"     QString (Code)
+//   "text"     QString (Text/Heading, markdown source)
+//   "latex"    QString (Math, without the $$ delimiters)
+class MarkdownParser {
+	public:
+	[[nodiscard]] static QVariantList parse(const QString& source);
+};
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/mathtext.cpp

@@ -0,0 +1,176 @@
+#include "mathtext.hpp"
+
+#include "latinmodern-fonts.hpp"
+
+#include <jkqtmathtext/jkqtmathtext.h>
+
+#include <QBuffer>
+#include <QDir>
+#include <QFile>
+#include <QFontDatabase>
+
+namespace ZShell::llm {
+
+namespace {
+// A few px of breathing room around the equation.
+constexpr int kRenderMargin = 2;
+constexpr unsigned int kResolutionDpi = 96;
+
+// Registers the embedded Latin Modern faces with the font database
+// (once per process) and reports which families became available.
+struct LatinModern {
+	bool roman = false;
+	bool math = false;
+};
+
+const LatinModern& loadLatinModern() {
+	static const LatinModern fonts = [] {
+		LatinModern result;
+		// addApplicationFont(QByteArray) fails in this environment, so
+		// the embedded bytes are staged to a per-process temp file and
+		// registered through the (stable) file-based API.
+		const auto add = [](const unsigned char* data, size_t size,
+		                    const QString& fileName, const QString& family) {
+			const QString path = QDir::tempPath() + QLatin1Char('/') + fileName;
+			{
+				QFile f(path);
+				if (!f.open(QIODevice::WriteOnly) ||
+				    f.write(reinterpret_cast<const char*>(data),
+				          static_cast<qint64>(size)) != static_cast<qint64>(size))
+					return false;
+			}
+			const int key = QFontDatabase::addApplicationFont(path);
+			if (key < 0)
+				return false;
+			return QFontDatabase::applicationFontFamilies(key).contains(family);
+		};
+		result.roman =
+			add(lmfont::lmroman10_regular, sizeof(lmfont::lmroman10_regular),
+			    QStringLiteral("lmroman10-regular.otf"), QStringLiteral("LMRoman10"))
+			&& add(lmfont::lmroman10_italic, sizeof(lmfont::lmroman10_italic),
+			       QStringLiteral("lmroman10-italic.otf"), QStringLiteral("LMRoman10"))
+			&& add(lmfont::lmroman10_bold, sizeof(lmfont::lmroman10_bold),
+			       QStringLiteral("lmroman10-bold.otf"), QStringLiteral("LMRoman10"))
+			&& add(lmfont::lmroman10_bolditalic, sizeof(lmfont::lmroman10_bolditalic),
+			       QStringLiteral("lmroman10-bolditalic.otf"),
+			       QStringLiteral("LMRoman10"));
+		result.math = add(
+			lmfont::latinmodern_math, sizeof(lmfont::latinmodern_math),
+			QStringLiteral("latinmodern-math.otf"), QStringLiteral("Latin Modern Math"));
+		return result;
+	}();
+	return fonts;
+}
+
+} // namespace
+
+LlmMathText::LlmMathText(QObject* parent)
+: QObject(parent), m_renderer(parent, /* useFontsForGUI */ true) {
+	// Latin Modern is the default font of modern LaTeX; use the embedded
+	// faces instead of whatever the system happens to have installed.
+	const LatinModern& fonts = loadLatinModern();
+	if (fonts.roman)
+		m_renderer.setFontRomanAndMath(QStringLiteral("LMRoman10"),
+		                               JKQTMathTextFontEncoding::MTFEUnicode);
+	if (fonts.math) {
+		// Same pattern as JKQTMathText's useXITS(): the OpenType math
+		// font supplies the math alphabet and operators from its MATH
+		// table.
+		m_renderer.setFontMathRoman(QStringLiteral("Latin Modern Math"),
+		                            JKQTMathTextFontEncoding::MTFEUnicode);
+		m_renderer.setFallbackFontSymbols(
+			QStringLiteral("Latin Modern Math"),
+			JKQTMathTextFontEncoding::MTFEUnicode);
+	}
+}
+
+void LlmMathText::setLatex(const QString& value) {
+	if (m_latex == value)
+		return;
+	m_latex = value;
+	reRender();
+}
+
+void LlmMathText::setColor(const QColor& value) {
+	if (m_color == value)
+		return;
+	m_color = value;
+	reRender();
+}
+
+void LlmMathText::setFontPointSize(double value) {
+	if (qFuzzyCompare(m_fontPointSize, value))
+		return;
+	m_fontPointSize = value;
+	reRender();
+}
+
+void LlmMathText::setDevicePixelRatio(qreal value) {
+	if (qFuzzyCompare(m_devicePixelRatio, value))
+		return;
+	m_devicePixelRatio = value;
+	reRender();
+}
+
+void LlmMathText::reRender() {
+	if (m_latex.trimmed().isEmpty()) {
+		m_image = QImage();
+		m_imageUrl = QUrl();
+		m_width = 0;
+		m_height = 0;
+		m_ok = false;
+		Q_EMIT changed();
+		return;
+	}
+
+	m_renderer.setFontPointSize(m_fontPointSize);
+	m_renderer.setFontColor(m_color);
+
+	const bool ok = m_renderer.parse(
+		m_latex,
+		JKQTMathText::LatexParser,
+		JKQTMathText::DefaultParseOptions);
+	if (!ok) {
+		m_image = QImage();
+		m_imageUrl = QUrl();
+		m_width = 0;
+		m_height = 0;
+		m_ok = false;
+		Q_EMIT changed();
+		return;
+	}
+
+	const QImage image = m_renderer.drawIntoImage(
+		/* drawBoxes */ false,
+		QColor(Qt::transparent),
+		kRenderMargin,
+		m_devicePixelRatio,
+		kResolutionDpi);
+	if (image.isNull()) {
+		m_image = QImage();
+		m_imageUrl = QUrl();
+		m_width = 0;
+		m_height = 0;
+		m_ok = false;
+		Q_EMIT changed();
+		return;
+	}
+
+	m_image = image;
+	QByteArray png;
+	{
+		QBuffer buffer(&png);
+		buffer.open(QIODevice::WriteOnly);
+		image.save(&buffer, "PNG");
+	}
+	m_imageUrl = QUrl(
+		QStringLiteral("data:image/png;base64,") + QString::fromLatin1(png.toBase64()));
+	// drawIntoImage renders at devicePixelRatio; convert back to
+	// logical pixels.
+	m_width = image.width() / m_devicePixelRatio;
+	m_height = image.height() / m_devicePixelRatio;
+	m_ok = true;
+	Q_EMIT changed();
+}
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/mathtext.hpp

