chore(package): bump version to 1.0.10-express

.env.example

@@ -17,6 +17,9 @@ ADMIN_API_TOKEN_TTL_DAYS=30
 CORS_ORIGINS=http://localhost:3000,http://127.0.0.1:3000
 ADMIN_TRUSTED_PROXY_IPS=
 
+# Model routing
+MODEL_LIST_INCLUDE_ROUTING_METADATA=false
+
 # OpenID Connect
 OIDC_ISSUER_URL=
 OIDC_CLIENT_ID=

.forgejo/workflows/publish-images.yml

@@ -80,11 +80,12 @@ jobs:
 
           docker_args=(
             build
-            .
             --file Dockerfile
+            --progress=plain
             --target app
             --label "org.opencontainers.image.source=${REPOSITORY_URL}"
             --label "org.opencontainers.image.revision=${GITHUB_SHA}"
+            .
           )
 
           for tag in "${tags[@]}"; do
@@ -96,5 +97,3 @@ jobs:
           for tag in "${tags[@]}"; do
             docker push "${tag}"
           done
-
-          # The single OCI image contains the public API and the admin dashboard runtime.

AGENTS.md

@@ -78,9 +78,11 @@ pnpm run bench      # 벤치마크 실행
 | `OIDC_ALLOWED_EMAILS` | empty | 관리자 접근을 허용할 이메일 목록 |
 | `OIDC_SCOPES` | `openid profile email` | OIDC authorization scope |
 | `MODEL_CATALOG_REFRESH_MIN_MS` | `300000` 예시 | 모델 카탈로그 refresh 최소 간격(ms) |
+| `MODEL_LIST_INCLUDE_ROUTING_METADATA` | `false` | `true`이면 `/v1/models` model object에 비표준 `kyush_router` routing metadata 추가 |
+| `DETAIL_STREAM_LOG_MODE` | `compact` | 상세 로그에서 stream response body 저장 방식 (`compact`, `raw`, `both`, `off`) |
 
 ## Detailed Docs
-관련 기능을 수정하기 전에 해당 문서를 반드시 먼저 읽으세요.
+관련 기능을 수정하기 전에 해당 문서를 반드시 UTF-8로 먼저 읽으세요.
 
 클라이언트 중심
 - [docs/client.md](docs/client.md) — 클라이언트 구조, `/dashboard` 라우팅, 관리자 UI 동작

client/src/api/client.ts

@@ -99,7 +99,7 @@ export const api = {
   users: {
     getAll: (): Promise<User[]> => fetchJson<User[]>(`${API_BASE}/admin/users`),
     getById: (id: number): Promise<User> => fetchJson<User>(`${API_BASE}/admin/users/${id}`),
-    create: (data: { name: string; email?: string; api_key?: string; detail_logging?: boolean }): Promise<User> =>
+    create: (data: { name: string; email?: string; api_key?: string; detail_logging?: boolean; copy_reasoning_to_reasoning_content?: boolean }): Promise<User> =>
       fetchJson<User>(`${API_BASE}/admin/users`, { method: 'POST', body: JSON.stringify(data) }),
     update: (id: number, data: Partial<User>): Promise<User> =>
       fetchJson<User>(`${API_BASE}/admin/users/${id}`, { method: 'PUT', body: JSON.stringify(data) }),

client/src/routes/Analytics.tsx

@@ -1,6 +1,8 @@
-import { Show, createMemo, createResource, createSignal, type Component } from 'solid-js';
+import RefreshCw from 'lucide-solid/icons/refresh-cw';
+import { createEffect, createMemo, createResource, createSignal, onCleanup, type Component } from 'solid-js';
 import { api } from '../api/client';
 import { Layout } from '../components/Layout';
+import { formatDurationMs } from '../ui/lib/format';
 import {
   BoxPlotChart,
   ChartLegend,
@@ -8,11 +10,11 @@ import {
   CommandBar,
   CommandBarGroup,
   HistogramChart,
-  MetaCluster,
   PageHeader,
   Panel,
   Select,
   SummaryStrip,
+  Switch,
   TimeSeriesChart,
 } from '../ui';
 
@@ -32,10 +34,15 @@ export const Analytics: Component = () => {
   const [hiddenDailySeries, setHiddenDailySeries] = createSignal<Set<string>>(new Set());
   const [hiddenResponseSeries, setHiddenResponseSeries] = createSignal<Set<string>>(new Set());
   const [hiddenModelSeries, setHiddenModelSeries] = createSignal<Set<string>>(new Set());
+  const [isAutoRefresh, setIsAutoRefresh] = createSignal(false);
+  const [refreshInterval, setRefreshInterval] = createSignal('10');
+  const [refreshKey, setRefreshKey] = createSignal(0);
+  const [dailyVolumeScale, setDailyVolumeScale] = createSignal<'linear' | 'log'>('linear');
 
   const filters = createMemo(() => ({
     days: Number(days()),
     backendId: backendFilter() === 'all' ? undefined : Number(backendFilter()),
+    key: refreshKey(),
   }));
 
