Don't count tool waits in output speed

src-tauri/src/commands/token_usage.rs

@@ -323,6 +323,14 @@ fn bucket_key(date: NaiveDate, bucket: TokenUsageBucket) -> String {
 
 // ─── Aggregation ────────────────────────────────────────────────────────
 
+/// Per-conversation running sums, keyed by conversation id:
+/// `(total_tokens, output_tokens, duration_ms, turns, last_activity, agent, folder_id)`.
+type ConversationAcc = (u64, u64, u64, u64, DateTime<Utc>, String, i32);
+
+/// [`ConversationAcc`] with the conversation id folded in as the first element,
+/// so the top-conversations list can be sorted as one flat tuple.
+type ConversationTotals = (i32, u64, u64, u64, u64, DateTime<Utc>, String, i32);
+
 /// Running sums for one group (a bucket, or one slice of a breakdown).
 #[derive(Debug, Default, Clone)]
 struct Acc {
@@ -416,7 +424,7 @@ pub(crate) fn aggregate_report(
     let mut by_agent: HashMap<String, Acc> = HashMap::new();
     let mut by_model: HashMap<String, Acc> = HashMap::new();
     let mut heat: HashMap<(u8, u8), (u64, u64)> = HashMap::new();
-    let mut per_conversation: HashMap<i32, (u64, u64, DateTime<Utc>, String, i32)> = HashMap::new();
+    let mut per_conversation: HashMap<i32, ConversationAcc> = HashMap::new();
     let mut active_dates: HashSet<NaiveDate> = HashSet::new();
     let mut first_activity: Option<DateTime<Utc>> = None;
     let mut last_activity: Option<DateTime<Utc>> = None;
@@ -452,16 +460,20 @@ pub(crate) fn aggregate_report(
         cell.1 += 1;
 
         let entry = per_conversation.entry(row.conversation_id).or_insert((
+            0,
+            0,
             0,
             0,
             row.occurred_at,
             row.agent_type.clone(),
             row.folder_id,
         ));
         entry.0 += row.total_tokens.max(0) as u64;
-        entry.1 += 1;
-        if row.occurred_at > entry.2 {
-            entry.2 = row.occurred_at;
+        entry.1 += row.output_tokens.max(0) as u64;
+        entry.2 += row.duration_ms.max(0) as u64;
+        entry.3 += 1;
+        if row.occurred_at > entry.4 {
+            entry.4 = row.occurred_at;
         }
 
         active_dates.insert(date);
@@ -518,10 +530,10 @@ pub(crate) fn aggregate_report(
         .collect();
     heatmap.sort_by_key(|a| (a.weekday, a.hour));
 
-    let mut top: Vec<(i32, u64, u64, DateTime<Utc>, String, i32)> = per_conversation
+    let mut top: Vec<ConversationTotals> = per_conversation
         .into_iter()
-        .map(|(id, (tokens, turns, last, agent, folder_id))| {
-            (id, tokens, turns, last, agent, folder_id)
+        .map(|(id, (tokens, output, duration_ms, turns, last, agent, folder_id))| {
+            (id, tokens, output, duration_ms, turns, last, agent, folder_id)
         })
         .collect();
     // Ties broken by id so the list is stable across identical requests.
@@ -530,15 +542,19 @@ pub(crate) fn aggregate_report(
     let top_conversations: Vec<TokenUsageConversationItem> = top
         .into_iter()
         .map(
-            |(id, tokens, turns, last, agent, folder_id)| TokenUsageConversationItem {
-                conversation_id: id,
-                // Filled by the command layer, which owns the DB handle.
-                title: None,
-                agent_type: agent,
-                folder_label: opts.folder_labels.get(&folder_id).cloned(),
-                total_tokens: tokens,
-                turn_count: turns,
-                last_activity_at: last,
+            |(id, tokens, output, duration_ms, turns, last, agent, folder_id)| {
+                TokenUsageConversationItem {
+                    conversation_id: id,
+                    // Filled by the command layer, which owns the DB handle.
+                    title: None,
+                    agent_type: agent,
+                    folder_label: opts.folder_labels.get(&folder_id).cloned(),
+                    total_tokens: tokens,
+                    output_tokens: output,
+                    turn_count: turns,
+                    duration_ms,
+                    last_activity_at: last,
+                }
             },
         )
         .collect();
@@ -1625,6 +1641,30 @@ mod tests {
         assert_eq!(report.by_model[0].key, UNKNOWN_MODEL);
     }
 