@@ -0,0 +1,72 @@
+#pragma once
+
+#include <QColor>
+#include <QImage>
+#include <QObject>
+#include <QString>
+#include <QUrl>
+#include <QtQml>
+
+#include <jkqtmathtext/jkqtmathtext.h>
+
+namespace ZShell::llm {
+
+// QML wrapper around JKQTMathText (JKQtPlotter's LaTeX renderer).
+//
+// Parses a display-math string and renders it into a transparent
+// QImage at the given device pixel ratio. QML displays the image
+// (scaling it to the bubble width when needed) and falls back to the
+// raw LaTeX when parsing fails.
+class LlmMathText : public QObject {
+	Q_OBJECT
+	QML_ELEMENT
+
+	Q_PROPERTY(QString latex READ latex WRITE setLatex NOTIFY changed)
+	Q_PROPERTY(QColor color READ color WRITE setColor NOTIFY changed)
+	Q_PROPERTY(double fontPointSize READ fontPointSize WRITE setFontPointSize NOTIFY changed)
+	Q_PROPERTY(qreal devicePixelRatio READ devicePixelRatio WRITE setDevicePixelRatio NOTIFY changed)
+	Q_PROPERTY(QImage image READ image NOTIFY changed)
+	// data: URL of the rendered equation; usable directly as
+	// Image.source (a raw QImage is not).
+	Q_PROPERTY(QUrl imageUrl READ imageUrl NOTIFY changed)
+	// Logical (CSS pixel) size of the rendered equation.
+	Q_PROPERTY(qreal width READ width NOTIFY changed)
+	Q_PROPERTY(qreal height READ height NOTIFY changed)
+	Q_PROPERTY(bool ok READ ok NOTIFY changed)
+
+	public:
+	explicit LlmMathText(QObject* parent = nullptr);
+
+	[[nodiscard]] QString latex() const { return m_latex; }
+	void setLatex(const QString& value);
+	[[nodiscard]] QColor color() const { return m_color; }
+	void setColor(const QColor& value);
+	[[nodiscard]] double fontPointSize() const { return m_fontPointSize; }
+	void setFontPointSize(double value);
+	[[nodiscard]] qreal devicePixelRatio() const { return m_devicePixelRatio; }
+	void setDevicePixelRatio(qreal value);
+	[[nodiscard]] QImage image() const { return m_image; }
+	[[nodiscard]] QUrl imageUrl() const { return m_imageUrl; }
+	[[nodiscard]] qreal width() const { return m_width; }
+	[[nodiscard]] qreal height() const { return m_height; }
+	[[nodiscard]] bool ok() const { return m_ok; }
+
+	Q_SIGNALS:
+	void changed();
+
+	private:
+	void reRender();
+
+	JKQTMathText m_renderer;
+	QString m_latex;
+	QColor m_color;
+	double m_fontPointSize = 12.0;
+	qreal m_devicePixelRatio = 1.0;
+	QImage m_image;
+	QUrl m_imageUrl;
+	qreal m_width = 0;
+	qreal m_height = 0;
+	bool m_ok = false;
+};
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/message.cpp

@@ -18,16 +18,8 @@ ChatSession* sessionOf(const ChatMessage* message) {
 ChatMessage::ChatMessage(Role role, qint64 timestamp, QObject* parent)
 : QObject(parent), m_role(role), m_timestamp(timestamp) {}
 
-ChatGeneration* ChatMessage::addGeneration(
-	qint64 timestamp,
-	const QString& content,
-	const QString& reasoning,
-	qint64 reasoningElapsedMs,
-	qint64 contentElapsedMs) {
+ChatGeneration* ChatMessage::addGeneration(qint64 timestamp) {
 	auto* generation = new ChatGeneration(timestamp, this);
-	generation->setContent(content);
-	generation->setReasoning(reasoning);
-	generation->setElapsedMs(reasoningElapsedMs, contentElapsedMs);
 	m_generations.append(generation);
 	if (m_active < 0)
 		m_active = static_cast<int>(m_generations.size() - 1);
@@ -36,8 +28,7 @@ ChatGeneration* ChatMessage::addGeneration(
 }
 