   const [backends] = createResource(() => api.backends.getAll());
@@ -44,6 +51,19 @@ export const Analytics: Component = () => {
   const [modelTrends] = createResource(filters, (params) => api.analytics.getModelTrends({ backendId: params.backendId, days: params.days, limit: 8 }));
   const [histogram] = createResource(filters, (params) => api.analytics.getResponseLengthHistogram({ backendId: params.backendId, days: params.days, bins: 20 }));
   const [boxPlot] = createResource(filters, (params) => api.analytics.getResponseLengthBoxPlot(params.backendId, params.days));
+  const currentDailyTotals = createMemo(() => dailyTotals.latest ?? dailyTotals());
+  const currentBackendQuality = createMemo(() => backendQuality.latest ?? backendQuality());
+  const currentModelTrends = createMemo(() => modelTrends.latest ?? modelTrends());
+  const currentHistogram = createMemo(() => histogram.latest ?? histogram());
+  const currentBoxPlot = createMemo(() => boxPlot.latest ?? boxPlot());
+  const analyticsLoading = createMemo(() => dailyTotals.loading || backendQuality.loading || modelTrends.loading || histogram.loading || boxPlot.loading);
+
+  createEffect(() => {
+    if (!isAutoRefresh()) return;
+    const ms = Number(refreshInterval()) * 1000;
+    const id = setInterval(() => setRefreshKey((key) => key + 1), ms);
+    onCleanup(() => clearInterval(id));
+  });
 
   const backendOptions = createMemo(() => [
     { value: 'all', label: 'All Backends' },
@@ -62,7 +82,7 @@ export const Analytics: Component = () => {
   });
 
   const dailyVolumeRows = createMemo(() =>
-    (dailyTotals() ?? []).map((row) => ({
+    (currentDailyTotals() ?? []).map((row) => ({
       date: row.date,
       requests: row.total_requests,
       tokens: row.total_tokens,
@@ -71,7 +91,7 @@ export const Analytics: Component = () => {
 
   const responseTimeRows = createMemo(() => {
     const grouped = new Map<string, AnalyticsChartRow>();
-    for (const row of backendQuality() ?? []) {
+    for (const row of currentBackendQuality() ?? []) {
       const entry: AnalyticsChartRow = grouped.get(row.date) ?? { date: row.date };
       entry[`backend_${row.backend_id}`] = row.avg_response_time_ms;
       grouped.set(row.date, entry);
@@ -80,7 +100,7 @@ export const Analytics: Component = () => {
   });
 
   const responseTimeSeries = createMemo(() => {
-    const ids = Array.from(new Set((backendQuality() ?? []).map((row) => row.backend_id))).sort((left, right) => left - right);
+    const ids = Array.from(new Set((currentBackendQuality() ?? []).map((row) => row.backend_id))).sort((left, right) => left - right);
     return ids.map((backendId, index) => ({
       key: `backend_${backendId}`,
       label: backendNameById().get(backendId) ?? `Backend ${backendId}`,
@@ -90,7 +110,7 @@ export const Analytics: Component = () => {
 
   const reliabilityRows = createMemo(() => {
     const grouped = new Map<string, { requests: number; errors: number }>();
-    for (const row of backendQuality() ?? []) {
+    for (const row of currentBackendQuality() ?? []) {
       const entry = grouped.get(row.date) ?? { requests: 0, errors: 0 };
       entry.requests += row.total_requests;
       entry.errors += row.error_count;
@@ -108,7 +128,7 @@ export const Analytics: Component = () => {
 
   const modelTrendRows = createMemo(() => {
     const grouped = new Map<string, AnalyticsChartRow>();
-    for (const row of modelTrends() ?? []) {
+    for (const row of currentModelTrends() ?? []) {
       const entry: AnalyticsChartRow = grouped.get(row.date) ?? { date: row.date };
       entry[`model_${row.model}`] = row.request_count;
       grouped.set(row.date, entry);
@@ -117,7 +137,7 @@ export const Analytics: Component = () => {
   });
 
   const modelTrendSeries = createMemo(() => {
-    const models = Array.from(new Set((modelTrends() ?? []).map((row) => row.model)));
+    const models = Array.from(new Set((currentModelTrends() ?? []).map((row) => row.model)));
     return models.map((model, index) => ({
       key: `model_${model}`,
       label: model,
@@ -126,7 +146,7 @@ export const Analytics: Component = () => {
   });
 
   const summaryItems = createMemo(() => {
-    const totals = (dailyTotals() ?? []).reduce(
+    const totals = (currentDailyTotals() ?? []).reduce(
       (acc, row) => {
         acc.requests += row.total_requests;
         acc.tokens += row.total_tokens;
@@ -134,15 +154,15 @@ export const Analytics: Component = () => {
       },
       { requests: 0, tokens: 0 }
     );
-    const qualityRows = backendQuality() ?? [];
+    const qualityRows = currentBackendQuality() ?? [];
     const avgLatency =
       qualityRows.length === 0 ? 0 : qualityRows.reduce((sum, row) => sum + row.avg_response_time_ms, 0) / qualityRows.length;
     const errorCount = qualityRows.reduce((sum, row) => sum + row.error_count, 0);
 