+    #[test]
+    fn top_conversations_carry_output_tokens_and_duration() {
+        let labels = HashMap::new();
+        let mut heavy = row("2026-08-01T10:00:00Z", 100);
+        heavy.output_tokens = 40;
+        heavy.duration_ms = 2000;
+        let mut light = row("2026-08-01T11:00:00Z", 10);
+        light.conversation_id = 2;
+        light.output_tokens = 5;
+        light.duration_ms = 500;
+
+        let report = aggregate_report(
+            &[heavy, light],
+            &[],
+            &opts(&labels, TokenUsageBucket::Day, 0, None, None),
+        );
+        assert_eq!(report.top_conversations[0].conversation_id, 1);
+        assert_eq!(report.top_conversations[0].output_tokens, 40);
+        assert_eq!(report.top_conversations[0].duration_ms, 2000);
+        assert_eq!(report.top_conversations[1].conversation_id, 2);
+        assert_eq!(report.top_conversations[1].output_tokens, 5);
+        assert_eq!(report.top_conversations[1].duration_ms, 500);
+    }
+
     #[test]
     fn heatmap_uses_local_weekday_and_hour() {
         let labels = HashMap::new();

src-tauri/src/models/token_usage.rs

@@ -154,7 +154,10 @@ pub struct TokenUsageConversationItem {
     pub agent_type: String,
     pub folder_label: Option<String>,
     pub total_tokens: u64,
+    pub output_tokens: u64,
     pub turn_count: u64,
+    /// Summed recorded generation time of the counted turns.
+    pub duration_ms: u64,
     pub last_activity_at: DateTime<Utc>,
 }
 

src/components/conversations/session-details-content.tsx

@@ -21,6 +21,8 @@ import {
 import { getFolderConversation } from "@/lib/api"
 import { useCopiedFlag } from "@/hooks/use-copied-flag"
 import { pickModelFromTurns } from "./active-session-details"
+import { useSessionOutputSpeed } from "@/hooks/use-session-output-speed"
+import { averageOutputTps, formatTokPerSec } from "@/lib/token-speed"
 import { AgentIcon } from "@/components/agent-icon"
 import { ConversationStatusDot } from "./conversation-status-dot"
 
@@ -141,15 +143,17 @@ export function InfoItem({
   children,
   className,
   valueClassName,
+  title,
 }: {
   /** Usually plain text; a node so a caller can hang a badge off the label. */
   label: ReactNode
   children: ReactNode
   className?: string
   valueClassName?: string
+  title?: string
 }) {
   return (
-    <div className={cn("min-w-0 space-y-0.5", className)}>
+    <div className={cn("min-w-0 space-y-0.5", className)} title={title}>
       <dt className="flex min-w-0 items-center gap-1.5 text-xs text-muted-foreground">
         {label}
       </dt>
@@ -312,6 +316,12 @@ export function SessionDetailsContent({
     ctxMax
   )
   const durationMs = resolveSessionDurationMs(summary, stats)
+  const liveOutputSpeed = useSessionOutputSpeed(summary.id)
+  const recordedOutputTps = averageOutputTps(
+    usage?.output_tokens ?? 0,
+    stats?.total_duration_ms ?? 0
+  )
+  const outputTps = liveOutputSpeed?.averageTps ?? recordedOutputTps
   // Never coerce an unknown `used` to 0 — some parsers infer the model's
   // context cap without any usage figure, so render "— / max" rather than a
   // bogus "0 / max".
@@ -324,11 +334,12 @@ export function SessionDetailsContent({
         ? formatTokenCount(ctxUsed)
         : null
   const hasTokenInfo =
-    stats != null &&
-    (totalTokens != null ||
-      usage != null ||
-      contextWindowValue != null ||
-      durationMs > 0)
+    outputTps != null ||
+    (stats != null &&
+      (totalTokens != null ||
+        usage != null ||
+        contextWindowValue != null ||
+        durationMs > 0))
 
   const numeric = "font-mono tabular-nums"
 