 ChatGeneration* ChatMessage::appendGeneration(qint64 timestamp) {
-	auto* generation =
-		addGeneration(timestamp, QString(), QString(), 0, 0);
+	auto* generation = addGeneration(timestamp);
 	setActiveInternal(static_cast<int>(m_generations.size() - 1));
 	return generation;
 }

Plugins/ZShell/Llm/message.hpp

@@ -58,12 +58,8 @@ class ChatMessage : public QObject {
 	Q_INVOKABLE void retry();
 	Q_INVOKABLE void generate();
 
-	ChatGeneration* addGeneration(
-		qint64 timestamp,
-		const QString& content,
-		const QString& reasoning,
-		qint64 reasoningElapsedMs,
-		qint64 contentElapsedMs);
+	// Creates an empty generation; callers fill it with segments.
+	ChatGeneration* addGeneration(qint64 timestamp);
 	ChatGeneration* appendGeneration(qint64 timestamp);
 	void removeGeneration(ChatGeneration* generation);
 

Plugins/ZShell/Llm/messagemodel.cpp

@@ -44,7 +44,8 @@ ChatMessage* ChatMessageModel::createMessage(
 ChatMessage* ChatMessageModel::appendNewest(
 	ChatMessage::Role role, const QString& content, qint64 timestamp) {
 	auto* message = new ChatMessage(role, timestamp, this);
-	message->addGeneration(timestamp, content, QString(), 0, 0);
+	auto* generation = message->addGeneration(timestamp);
+	generation->setContent(content);
 
 	beginInsertRows(QModelIndex(), 0, 0);
 	m_messages.prepend(message);

Plugins/ZShell/Llm/segment.cpp

@@ -0,0 +1,138 @@
+#include "segment.hpp"
+
+#include "markdownparser.hpp"
+
+#include <QDateTime>
+
+namespace ZShell::llm {
+
+namespace {
+// Re-parse cadence while a segment streams; content between refreshes is
+// at most this stale.
+constexpr int kMarkdownRefreshMs = 150;
+} // namespace
+
+LlmSegment::LlmSegment(Type type, qint64 timestamp, QObject* parent)
+: QObject(parent), m_type(type), m_timestamp(timestamp) {
+	m_markdownTimer.setInterval(kMarkdownRefreshMs);
+	connect(
+		&m_markdownTimer, &QTimer::timeout, this, [this]() {
+			if (!m_markdownDirty)
+				return;
+			m_markdownDirty = false;
+			parseMarkdown();
+		});
+}
+
+qint64 LlmSegment::elapsedMs() const {
+	if (m_startedAt <= 0)
+		return 0;
+	const qint64 end = m_endedAt > 0 ? m_endedAt
+	                                  : QDateTime::currentMSecsSinceEpoch();
+	return end - m_startedAt;
+}
+
+void LlmSegment::begin() {
+	if (m_running)
+		return;
+	m_running = true;
+	m_startedAt = QDateTime::currentMSecsSinceEpoch();
+	m_endedAt = 0;
+	Q_EMIT runningChanged();
+	Q_EMIT elapsedMsChanged();
+}
+
+void LlmSegment::close() {
+	if (m_running) {
+		m_running = false;
+		m_endedAt = QDateTime::currentMSecsSinceEpoch();
+		Q_EMIT runningChanged();
+		Q_EMIT elapsedMsChanged();
+	}
+	// The segment is final; parse immediately so the UI does not wait
+	// out the debounce.
+	if (m_type == Type::Content && m_markdownDirty) {
+		m_markdownTimer.stop();
+		parseMarkdown();
+	}
+}
+
+void LlmSegment::appendText(const QString& piece) {
+	if (piece.isEmpty())
+		return;
+	m_text += piece;
+	Q_EMIT textChanged();
+	scheduleMarkdown();
+}
+
+void LlmSegment::setText(const QString& value) {
+	if (m_text == value)
+		return;
+	m_text = value;
+	Q_EMIT textChanged();
+	scheduleMarkdown();
+}
+
+void LlmSegment::setName(const QString& value) {
+	if (m_name == value)
+		return;
+	m_name = value;
+	Q_EMIT nameChanged();
+}
+
+void LlmSegment::setToolCallId(const QString& value) {
+	if (m_toolCallId == value)
+		return;
+	m_toolCallId = value;
+	Q_EMIT toolCallIdChanged();
+}
+
+void LlmSegment::appendArguments(const QString& piece) {
+	if (piece.isEmpty())
+		return;
+	m_arguments += piece;
+	Q_EMIT argumentsChanged();
+}
+
+void LlmSegment::setResult(const QString& value) {
+	if (m_result == value)
+		return;
+	m_result = value;
+	Q_EMIT resultChanged();
+}
+
+void LlmSegment::setStatus(Status value) {
+	if (m_status == value)
+		return;
+	m_status = value;
+	Q_EMIT statusChanged();
+}
+
+void LlmSegment::finishTool(const QString& resultText, bool success) {
+	setResult(resultText);
+	setStatus(success ? Status::Success : Status::Error);
+	close();
+}
+
+void LlmSegment::restore(qint64 elapsedMs) {
+	m_startedAt = m_timestamp;
+	m_endedAt = m_timestamp + qMax<qint64>(0, elapsedMs);
+}
+
+void LlmSegment::scheduleMarkdown() {
+	// Content segments only; reasoning/tool output is never parsed.
+	// (User content is parsed too, so it can later be rendered as blocks
+	// as well; the QML currently only does that for assistant messages.)
+	if (m_type != Type::Content)
+		return;
+	m_markdownDirty = true;
+	if (!m_markdownTimer.isActive())
+		m_markdownTimer.start();
+}
+
+void LlmSegment::parseMarkdown() {
+	m_markdown = MarkdownParser::parse(m_text);
+	Q_EMIT markdownChanged();
+}
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/segment.hpp