     return [
       { label: 'Requests', value: formatInteger.format(totals.requests), hint: `Last ${days()} days` },
-      { label: 'Tokens', value: formatInteger.format(totals.tokens), hint: 'Aggregated daily total tokens' },
-      { label: 'Avg Response', value: `${avgLatency.toFixed(1)}ms`, hint: 'Across visible backend series' },
+      { label: 'Tokens', value: formatInteger.format(totals.tokens), hint: `Selected ${days()}-day window total` },
+      { label: 'Avg Response', value: formatDurationMs(avgLatency), hint: 'Across visible backend series' },
       { label: 'Errors', value: formatInteger.format(errorCount), hint: 'Absolute backend error count' },
     ];
   });
@@ -175,6 +195,37 @@ export const Analytics: Component = () => {
             <Select label="Window" value={days()} options={dayOptions} onChange={setDays} />
             <Select label="Backend" value={backendFilter()} options={backendOptions()} onChange={setBackendFilter} />
           </CommandBarGroup>
+          <CommandBarGroup>
+            <Switch
+              label="Auto refresh"
+              checked={isAutoRefresh()}
+              onChange={setIsAutoRefresh}
+            />
+            <Select
+              label="Refresh Interval"
+              value={refreshInterval()}
+              options={[
+                { value: '5', label: 'Every 5s' },
+                { value: '10', label: 'Every 10s' },
+                { value: '30', label: 'Every 30s' },
+                { value: '60', label: 'Every 60s' },
+                { value: '600', label: 'Every 10m' },
+              ]}
+              onChange={setRefreshInterval}
+            />
+            <div class="ui-divider--vertical" />
+            <button
+              class="ui-button analytics__refresh-button"
+              classList={{ 'ui-button--loading': analyticsLoading() }}
+              type="button"
+              onClick={() => setRefreshKey((key) => key + 1)}
+              disabled={analyticsLoading()}
+              aria-busy={analyticsLoading()}
+            >
+              <RefreshCw />
+              {analyticsLoading() ? 'Refreshing' : 'Refresh'}
+            </button>
+          </CommandBarGroup>
         </CommandBar>
 
         <SummaryStrip items={summaryItems()} />
@@ -184,14 +235,25 @@ export const Analytics: Component = () => {
             title="Daily Volume"
             description="Daily request and token totals on shared time axis."
             actions={
-              <ChartLegend
-                items={[
-                  { key: 'requests', label: 'Requests', color: '#2357d8' },
-                  { key: 'tokens', label: 'Tokens', color: '#1f7a45' },
-                ]}
-                mutedKeys={hiddenDailySeries()}
-                onToggle={(key) => toggleHiddenKey(setHiddenDailySeries, key)}
-              />
+              <div style="display: flex; align-items: center; gap: 16px;">
+                <ChartLegend
+                  items={[
+                    { key: 'requests', label: 'Requests', color: '#2357d8' },
+                    { key: 'tokens', label: 'Tokens', color: '#1f7a45' },
+                  ]}
+                  mutedKeys={hiddenDailySeries()}
+                  onToggle={(key) => toggleHiddenKey(setHiddenDailySeries, key)}
+                />
+                <Select
+                  label="Scale"
+                  value={dailyVolumeScale()}
+                  options={[
+                    { value: 'linear', label: 'Linear' },
+                    { value: 'log', label: 'Log' },
+                  ]}
+                  onChange={setDailyVolumeScale}
+                />
+              </div>
             }
           >
             <TimeSeriesChart
@@ -208,6 +270,7 @@ export const Analytics: Component = () => {
               formatLeftValue={(value) => new Intl.NumberFormat('en-US').format(Math.round(value))}
               formatRightValue={(value) => new Intl.NumberFormat('en-US', { notation: 'compact', maximumFractionDigits: 1 }).format(value)}
               tooltipTitle="Daily request and token totals"
+              yScaleType={dailyVolumeScale()}
             />
           </Panel>
 
@@ -252,8 +315,8 @@ export const Analytics: Component = () => {
               showLegend={false}
               hiddenKeys={hiddenResponseSeries()}
               onToggleLegend={(key) => toggleHiddenKey(setHiddenResponseSeries, key)}
-              yLeftLabel="Milliseconds"
-              formatLeftValue={(value) => `${value.toFixed(0)}ms`}
+              yLeftLabel="Response time"
+              formatLeftValue={formatDurationMs}
               tooltipTitle="Average backend response time"
             />
           </Panel>
@@ -283,22 +346,16 @@ export const Analytics: Component = () => {
         </div>
 
         <div class="ui-section-grid analytics__grid--spread-wide">
-          <Panel title="Response Length Distribution" description="Histogram of completion token lengths across the selected window.">
-            <MetaCluster
-              items={[
-                { key: 'Metric', value: 'completion_tokens' },
-              ]}
+          <Panel title="Response Length Distribution" description="Log-scaled completion_tokens histogram across the selected window.">
+            <HistogramChart
+              data={currentHistogram() ?? []}
+              xTickUnit="tok"
+              yTickUnit="req"
             />
-            <HistogramChart data={histogram() ?? []} />
           </Panel>
 