@@ -434,6 +445,15 @@ export function SessionDetailsContent({
                 {formatDuration(durationMs)}
               </InfoItem>
             )}
+            {outputTps != null && (
+              <InfoItem
+                label={t("outputSpeed")}
+                valueClassName={numeric}
+                title={t("outputSpeedTooltip")}
+              >
+                {formatTokPerSec(outputTps)}
+              </InfoItem>
+            )}
           </dl>
         ) : (
           <div className="text-muted-foreground">{t("noStats")}</div>

src/components/message/live-turn-stats.tsx

@@ -16,11 +16,13 @@ import { FilePenLine, Plane, Timer } from "lucide-react"
 import type { AgentType } from "@/lib/types"
 import { AgentIcon } from "@/components/agent-icon"
 import { useTokenOutputSpeed } from "@/hooks/use-token-output-speed"
+import { formatTokPerSec } from "@/lib/token-speed"
 
 interface LiveTurnStatsProps {
   message: LiveMessage
   agentType: AgentType
   isStreaming?: boolean
+  conversationId?: number | null
 }
 
 interface LineChangeStats {
@@ -311,12 +313,13 @@ export function LiveTurnStats({
   message,
   agentType,
   isStreaming = true,
+  conversationId = null,
 }: LiveTurnStatsProps) {
   const locale = useLocale()
   const t = useTranslations("Folder.chat.liveTurnStats")
   const [elapsed, setElapsed] = useState(() => Date.now() - message.startedAt)
   const editStats = useMemo(() => extractLiveEditStats(message), [message])
-  const tps = useTokenOutputSpeed(message)
+  const tps = useTokenOutputSpeed(message, { conversationId })
   const compactNumberFormatter = useMemo(
     () =>
       new Intl.NumberFormat(locale, {
@@ -392,7 +395,7 @@ export function LiveTurnStats({
                 aria-label={t("outputSpeedAria")}
                 className="h-3 w-3 shrink-0"
               />
-              {tps.toFixed(1)} tok/s
+              {formatTokPerSec(tps)}
             </span>
           </>
         )}

src/components/message/message-list-view.tsx

@@ -1154,6 +1154,7 @@ export function MessageListView({
             message={liveMessage}
             agentType={agentType}
             isStreaming={connStatus === "prompting"}
+            conversationId={conversationId}
           />
         )}
         {/* Shared overlay stack pinned to the inline-start edge (top-left in LTR,

src/components/token-usage/token-usage-page.tsx

@@ -54,6 +54,7 @@ import { toErrorMessage } from "@/lib/app-error"
 import { getAgentLabel } from "@/lib/custom-agents"
 import { subscribe } from "@/lib/platform"
 import { formatTokenCount } from "@/lib/token-format"
+import { averageOutputTps, formatTokPerSec } from "@/lib/token-speed"
 import {
   averagePerActiveDay,
   averagePerConversation,
@@ -561,6 +562,9 @@ export function TokenUsagePage() {
 
   const totals = report?.totals
   const cache = totals ? cacheHitRate(totals) : null
+  const averageTps = totals
+    ? averageOutputTps(totals.output_tokens, totals.duration_ms)
+    : null
   const heat = useMemo(
     () => buildHeatMatrix(report?.heatmap ?? []),
     [report?.heatmap]
@@ -1204,9 +1208,15 @@ export function TokenUsagePage() {
                           className="border-s border-t lg:border-t-0"
                           label={t("tileGenTime")}
                           value={formatDuration(totals.duration_ms)}
-                          hint={`${t("avgPerActiveDay")} ${formatTokenCount(
-                            Math.round(averagePerActiveDay(totals))
-                          )}`}
+                          hint={
+                            averageTps != null
+                              ? t("avgOutputSpeed", {
+                                  speed: formatTokPerSec(averageTps),
+                                })
+                              : `${t("avgPerActiveDay")} ${formatTokenCount(
+                                  Math.round(averagePerActiveDay(totals))
+                                )}`
+                          }
                         />
                       </section>
 