@@ -0,0 +1,113 @@
+#pragma once
+
+#include <QObject>
+#include <QTimer>
+#include <QString>
+#include <QVariantList>
+#include <QtQml>
+
+namespace ZShell::llm {
+
+// One unit of activity within a ChatGeneration. A generation keeps its
+// segments in chronological order: zero or more reasoning bursts and
+// tool calls interleaved, plus at most one content segment holding the
+// final answer.
+class LlmSegment : public QObject {
+	Q_OBJECT
+	QML_ELEMENT
+	QML_UNCREATABLE("Segments are managed by ChatGeneration")
+
+	Q_PROPERTY(Type type READ type CONSTANT)
+	Q_PROPERTY(qint64 timestamp READ timestamp CONSTANT)
+	Q_PROPERTY(QString text READ text NOTIFY textChanged)
+	Q_PROPERTY(QString name READ name NOTIFY nameChanged)
+	Q_PROPERTY(QString toolCallId READ toolCallId NOTIFY toolCallIdChanged)
+	Q_PROPERTY(QString arguments READ arguments NOTIFY argumentsChanged)
+	Q_PROPERTY(QString result READ result NOTIFY resultChanged)
+	Q_PROPERTY(Status status READ status NOTIFY statusChanged)
+	Q_PROPERTY(bool running READ running NOTIFY runningChanged)
+	Q_PROPERTY(qint64 elapsedMs READ elapsedMs NOTIFY elapsedMsChanged)
+	// Parsed markdown blocks (QVariantList of maps, see MarkdownParser).
+	// Only Content segments are parsed. Refreshes are debounced so a
+	// streaming segment re-parses at a steady cadence rather than per
+	// chunk.
+	Q_PROPERTY(QVariantList markdown READ markdown NOTIFY markdownChanged)
+
+	public:
+	enum class Type : int {
+		Reasoning = 0,
+		ToolCall,
+		Content
+	};
+	Q_ENUM(Type)
+
+	enum class Status : int {
+		None = 0,
+		Running,
+		Success,
+		Error
+	};
+	Q_ENUM(Status)
+
+	explicit LlmSegment(Type type, qint64 timestamp, QObject* parent = nullptr);
+
+	[[nodiscard]] Type type() const { return m_type; }
+	[[nodiscard]] qint64 timestamp() const { return m_timestamp; }
+	[[nodiscard]] QString text() const { return m_text; }
+	[[nodiscard]] QString name() const { return m_name; }
+	[[nodiscard]] QString toolCallId() const { return m_toolCallId; }
+	[[nodiscard]] QString arguments() const { return m_arguments; }
+	[[nodiscard]] QString result() const { return m_result; }
+	[[nodiscard]] Status status() const { return m_status; }
+	[[nodiscard]] bool running() const { return m_running; }
+	[[nodiscard]] qint64 elapsedMs() const;
+	[[nodiscard]] QVariantList markdown() const { return m_markdown; }
+
+	// Starts the segment's clock; a no-op while already running.
+	void begin();
+	// Stops the segment's clock; a no-op when not running.
+	void close();
+	void appendText(const QString& piece);
+	void setText(const QString& value);
+	void setName(const QString& value);
+	void setToolCallId(const QString& value);
+	void appendArguments(const QString& piece);
+	void setResult(const QString& value);
+	void setStatus(Status value);
+	// Completes a tool call with the model-facing result text.
+	void finishTool(const QString& resultText, bool success);
+	// Restores persisted timing without a live clock.
+	void restore(qint64 elapsedMs);
+
+	Q_SIGNALS:
+	void textChanged();
+	void markdownChanged();
+	void nameChanged();
+	void toolCallIdChanged();
+	void argumentsChanged();
+	void resultChanged();
+	void statusChanged();
+	void runningChanged();
+	void elapsedMsChanged();
+
+	private:
+	void scheduleMarkdown();
+	void parseMarkdown();
+
+	Type m_type;
+	qint64 m_timestamp;
+	QString m_text;
+	QString m_name;
+	QString m_toolCallId;
+	QString m_arguments;
+	QString m_result;
+	Status m_status = Status::None;
+	bool m_running = false;
+	qint64 m_startedAt = 0;
+	qint64 m_endedAt = 0;
+	QVariantList m_markdown;
+	QTimer m_markdownTimer;
+	bool m_markdownDirty = false;
+};
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/tool.cpp