-          <Panel title="Daily Response Length Spread" description="Completion token box plot by day using min / q1 / median / q3 / max summary.">
-            <MetaCluster
-              items={[
-                { key: 'Outliers', value: 'Hidden in this view' },
-              ]}
-            />
-            <BoxPlotChart data={boxPlot() ?? []} />
+          <Panel title="Daily Response Length Spread" description="Log-scaled daily completion_tokens spread; outliers are hidden.">
+            <BoxPlotChart data={currentBoxPlot() ?? []} />
           </Panel>
         </div>
       </div>

client/src/routes/Backends.tsx

@@ -1,4 +1,4 @@
-import { For, createResource, createSignal, Show, type Component } from 'solid-js';
+import { For, createMemo, createResource, createSignal, Show, type Component } from 'solid-js';
 import Pencil from 'lucide-solid/icons/pencil';
 import Plus from 'lucide-solid/icons/plus';
 import RefreshCw from 'lucide-solid/icons/refresh-cw';
@@ -39,6 +39,7 @@ const emptyForm = (): BackendFormState => ({
 
 export const Backends: Component = () => {
   const [backends, { refetch }] = createResource(() => api.backends.getAll());
+  const currentBackends = createMemo(() => backends.state === 'ready' || backends.state === 'refreshing' ? backends.latest : undefined);
   const [dialogOpen, setDialogOpen] = createSignal(false);
   const [confirmOpen, setConfirmOpen] = createSignal(false);
   const [editingBackend, setEditingBackend] = createSignal<Backend | null>(null);
@@ -202,11 +203,11 @@ export const Backends: Component = () => {
 
         <Panel title="Backend catalog" description="Operational list with overflow-safe URL presentation and compact actions.">
           <Show
-            when={!backends.loading || (backends()?.length ?? 0) > 0}
+            when={!backends.loading || (currentBackends()?.length ?? 0) > 0}
             fallback={<EmptyState title="Loading backends" description="Reading upstream routing targets from the admin API." />}
           >
             <Show
-              when={(backends()?.length ?? 0) > 0}
+              when={(currentBackends()?.length ?? 0) > 0}
               fallback={
                 <EmptyState
                   title="No backends yet"
@@ -216,7 +217,7 @@ export const Backends: Component = () => {
               }
             >
               <DataGrid
-                rows={backends() ?? []}
+                rows={currentBackends() ?? []}
                 columns={[
                   { id: 'id', header: 'ID', mono: true, cell: (backend) => <span>{backend.id}</span> },
                   { id: 'name', header: 'Name', cell: (backend) => <span>{backend.name}</span> },
@@ -248,7 +249,7 @@ export const Backends: Component = () => {
                   },
                 ]}
                 getRowKey={(backend) => backend.id}
-                loading={backends.loading}
+                loading={backends.loading && (currentBackends()?.length ?? 0) === 0}
                 rowActions={(backend) => (
                   <div class="ui-row-actions">
                     <IconButton

client/src/routes/Dashboard.tsx

@@ -1,7 +1,8 @@
-import RefreshCcw from 'lucide-solid/icons/refresh-ccw';
-import { Show, createMemo, createResource, createSignal, type Component } from 'solid-js';
+import RefreshCw from 'lucide-solid/icons/refresh-cw';
+import { Show, createEffect, createMemo, createResource, createSignal, onCleanup, type Component } from 'solid-js';
 import { api } from '../api/client';
 import { Layout } from '../components/Layout';
+import { formatDurationMs } from '../ui/lib/format';
 import {
   ChartLegend,
   ComboChart,
@@ -13,6 +14,7 @@ import {
   Panel,
   Select,
   SummaryStrip,
+  Switch,
   TimeSeriesChart,
 } from '../ui';
 
@@ -32,10 +34,23 @@ export const Dashboard: Component = () => {
   const [hiddenTrafficSeries, setHiddenTrafficSeries] = createSignal<Set<string>>(new Set());
   const [hiddenLatencySeries, setHiddenLatencySeries] = createSignal<Set<string>>(new Set());
   const [hiddenModelSeries, setHiddenModelSeries] = createSignal<Set<string>>(new Set());
+  const [isAutoRefresh, setIsAutoRefresh] = createSignal(false);
+  const [refreshInterval, setRefreshInterval] = createSignal('10');
+  const [refreshKey, setRefreshKey] = createSignal(0);
+  const [trafficVolumeScale, setTrafficVolumeScale] = createSignal<'linear' | 'log'>('linear');
 
   const windowDays = createMemo(() => Number(days()));
-  const [summary, { refetch }] = createResource(windowDays, (value) => api.dashboard.getSummary(value));
+  const summarySource = createMemo(() => ({ days: windowDays(), key: refreshKey() }));
+  const [summary] = createResource(summarySource, (value) => api.dashboard.getSummary(value.days));
   const [backends] = createResource(() => api.backends.getAll());
+  const currentSummary = createMemo(() => summary.latest ?? summary());
+
+  createEffect(() => {
+    if (!isAutoRefresh()) return;
+    const ms = Number(refreshInterval()) * 1000;
+    const id = setInterval(() => setRefreshKey((k) => k + 1), ms);
+    onCleanup(() => clearInterval(id));
+  });
 
   const backendNameById = createMemo(() => {
     const entries = new Map<number, string>();
@@ -46,7 +61,7 @@ export const Dashboard: Component = () => {
   });
 