@@ -1384,31 +1394,45 @@ export function TokenUsagePage() {
                           </p>
                         ) : (
                           <ol className="divide-y divide-border">
-                            {report.top_conversations.map((c, i) => (
-                              <li
-                                key={c.conversation_id}
-                                className="flex items-center gap-3 py-2 first:pt-0 last:pb-0"
-                              >
-                                <span
-                                  aria-hidden="true"
-                                  className="w-5 shrink-0 font-mono text-[0.625rem] tabular-nums text-muted-foreground/70"
+                            {report.top_conversations.map((c, i) => {
+                              const sessionTps = averageOutputTps(
+                                c.output_tokens,
+                                c.duration_ms
+                              )
+                              return (
+                                <li
+                                  key={c.conversation_id}
+                                  className="flex items-center gap-3 py-2 first:pt-0 last:pb-0"
                                 >
-                                  {String(i + 1).padStart(2, "0")}
-                                </span>
-                                <span className="min-w-0 flex-1 truncate text-[0.8125rem]">
-                                  {c.title || t("untitledSession")}
-                                </span>
-                                <span className="hidden shrink-0 truncate text-xs text-muted-foreground sm:block sm:max-w-[10rem]">
-                                  {c.folder_label}
-                                </span>
-                                <span className="shrink-0 text-xs text-muted-foreground">
-                                  {getAgentLabel(c.agent_type as AgentType)}
-                                </span>
-                                <span className="shrink-0 font-mono text-xs tabular-nums">
-                                  {formatTokenCount(c.total_tokens)}
-                                </span>
-                              </li>
-                            ))}
+                                  <span
+                                    aria-hidden="true"
+                                    className="w-5 shrink-0 font-mono text-[0.625rem] tabular-nums text-muted-foreground/70"
+                                  >
+                                    {String(i + 1).padStart(2, "0")}
+                                  </span>
+                                  <span className="min-w-0 flex-1 truncate text-[0.8125rem]">
+                                    {c.title || t("untitledSession")}
+                                  </span>
+                                  <span className="hidden shrink-0 truncate text-xs text-muted-foreground sm:block sm:max-w-[10rem]">
+                                    {c.folder_label}
+                                  </span>
+                                  <span className="shrink-0 text-xs text-muted-foreground">
+                                    {getAgentLabel(c.agent_type as AgentType)}
+                                  </span>
+                                  <span className="shrink-0 font-mono text-xs tabular-nums">
+                                    {formatTokenCount(c.total_tokens)}
+                                  </span>
+                                  {sessionTps != null && (
+                                    <span
+                                      className="hidden shrink-0 font-mono text-xs tabular-nums text-muted-foreground sm:block"
+                                      title={t("outputSpeedTooltip")}
+                                    >
+                                      {formatTokPerSec(sessionTps)}
+                                    </span>
+                                  )}
+                                </li>
+                              )
+                            })}
                           </ol>
                         )}
                       </Panel>

src/hooks/use-session-output-speed.ts

@@ -0,0 +1,23 @@
+"use client"
+
+import { useCallback, useSyncExternalStore } from "react"
+
+import {
+  getSessionOutputSpeed,
+  subscribeSessionOutputSpeed,
+  type SessionOutputSpeed,
+} from "@/lib/session-output-speed"
+
+export function useSessionOutputSpeed(
+  conversationId: number | null | undefined
+): SessionOutputSpeed | null {
+  const subscribe = useCallback(
+    (onStoreChange: () => void) => subscribeSessionOutputSpeed(onStoreChange),
+    []
+  )
+  const getSnapshot = useCallback(
+    () => getSessionOutputSpeed(conversationId),
+    [conversationId]
+  )
+  return useSyncExternalStore(subscribe, getSnapshot, getSnapshot)
+}