@@ -0,0 +1,59 @@
+#include "tool.hpp"
+
+namespace ZShell::llm {
+
+LlmTool::LlmTool(QObject* parent) : QObject(parent) {}
+
+LlmTool::~LlmTool() = default;
+
+void LlmTool::cancel() {}
+
+QJsonObject LlmTool::specification() const {
+	QJsonObject function;
+	function[QStringLiteral("name")] = name();
+	function[QStringLiteral("description")] = description();
+	function[QStringLiteral("parameters")] = parameters();
+	QJsonObject spec;
+	spec[QStringLiteral("type")] = QStringLiteral("function");
+	spec[QStringLiteral("function")] = function;
+	return spec;
+}
+
+ToolRegistry::ToolRegistry(QObject* parent) : QObject(parent) {}
+
+void ToolRegistry::setEnabled(bool value) {
+	if (m_enabled == value)
+		return;
+	m_enabled = value;
+	Q_EMIT enabledChanged();
+}
+
+void ToolRegistry::registerTool(LlmTool* tool) {
+	if (!tool || m_tools.contains(tool))
+		return;
+	tool->setParent(this);
+	m_tools.append(tool);
+}
+
+LlmTool* ToolRegistry::tool(const QString& name) const {
+	for (const auto* tool : m_tools)
+		if (tool->name() == name)
+			return const_cast<LlmTool*>(tool);
+	return nullptr;
+}
+
+QJsonArray ToolRegistry::specifications() const {
+	if (!m_enabled)
+		return {};
+	QJsonArray specs;
+	for (const auto* tool : m_tools)
+		specs.append(tool->specification());
+	return specs;
+}
+
+void ToolRegistry::cancelAll() {
+	for (auto* tool : m_tools)
+		tool->cancel();
+}
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/tool.hpp

@@ -0,0 +1,70 @@
+#pragma once
+
+#include <QJsonArray>
+#include <QJsonObject>
+#include <QList>
+#include <QObject>
+#include <QString>
+
+#include <functional>
+
+namespace ZShell::llm {
+
+// A capability the model may invoke mid-turn. Tools run asynchronously
+// and report exactly one result: `{"output": ...}` on success or
+// `{"error": ...}` on failure.
+class LlmTool : public QObject {
+	Q_OBJECT
+
+	public:
+	explicit LlmTool(QObject* parent = nullptr);
+	~LlmTool() override;
+
+	[[nodiscard]] virtual QString name() const = 0;
+	[[nodiscard]] virtual QString description() const = 0;
+	// JSON Schema describing the tool's `arguments` object.
+	[[nodiscard]] virtual QJsonObject parameters() const = 0;
+
+	// Runs the tool; `done` is invoked exactly once, with
+	// `{"output": ...}` on success or `{"error": ...}` on failure.
+	// `done` must be invoked asynchronously (on a later event loop
+	// iteration), never synchronously within execute().
+	virtual void execute(
+		const QJsonObject& args,
+		std::function<void(const QJsonObject& result)> done) = 0;
+	// Abandons in-flight work, if any.
+	virtual void cancel();
+
+	// The OpenAI-compatible `tools` entry for this tool.
+	[[nodiscard]] QJsonObject specification() const;
+};
+
+// Owns the set of tools available to the model.
+class ToolRegistry : public QObject {
+	Q_OBJECT
+
+	Q_PROPERTY(bool enabled READ enabled WRITE setEnabled NOTIFY enabledChanged)
+
+	public:
+	explicit ToolRegistry(QObject* parent = nullptr);
+
+	[[nodiscard]] bool enabled() const { return m_enabled; }
+	void setEnabled(bool value);
+
+	// Takes ownership; tools become children of the registry.
+	void registerTool(LlmTool* tool);
+	[[nodiscard]] LlmTool* tool(const QString& name) const;
+	// The request body's `tools` array; empty while disabled.
+	[[nodiscard]] QJsonArray specifications() const;
+	// Abandons in-flight work in every tool.
+	void cancelAll();
+
+	Q_SIGNALS:
+	void enabledChanged();
+
+	private:
+	bool m_enabled = true;
+	QList<LlmTool*> m_tools;
+};
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/webfetchtool.cpp