   const trafficRows = createMemo(() =>
-    (summary()?.series.daily_totals ?? []).map((row) => ({
+    (currentSummary()?.series.daily_totals ?? []).map((row) => ({
       date: row.date,
       requests: row.total_requests,
       tokens: row.total_tokens,
@@ -55,7 +70,7 @@ export const Dashboard: Component = () => {
 
   const reliabilityRows = createMemo(() => {
     const grouped = new Map<string, { requests: number; errors: number }>();
-    for (const row of summary()?.series.backend_quality ?? []) {
+    for (const row of currentSummary()?.series.backend_quality ?? []) {
       const entry = grouped.get(row.date) ?? { requests: 0, errors: 0 };
       entry.requests += row.total_requests;
       entry.errors += row.error_count;
@@ -73,7 +88,7 @@ export const Dashboard: Component = () => {
 
   const latencyRows = createMemo(() => {
     const grouped = new Map<string, DashboardChartRow>();
-    for (const row of summary()?.series.backend_quality ?? []) {
+    for (const row of currentSummary()?.series.backend_quality ?? []) {
       const entry: DashboardChartRow = grouped.get(row.date) ?? { date: row.date };
       entry[`backend_${row.backend_id}`] = row.avg_response_time_ms;
       grouped.set(row.date, entry);
@@ -82,7 +97,7 @@ export const Dashboard: Component = () => {
   });
 
   const latencySeries = createMemo(() => {
-    const ids = Array.from(new Set((summary()?.series.backend_quality ?? []).map((row) => row.backend_id))).sort((left, right) => left - right);
+    const ids = Array.from(new Set((currentSummary()?.series.backend_quality ?? []).map((row) => row.backend_id))).sort((left, right) => left - right);
     return ids.map((backendId, index) => ({
       key: `backend_${backendId}`,
       label: backendNameById().get(backendId) ?? `Backend ${backendId}`,
@@ -92,7 +107,7 @@ export const Dashboard: Component = () => {
 
   const modelRows = createMemo(() => {
     const grouped = new Map<string, DashboardChartRow>();
-    for (const row of summary()?.series.model_trends ?? []) {
+    for (const row of currentSummary()?.series.model_trends ?? []) {
       const entry: DashboardChartRow = grouped.get(row.date) ?? { date: row.date };
       entry[`model_${row.model}`] = row.request_count;
       grouped.set(row.date, entry);
@@ -101,7 +116,7 @@ export const Dashboard: Component = () => {
   });
 
   const modelSeries = createMemo(() => {
-    const models = Array.from(new Set((summary()?.series.model_trends ?? []).map((row) => row.model)));
+    const models = Array.from(new Set((currentSummary()?.series.model_trends ?? []).map((row) => row.model)));
     return models.map((model, index) => ({
       key: `model_${model}`,
       label: model,
@@ -110,7 +125,7 @@ export const Dashboard: Component = () => {
   });
 
   const summaryItems = createMemo(() => {
-    const payload = summary();
+    const payload = currentSummary();
     const latestTraffic = payload?.series.daily_totals[payload.series.daily_totals.length - 1];
 
     return [
@@ -122,7 +137,7 @@ export const Dashboard: Component = () => {
   });
 
   const cacheStateItems = createMemo(() => {
-    const counts = summary()?.health.cache_state_counts;
+    const counts = currentSummary()?.health.cache_state_counts;
     if (!counts) return [];
     return [
       { key: 'Ready', value: String(counts.ready) },
@@ -133,7 +148,7 @@ export const Dashboard: Component = () => {
   });
 
   const scriptItems = createMemo(() => {
-    const payload = summary();
+    const payload = currentSummary();
     if (!payload) return [];
 
     return [
@@ -144,7 +159,7 @@ export const Dashboard: Component = () => {
   });
 
   const accessItems = createMemo(() => {
-    const payload = summary();
+    const payload = currentSummary();
     if (!payload) return [];
 
     return [
@@ -176,13 +191,43 @@ export const Dashboard: Component = () => {
         <PageHeader
           title="Dashboard"
           description="Operations cockpit for router health, traffic shape, and the configuration context behind current behavior."
-          actions={<button class="ui-button" type="button" onClick={() => void refetch()}><RefreshCcw />Refresh</button>}
         />
 
         <CommandBar class="analytics__filters">
           <CommandBarGroup>
             <Select label="Window" value={days()} options={dayOptions} onChange={setDays} />
           </CommandBarGroup>
+          <CommandBarGroup>
+            <Switch
+              label="Auto refresh"
+              checked={isAutoRefresh()}
+              onChange={setIsAutoRefresh}
+            />
+            <Select
+              label="Refresh Interval"
+              value={refreshInterval()}
+              options={[
+                { value: '5', label: 'Every 5s' },
+                { value: '10', label: 'Every 10s' },
+                { value: '30', label: 'Every 30s' },
+                { value: '60', label: 'Every 60s' },
+                { value: '600', label: 'Every 10m' },
+              ]}
+              onChange={setRefreshInterval}
+            />
+            <div class="ui-divider--vertical" />
+            <button
+              class="ui-button dashboard__refresh-button"
+              classList={{ 'ui-button--loading': summary.loading }}
+              type="button"
+              onClick={() => setRefreshKey((k) => k + 1)}
+              disabled={summary.loading}
+              aria-busy={summary.loading}
+            >
+              <RefreshCw />
+              {summary.loading ? 'Refreshing' : 'Refresh'}
+            </button>
+          </CommandBarGroup>
         </CommandBar>
 
