fix(webui): preserve activity duration on replay

webui/src/components/thread/AgentActivityCluster.tsx

@@ -173,6 +173,8 @@ interface AgentActivityClusterProps {
   /** True while the session turn is still running (drives “Working…” copy + header sheen). */
   isTurnStreaming: boolean;
   hasBodyBelow: boolean;
+  /** Persisted end-to-end turn latency from the assistant answer, used for history replay. */
+  turnLatencyMs?: number;
   cliApps?: CliAppInfo[];
   mcpPresets?: McpPresetInfo[];
 }
@@ -185,6 +187,7 @@ export function AgentActivityCluster({
   messages,
   isTurnStreaming,
   hasBodyBelow,
+  turnLatencyMs,
   cliApps = [],
   mcpPresets = [],
 }: AgentActivityClusterProps) {
@@ -242,12 +245,15 @@ export function AgentActivityCluster({
   const singleFilePath = fileCount === 1 ? primaryFilePath : undefined;
   const singleFileTooltipPath = fileCount === 1 ? primaryFileTooltipPath : undefined;
   const hasVisibleActivity = reasoningSteps > 0 || toolCalls > 0 || cliCount > 0 || mcpCount > 0 || fileCount > 0;
-  const activityDuration = formatActivityDuration(activityDurationMs(messages, isTurnStreaming, now));
+  const durationMs = activityDurationMs(messages, isTurnStreaming, now, turnLatencyMs);
+  const activityDuration = formatActivityDuration(durationMs);
   const thoughtLabel = isTurnStreaming
     ? t("message.activityThinkingFor", {
         duration: activityDuration,
         defaultValue: "Thinking for {{duration}}",
       })
+    : durationMs <= 0
+      ? t("message.activityThought", { defaultValue: "Thought" })
     : t("message.activityThoughtFor", {
         duration: activityDuration,
         defaultValue: "Thought for {{duration}}",
@@ -500,7 +506,15 @@ function shortFileName(path: string): string {
   return path.split(/[\\/]/).pop() || path;
 }
 
-function activityDurationMs(messages: UIMessage[], active: boolean, now: number): number {
+function activityDurationMs(
+  messages: UIMessage[],
+  active: boolean,
+  now: number,
+  completedLatencyMs?: number,
+): number {
+  if (!active && Number.isFinite(completedLatencyMs) && completedLatencyMs! >= 0) {
+    return Math.round(completedLatencyMs!);
+  }
   const timestamps = messages
     .map((message) => message.createdAt)
     .filter((value) => Number.isFinite(value));
@@ -513,7 +527,7 @@ function activityDurationMs(messages: UIMessage[], active: boolean, now: number)
 }
 
 function formatActivityDuration(ms: number): string {
-  const seconds = Math.max(0, Math.round(ms / 1000));
+  const seconds = ms > 0 && ms < 1000 ? 1 : Math.max(0, Math.round(ms / 1000));
   if (seconds < 60) return `${seconds}s`;
   const minutes = Math.floor(seconds / 60);
   const rest = seconds % 60;

webui/src/components/thread/ThreadMessages.tsx

@@ -134,6 +134,7 @@ function reasoningOnlyMessageFromAnswer(message: UIMessage): UIMessage {
     reasoningStreaming: message.reasoningStreaming,
     isStreaming: message.reasoningStreaming,
     activitySegmentId: message.activitySegmentId,
+    latencyMs: message.latencyMs,
   };
 }
 
@@ -204,6 +205,8 @@ export function ThreadMessages({
           unit.type === "cluster"
           && next?.type === "single"
           && next.message.role === "assistant";
+        const turnLatencyMs =
+          unit.type === "cluster" ? activityClusterTurnLatencyMs(unit.messages, next) : undefined;
 
         return (
           <div key={unitKey(unit, index)} className={marginTop}>
@@ -212,6 +215,7 @@ export function ThreadMessages({
                 messages={unit.messages}
                 isTurnStreaming={index === liveActivityClusterIndex}
                 hasBodyBelow={hasBodyBelow}
+                turnLatencyMs={turnLatencyMs}
                 cliApps={cliApps}
                 mcpPresets={mcpPresets}
               />
@@ -234,6 +238,28 @@ export function ThreadMessages({
   );
 }
 
+function activityClusterTurnLatencyMs(
+  messages: UIMessage[],
+  next: DisplayUnit | undefined,
+): number | undefined {
+  for (let i = messages.length - 1; i >= 0; i -= 1) {
+    const latency = messages[i].latencyMs;
+    if (typeof latency === "number" && Number.isFinite(latency) && latency >= 0) {
+      return latency;
+    }
+  }
+  if (
+    next?.type === "single"
+    && next.message.role === "assistant"
+    && typeof next.message.latencyMs === "number"
+    && Number.isFinite(next.message.latencyMs)
+    && next.message.latencyMs >= 0
+  ) {
+    return next.message.latencyMs;
+  }
+  return undefined;
+}
+
 function currentActivityClusterIndex(units: DisplayUnit[]): number {
   const last = units.length - 1;
   return units[last]?.type === "cluster" ? last : -1;