@@ -0,0 +1,428 @@
+#include "webfetchtool.hpp"
+
+#include <QJsonArray>
+#include <QNetworkReply>
+#include <QNetworkRequest>
+#include <QSet>
+#include <QTimer>
+
+namespace ZShell::llm {
+
+namespace {
+
+const char* kUserAgent =
+	"Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
+	"(KHTML, like Gecko) Chrome/143.0.0.0 Safari/537.36";
+
+QJsonObject makeOutput(const QString& text) {
+	QJsonObject obj;
+	obj[QStringLiteral("output")] = text;
+	return obj;
+}
+
+QJsonObject makeError(const QString& message) {
+	QJsonObject obj;
+	obj[QStringLiteral("error")] = message;
+	return obj;
+}
+
+} // namespace
+
+WebFetchTool::WebFetchTool(QObject* parent) : LlmTool(parent) {}
+
+WebFetchTool::~WebFetchTool() {
+	for (auto* job : m_jobs) {
+		if (job->timer)
+			job->timer->stop();
+		if (job->reply)
+			job->reply->abort();
+	}
+	// Pending result callbacks are dropped; the client is going away.
+	qDeleteAll(m_jobs);
+}
+
+QString WebFetchTool::name() const {
+	return QStringLiteral("webfetch");
+}
+
+QString WebFetchTool::description() const {
+	return QStringLiteral(
+		"Fetch content from an HTTP or HTTPS URL and return it as plain "
+		"text or raw HTML. HTML pages are reduced to their visible text "
+		"by default. This tool is read-only.");
+}
+
+QJsonObject WebFetchTool::parameters() const {
+	QJsonObject url;
+	url[QStringLiteral("type")] = QStringLiteral("string");
+	url[QStringLiteral("description")] =
+		QStringLiteral("The HTTP or HTTPS URL to fetch content from");
+
+	QJsonArray formats;
+	formats.append(QStringLiteral("text"));
+	formats.append(QStringLiteral("html"));
+	QJsonObject format;
+	format[QStringLiteral("type")] = QStringLiteral("string");
+	format[QStringLiteral("enum")] = formats;
+	format[QStringLiteral("description")] =
+		QStringLiteral("The format to return the content in. Defaults to "
+		               "text.");
+
+	QJsonObject timeout;
+	timeout[QStringLiteral("type")] = QStringLiteral("integer");
+	timeout[QStringLiteral("minimum")] = 1;
+	timeout[QStringLiteral("maximum")] = MaxTimeoutSeconds;
+	timeout[QStringLiteral("description")] =
+		QStringLiteral("Optional timeout in seconds");
+
+	QJsonObject properties;
+	properties[QStringLiteral("url")] = url;
+	properties[QStringLiteral("format")] = format;
+	properties[QStringLiteral("timeout")] = timeout;
+
+	QJsonObject schema;
+	schema[QStringLiteral("type")] = QStringLiteral("object");
+	schema[QStringLiteral("properties")] = properties;
+	QJsonArray required;
+	required.append(QStringLiteral("url"));
+	schema[QStringLiteral("required")] = required;
+	return schema;
+}
+
+void WebFetchTool::completeJob(Job* job, QJsonObject result) {
+	// Deliver on a later event loop iteration; LlmClient relies on tool
+	// results never arriving synchronously within execute().
+	QMetaObject::invokeMethod(
+		this,
+		[this, job, result = std::move(result)]() mutable {
+			if (!m_jobs.contains(job))
+				return;
+			auto done = std::move(job->done);
+			m_jobs.removeAll(job);
+			job->timer->deleteLater();
+			if (job->reply)
+				job->reply->deleteLater();
+			delete job;
+			done(std::move(result));
+		},
+		Qt::QueuedConnection);
+}
+
+void WebFetchTool::execute(
+	const QJsonObject& args, std::function<void(const QJsonObject&)> done) {
+	auto* job = new Job;
+	job->done = std::move(done);
+	m_jobs.append(job);
+
+	auto fail = [this, job](const QString& message) {
+		job->timer->stop();
+		completeJob(job, makeError(message));
+	};
+
+	const QString urlText = args[QStringLiteral("url")].toString().trimmed();
+	const QUrl url = QUrl::fromUserInput(urlText);
+	if (!url.isValid() || url.host().isEmpty()) {
+		fail(QStringLiteral("Invalid URL: %1").arg(urlText));
+		return;
+	}
+	if (url.scheme() != QLatin1String("http") &&
+	    url.scheme() != QLatin1String("https")) {
+		fail(QStringLiteral("URL must use http:// or https://"));
+		return;
+	}
+
+	job->format =
+		args[QStringLiteral("format")].toString(QStringLiteral("text"));
+	if (job->format != QLatin1String("html"))
+		job->format = QStringLiteral("text");
+
+	const int timeoutMs = qBound(
+	                             1,
+	                             args[QStringLiteral("timeout")].toInt(
+		                             DefaultTimeoutSeconds),
+	                             MaxTimeoutSeconds) *
+	                     1000;
+
+	job->timer = new QTimer(this);
+	job->timer->setSingleShot(true);
+	connect(
+		job->timer, &QTimer::timeout, this, [this, job]() {
+			if (job->reply)
+				job->reply->abort();
+		});
+	job->timer->start(timeoutMs);
+
+	QNetworkRequest request(url);
+	request.setRawHeader("User-Agent", QByteArray(kUserAgent));
+	request.setRawHeader(
+		"Accept",
+		job->format == QLatin1String("html")
+			? "text/html;q=1.0, application/xhtml+xml;q=0.9, */*;q=0.1"
+			: "text/plain;q=1.0, text/markdown;q=0.9, text/html;q=0.8, "
+			      "*/*;q=0.1");
+	request.setRawHeader("Accept-Language", "en-US,en;q=0.9");
+
+	job->reply = m_manager.get(request);
+	connect(job->reply, &QNetworkReply::readyRead, this, [this, job]() {
+		if (!job->reply)
+			return;
+		job->body += job->reply->readAll();
+		if (job->body.size() > MaxResponseBytes) {
+			job->tooLarge = true;
+			job->reply->abort();
+		}
+	});
+	connect(job->reply, &QNetworkReply::finished, this, [this, job]() {
+		QNetworkReply* reply = job->reply;
+		job->reply = nullptr;
+		job->timer->stop();
+		if (!reply)
+			return;
+
+		const QNetworkReply::NetworkError error = reply->error();
+		const QString errorString = reply->errorString();
+		QByteArray body = job->body;
+		body += reply->readAll();
+		job->body.clear();
+		const QByteArray contentType =
+			reply->rawHeader("Content-Type").toLower();
+		const int status =
+			reply->attribute(QNetworkRequest::HttpStatusCodeAttribute)
+				.toInt();
+
+		if (job->tooLarge) {
+			completeJob(job, makeError(
+				QStringLiteral("Response too large (exceeds the 5 MB "
+				               "limit")));
+			return;
+		}
+		if (error != QNetworkReply::NoError) {
+			completeJob(
+				job,
+				makeError(
+					QStringLiteral("Request failed: %1").arg(errorString)));
+			return;
+		}
+		if (status < 200 || status >= 300) {
+			completeJob(