         <SummaryStrip items={summaryItems()} />
@@ -193,14 +238,25 @@ export const Dashboard: Component = () => {
               title="Traffic Volume"
               description="Daily request and token totals for the selected window."
               actions={
-                <ChartLegend
-                  items={[
-                    { key: 'requests', label: 'Requests', color: '#2357d8' },
-                    { key: 'tokens', label: 'Tokens', color: '#1f7a45' },
-                  ]}
-                  mutedKeys={hiddenTrafficSeries()}
-                  onToggle={(key) => toggleHiddenKey(setHiddenTrafficSeries, key)}
-                />
+                <div style="display: flex; align-items: center; gap: 16px;">
+                  <ChartLegend
+                    items={[
+                      { key: 'requests', label: 'Requests', color: '#2357d8' },
+                      { key: 'tokens', label: 'Tokens', color: '#1f7a45' },
+                    ]}
+                    mutedKeys={hiddenTrafficSeries()}
+                    onToggle={(key) => toggleHiddenKey(setHiddenTrafficSeries, key)}
+                  />
+                  <Select
+                    label="Scale"
+                    value={trafficVolumeScale()}
+                    options={[
+                      { value: 'linear', label: 'Linear' },
+                      { value: 'log', label: 'Log' },
+                    ]}
+                    onChange={setTrafficVolumeScale}
+                  />
+                </div>
               }
             >
               <TimeSeriesChart
@@ -217,6 +273,7 @@ export const Dashboard: Component = () => {
                 formatLeftValue={(value) => formatInteger.format(Math.round(value))}
                 formatRightValue={(value) => new Intl.NumberFormat('en-US', { notation: 'compact', maximumFractionDigits: 1 }).format(value)}
                 tooltipTitle="Traffic volume"
+                yScaleType={trafficVolumeScale()}
               />
             </Panel>
 
@@ -248,8 +305,8 @@ export const Dashboard: Component = () => {
                 showLegend={false}
                 hiddenKeys={hiddenLatencySeries()}
                 onToggleLegend={(key) => toggleHiddenKey(setHiddenLatencySeries, key)}
-                yLeftLabel="Milliseconds"
-                formatLeftValue={(value) => `${value.toFixed(0)}ms`}
+                yLeftLabel="Latency"
+                formatLeftValue={formatDurationMs}
                 tooltipTitle="Backend latency"
               />
             </Panel>
@@ -276,11 +333,11 @@ export const Dashboard: Component = () => {
             <Panel title="Backend Health" description="Cache readiness, liveness, and sync drift indicators for current backends.">
               <MetaCluster items={cacheStateItems()} />
               <Show
-                when={(summary()?.health.stale_backends.length ?? 0) > 0}
+                when={(currentSummary()?.health.stale_backends.length ?? 0) > 0}
                 fallback={<EmptyState title="No stale backend syncs" description="All active backends synced within the freshness window." />}
               >
                 <div class="dashboard__status-list">
-                  {summary()?.health.stale_backends.map((backend) => (
+                  {currentSummary()?.health.stale_backends.map((backend) => (
                     <div class="dashboard__status-item">
                       <div>
                         <strong>{backend.name}</strong>

client/src/routes/DetailLogs.tsx

@@ -1,9 +1,9 @@
-import { createMemo, createResource, createSignal, Show, type Component } from 'solid-js';
+import { createEffect, createMemo, createResource, createSignal, onCleanup, Show, type Component } from 'solid-js';
 import RefreshCcw from 'lucide-solid/icons/refresh-ccw';
 import { api } from '../api/client';
 import { Layout } from '../components/Layout';
 import type { RequestLog } from '../types';
-import { Button, CommandBar, CommandBarGroup, ConversationTimeline, DataGrid, EmptyState, MetaCluster, PageHeader, Panel, Select, StatusBadge, SummaryStrip, Tabs, TextField, hasRenderableConversation } from '../ui';
+import { Button, CommandBar, CommandBarGroup, ConversationTimeline, DataGrid, EmptyState, MetaCluster, PageHeader, Panel, Select, StatusBadge, SummaryStrip, Tabs, TextField, extractAssistantConversationPreview, hasRenderableConversation } from '../ui';
 
 interface FilterState {
   month: string;
@@ -15,6 +15,7 @@ interface FilterState {
 }
 
 const PAGE_SIZE_OPTIONS = [25, 50, 100];
+const SEARCH_DEBOUNCE_MS = 350;
 
 const emptyFilters = (): FilterState => ({
   month: '',
@@ -25,32 +26,6 @@ const emptyFilters = (): FilterState => ({
   endpoint: '',
 });
 
-function extractAssistantPreview(responseBody?: string): string {
-  if (!responseBody) return '-';
-
-  try {
-    const parsed = JSON.parse(responseBody) as {
-      choices?: Array<{
-        message?: {
-          content?: unknown;
-        };
-      }>;
-    };
-    const content = parsed.choices?.[0]?.message?.content;
-    if (typeof content !== 'string') return '-';
-
-    const normalized = content
-      .replace(/\r/g, '')
-      .replace(/\n+/g, ' ')
-      .trim();
-
-    if (!normalized) return '-';
-    return normalized.length > 50 ? `${normalized.slice(0, 50)}...` : normalized;
-  } catch {
-    return '-';
-  }
-}
-
 function prettyPrint(value?: string): string {
   if (!value) return '';
 
@@ -63,6 +38,7 @@ function prettyPrint(value?: string): string {
 
 export const DetailLogs: Component = () => {
   const [filters, setFilters] = createSignal<FilterState>(emptyFilters());
+  const [searchDraft, setSearchDraft] = createSignal('');
   const [page, setPage] = createSignal(1);
   const [pageSize, setPageSize] = createSignal(25);
   const [selectedLogId, setSelectedLogId] = createSignal<number | null>(null);
@@ -88,9 +64,32 @@ export const DetailLogs: Component = () => {
       })
   );
 
-  const requestPage = createMemo(() => logs());
+  const updateFilter = (key: keyof FilterState, value: string) => {
+    let changed = false;
+    setFilters((current) => {
+      if (current[key] === value) return current;
+      changed = true;
+      return { ...current, [key]: value };
+    });
+
+    if (changed) {
+      setPage(1);
+    }
+  };
+
+  createEffect(() => {
+    const nextQuery = searchDraft();
+    const id = window.setTimeout(() => updateFilter('q', nextQuery), SEARCH_DEBOUNCE_MS);
+    onCleanup(() => window.clearTimeout(id));
+  });
+
+  const requestPage = createMemo(() => (logs.state === 'ready' || logs.state === 'refreshing' ? logs.latest : undefined));
   const requestRows = createMemo(() => requestPage()?.rows ?? []);
   const totalRows = createMemo(() => requestPage()?.total ?? 0);
+  const logsError = createMemo(() => {
+    if (!logs.error) return null;
+    return logs.error instanceof Error ? logs.error.message : 'Failed to load detailed logs.';
+  });
   const pageCount = createMemo(() => Math.max(1, Math.ceil(totalRows() / pageSize())));
   const rangeStart = createMemo(() => (totalRows() === 0 ? 0 : (page() - 1) * pageSize() + 1));
   const rangeEnd = createMemo(() => Math.min(totalRows(), page() * pageSize()));
@@ -140,7 +139,7 @@ export const DetailLogs: Component = () => {
   const assistantPreviewById = createMemo(() => {
     const previews = new Map<number, string>();
     for (const row of requestRows()) {
-      previews.set(row.id, extractAssistantPreview(row.response_body));
+      previews.set(row.id, extractAssistantConversationPreview(row.response_body));
     }
     return previews;
   });
@@ -165,22 +164,29 @@ export const DetailLogs: Component = () => {
   ];
 
   const resetFilters = () => {
+    setSearchDraft('');
     setFilters(emptyFilters());
     setPage(1);
   };
 
-  const updateFilter = (key: keyof FilterState, value: string) => {
-    setFilters((current) => ({ ...current, [key]: value }));
-    setPage(1);
-  };
-
   return (
     <Layout>
       <div class="ui-app-page">
         <PageHeader
           title="Detail Logs"
           description="Inspect verbose request logs with monthly filters, text search, and full request/response payload views."
-          actions={<Button onClick={() => void refetch()}><RefreshCcw />Refresh</Button>}
+          actions={
+            <Button
+              class="detail-logs__refresh-button"
+              classList={{ 'ui-button--loading': logs.loading }}
+              onClick={() => void refetch()}
+              disabled={logs.loading}
+              aria-busy={logs.loading}
+            >
+              <RefreshCcw />
+              {logs.loading ? 'Refreshing' : 'Refresh'}
+            </Button>
+          }
         />
 
         <SummaryStrip
@@ -195,9 +201,9 @@ export const DetailLogs: Component = () => {
           <CommandBarGroup>
             <TextField
               label="Search"
-              value={filters().q}
+              value={searchDraft()}
               placeholder="Search body, headers, models, errors"
-              onInput={(event) => updateFilter('q', event.currentTarget.value)}
+              onInput={(event) => setSearchDraft(event.currentTarget.value)}
             />
             <TextField
               label="Month"
@@ -254,7 +260,8 @@ export const DetailLogs: Component = () => {
                 },
               ]}
               getRowKey={(row) => row.id}
-              loading={logs.loading}
+              loading={logs.loading && requestRows().length === 0}
+              error={logsError()}
               emptyMessage="No detailed logs matched the current filters."
               onRowClick={(row) => setSelectedLogId(row.id)}
               pagination={{
@@ -269,7 +276,7 @@ export const DetailLogs: Component = () => {
                 pageSizeOptions: PAGE_SIZE_OPTIONS,
               }}
             />
-            {!logs.loading && requestRows().length === 0 && (
+            {!logs.loading && !logsError() && requestRows().length === 0 && (
               <EmptyState title="No logs found" description="Try a different month, date, or search term." />
             )}
           </Panel>