+				job,
+				makeError(
+					QStringLiteral("Server returned status %1")
+						.arg(status)));
+			return;
+		}
+
+		const QString mime =
+			QString::fromLatin1(contentType).section(QLatin1Char(';'), 0, 0)
+				.trimmed();
+		if (mime.startsWith(QLatin1String("image/"))) {
+			completeJob(
+				job,
+				makeError(
+					QStringLiteral(
+						"Unsupported fetched image content type: %1")
+						.arg(mime)));
+			return;
+		}
+		const bool textual = mime.isEmpty() ||
+			mime.startsWith(QLatin1String("text/")) ||
+			mime == QLatin1String("application/json") ||
+			mime.endsWith(QLatin1String("+json")) ||
+			mime == QLatin1String("application/xml") ||
+			mime.endsWith(QLatin1String("+xml")) ||
+			mime.startsWith(QLatin1String("application/javascript")) ||
+			mime.startsWith(QLatin1String("text/javascript"));
+		if (!textual) {
+			completeJob(
+				job,
+				makeError(
+					QStringLiteral(
+						"Unsupported fetched file content type: %1")
+						.arg(mime)));
+			return;
+		}
+
+		QString content = QString::fromUtf8(body);
+		if (mime.contains(QLatin1String("text/html")) &&
+		    job->format == QLatin1String("text"))
+			content = extractTextFromHtml(content);
+		if (content.size() > MaxOutputChars)
+			content = content.left(MaxOutputChars) +
+				QStringLiteral("\n[... truncated ...]");
+		completeJob(job, makeOutput(content));
+	});
+}
+
+void WebFetchTool::cancel() {
+	for (auto* job : m_jobs) {
+		job->timer->stop();
+		if (job->reply)
+			job->reply->abort();
+	}
+}
+
+QString WebFetchTool::decodeEntities(const QString& text) {
+	if (!text.contains(QLatin1Char('&')))
+		return text;
+	QString out;
+	out.reserve(text.size());
+	for (qsizetype i = 0; i < text.size(); ++i) {
+		if (text.at(i) != QLatin1Char('&')) {
+			out += text.at(i);
+			continue;
+		}
+		const qsizetype semi = text.indexOf(QLatin1Char(';'), i);
+		if (semi < 0 || semi - i > 12) {
+			out += QLatin1Char('&');
+			continue;
+		}
+		const QString entity = text.mid(i + 1, semi - i - 1);
+		QString replacement;
+		if (entity == QLatin1String("amp"))
+			replacement = QLatin1Char('&');
+		else if (entity == QLatin1String("lt"))
+			replacement = QLatin1Char('<');
+		else if (entity == QLatin1String("gt"))
+			replacement = QLatin1Char('>');
+		else if (entity == QLatin1String("quot"))
+			replacement = QLatin1Char('"');
+		else if (entity == QLatin1String("apos"))
+			replacement = QLatin1Char('\'');
+		else if (entity == QLatin1String("nbsp"))
+			replacement = QLatin1Char(' ');
+		else {
+			bool ok = false;
+			const quint32 codePoint = entity.startsWith(QLatin1String("#x")) ||
+			                              entity.startsWith(QLatin1String("#X"))
+				? entity.mid(2).toUInt(&ok, 16)
+				: entity.toUInt(&ok);
+			if (ok && codePoint != 0) {
+				const char32_t ucs4[2] = {
+					static_cast<char32_t>(codePoint),
+					0,
+				};
+				replacement = QString::fromUcs4(ucs4);
+			}
+		}
+		if (replacement.isEmpty()) {
+			out += QLatin1Char('&');
+			continue;
+		}
+		out += replacement;
+		i = semi;
+	}
+	return out;
+}
+
+QString WebFetchTool::extractTextFromHtml(const QString& html) {
+	static const QSet<QString> kSkipTags = {
+		QStringLiteral("noscript"),
+		QStringLiteral("iframe"),
+		QStringLiteral("object"),
+		QStringLiteral("head"),
+	};
+	static const QSet<QString> kRawTags = {
+		QStringLiteral("script"),
+		QStringLiteral("style"),
+	};
+	static const QSet<QString> kVoidTags = {
+		QStringLiteral("area"),   QStringLiteral("base"),
+		QStringLiteral("br"),     QStringLiteral("col"),
+		QStringLiteral("embed"),  QStringLiteral("hr"),
+		QStringLiteral("img"),    QStringLiteral("input"),
+		QStringLiteral("link"),   QStringLiteral("meta"),
+		QStringLiteral("source"), QStringLiteral("track"),
+		QStringLiteral("wbr"),
+	};
+
+	QString text;
+	text.reserve(html.size() / 2);
+	int skipDepth = 0;
+	qsizetype i = 0;
+	while (i < html.size()) {
+		const qsizetype open = html.indexOf(QLatin1Char('<'), i);
+		if (open < 0) {
+			if (skipDepth == 0)
+				text += html.mid(i);
+			break;
+		}
+		if (skipDepth == 0)
+			text += html.mid(i, open - i);
+		const qsizetype close = html.indexOf(QLatin1Char('>'), open);
+		if (close < 0)
+			break;
+		const QString tag =
+			html.mid(open + 1, close - open - 1).trimmed().toLower();
+		i = close + 1;
+
+		if (tag.startsWith(QLatin1Char('!')) ||
+		    tag.startsWith(QLatin1Char('?')))
+			continue;
+
+		QString name = tag;
+		if (name.startsWith(QLatin1Char('/'))) {
+			if (skipDepth > 0)
+				--skipDepth;
+			continue;
+		}
+		qsizetype j = 0;
+		while (j < name.size() &&
+		       (name.at(j).isLetterOrNumber() ||
+		        name.at(j) == QLatin1Char(':') ||
+		        name.at(j) == QLatin1Char('-')))
+			++j;
+		name = name.left(j);
+
+		if (kRawTags.contains(name)) {
+			// Raw-text element: swallow everything up to its close tag.
+			const qsizetype rawEnd =
+				html.indexOf(QStringLiteral("</") + name, i,
+				           Qt::CaseInsensitive);
+			if (rawEnd < 0)
+				break;
+			const qsizetype rawClose = html.indexOf(QLatin1Char('>'), rawEnd);
+			if (rawClose < 0)
+				break;
+			i = rawClose + 1;
+			continue;
+		}
+		if (kVoidTags.contains(name))
+			continue;
+		if (skipDepth > 0) {
+			// Browsers implicitly close <head> at <body>; malformed pages
+			// without a </head> would otherwise swallow the whole page.
+			if (name == QLatin1String("body")) {
+				skipDepth = 0;
+				continue;
+			}
+			++skipDepth;
+			continue;
+		}
+		if (kSkipTags.contains(name)) {
+			++skipDepth;
+			continue;
+		}
+		// Normal tag: replace with a space so words do not merge.
+		text += QLatin1Char(' ');
+	}
+
+	QString out = decodeEntities(text);
+	QStringList lines;
+	for (const QString& line : out.split(QLatin1Char('\n'))) {
+		const QString flat = line.simplified();
+		if (flat.isEmpty()) {
+			if (!lines.isEmpty() && lines.last().isEmpty())
+				continue;
+			lines.append(QString());
+		} else {
+			lines.append(flat);
+		}
+	}
+	while (lines.size() > 1 && lines.first().isEmpty())
+		lines.removeFirst();
+	while (lines.size() > 1 && lines.last().isEmpty())
+		lines.removeLast();
+	return lines.join(QLatin1Char('\n'));
+}
+
+} // namespace ZShell::llm

Plugins/ZShell/Llm/webfetchtool.hpp

@@ -0,0 +1,64 @@
+#pragma once
+
+#include "tool.hpp"
+
+#include <QByteArray>
+#include <QList>
+#include <QNetworkAccessManager>
+
+#include <functional>
+
+class QNetworkReply;
+class QTimer;
+
+namespace ZShell::llm {
+
+// Fetches an http(s) URL and returns its content as plain text or raw
+// HTML. Read-only. Mirrors opencode's webfetch tool, without markdown
+// conversion and the permission prompt. Concurrent fetches are
+// supported; results are always delivered on a later event loop
+// iteration, never synchronously from execute().
+class WebFetchTool : public LlmTool {
+	Q_OBJECT
+
+	public:
+	static constexpr int MaxResponseBytes = 5 * 1024 * 1024;
+	static constexpr int DefaultTimeoutSeconds = 30;
+	static constexpr int MaxTimeoutSeconds = 120;
+	// Caps the characters handed to the model so a large page cannot
+	// blow out the context.
+	static constexpr int MaxOutputChars = 64 * 1024;
+
+	explicit WebFetchTool(QObject* parent = nullptr);
+	~WebFetchTool() override;
+
+	QString name() const override;
+	QString description() const override;
+	QJsonObject parameters() const override;
+	void execute(
+		const QJsonObject& args,
+		std::function<void(const QJsonObject& result)> done) override;
+	void cancel() override;
+
+	// Strips tags (skipping script/style/noscript/iframe/object/embed/
+	// head) and decodes common entities.
+	static QString extractTextFromHtml(const QString& html);
+	static QString decodeEntities(const QString& text);
+
+	private:
+	struct Job {
+		QNetworkReply* reply = nullptr;
+		QTimer* timer = nullptr;
+		QByteArray body;
+		bool tooLarge = false;
+		QString format;
+		std::function<void(const QJsonObject& result)> done;
+	};
+
+	void completeJob(Job* job, QJsonObject result);
+
+	QNetworkAccessManager m_manager;
+	QList<Job*> m_jobs;
+};
+
+} // namespace ZShell::llm