client/src/routes/Models.tsx

@@ -41,6 +41,9 @@ export const Models: Component = () => {
   const [overview, { refetch: refetchOverview }] = createResource(() => api.modelCache.getOverview());
   const [backends] = createResource(() => api.backends.getAll());
   const [rules, { refetch: refetchRules }] = createResource(() => api.modelRewrites.getAll());
+  const currentOverview = createMemo(() => overview.state === 'ready' || overview.state === 'refreshing' ? overview.latest : undefined);
+  const currentBackends = createMemo(() => backends.state === 'ready' || backends.state === 'refreshing' ? backends.latest : undefined);
+  const currentRules = createMemo(() => rules.state === 'ready' || rules.state === 'refreshing' ? rules.latest : undefined);
   const [dialogOpen, setDialogOpen] = createSignal(false);
   const [confirmOpen, setConfirmOpen] = createSignal(false);
   const [editingRule, setEditingRule] = createSignal<ModelRewriteRule | null>(null);
@@ -50,7 +53,7 @@ export const Models: Component = () => {
   const [notice, setNotice] = createSignal<{ tone: 'success' | 'danger'; message: string } | null>(null);
   const backendNameById = createMemo(() => {
     const names = new Map<number, string>();
-    for (const backend of backends() ?? []) {
+    for (const backend of currentBackends() ?? []) {
       names.set(backend.id, backend.name);
     }
     return names;
@@ -58,7 +61,7 @@ export const Models: Component = () => {
 
   const getBackendName = (backendId: number) => backendNameById().get(backendId) ?? `Backend ${backendId}`;
   const modelCatalogRows = createMemo(() =>
-    (overview()?.models ?? []).map((entry) => ({
+    (currentOverview()?.models ?? []).map((entry) => ({
       ...entry,
       backend_names: entry.backend_ids.map((backendId) => getBackendName(backendId)).join(', '),
     }))
@@ -145,15 +148,15 @@ export const Models: Component = () => {
       <div class="ui-app-page">
         <PageHeader
           title="Models"
-          description="Inspect cached backend model catalogs and manage global model rewrite rules."
+          description="Inspect cached backend model catalogs and manage chained global model rewrite rules."
           actions={<Button onClick={() => void Promise.all([refetchOverview(), refetchRules()])}>Refresh</Button>}
         />
 
         <SummaryStrip
           items={[
-            { label: 'Catalog Models', value: overview()?.models.length ?? 0, hint: 'Unique models across active backends' },
-            { label: 'Tracked Backends', value: overview()?.backends.length ?? 0, hint: 'Memory cache status by backend' },
-            { label: 'Rewrite Rules', value: rules()?.length ?? 0, hint: 'Global source -> target mappings' },
+            { label: 'Catalog Models', value: currentOverview()?.models.length ?? 0, hint: 'Unique models across active backends' },
+            { label: 'Tracked Backends', value: currentOverview()?.backends.length ?? 0, hint: 'Memory cache status by backend' },
+            { label: 'Rewrite Rules', value: currentRules()?.length ?? 0, hint: 'Global source -> target mappings' },
           ]}
         />
 
@@ -164,11 +167,11 @@ export const Models: Component = () => {
         <div class="ui-section-grid">
           <Panel title="Backend Cache Status" description="Memory-backed backend cache state used by request routing and `/v1/models`.">
             <Show
-              when={(overview()?.backends.length ?? 0) > 0}
+              when={(currentOverview()?.backends.length ?? 0) > 0 || overview.loading}
               fallback={<EmptyState title="No backend cache yet" description="Backend model states appear here after the server has seen active backends." />}
             >
               <DataGrid
-                rows={overview()?.backends ?? []}
+                rows={currentOverview()?.backends ?? []}
                 columns={[
                   {
                     id: 'backend_id',
@@ -182,14 +185,14 @@ export const Models: Component = () => {
                   { id: 'last_error', header: 'Last Error', cell: (item) => <span title={item.last_error ?? '-'}>{item.last_error ?? '-'}</span> },
                 ]}
                 getRowKey={(item) => item.backend_id}
-                loading={overview.loading}
+                loading={overview.loading && (currentOverview()?.backends.length ?? 0) === 0}
               />
             </Show>
           </Panel>
 
           <Panel title="Model Catalog" description="Unique models and the backend names currently advertising each one.">
             <Show
-              when={modelCatalogRows().length > 0}
+              when={modelCatalogRows().length > 0 || overview.loading}
               fallback={<EmptyState title="No cached models yet" description="Model catalog entries appear here after backend model snapshots are available." />}
             >
               <DataGrid
@@ -216,7 +219,7 @@ export const Models: Component = () => {
                   },
                 ]}
                 getRowKey={(item) => item.model_id}
-                loading={overview.loading}
+                loading={overview.loading && modelCatalogRows().length === 0}
               />
             </Show>
           </Panel>
@@ -224,16 +227,16 @@ export const Models: Component = () => {
 
         <Panel
           title="Model Rewrite Rules"
-          description="Force rules always rewrite. Fallback rules rewrite only when the original model has no usable backend."
+          description="Force rules always rewrite and continue through the chain. Fallback rules continue only when the current model has no usable backend."
           actions={<IconButton variant="primary" icon={<Plus />} label="Add Rule" onClick={openCreateDialog} />}
         >
           <div class="ui-stack ui-stack--tight">
             <Show
-              when={(rules()?.length ?? 0) > 0}
+              when={(currentRules()?.length ?? 0) > 0 || rules.loading}
               fallback={<EmptyState title="No rewrite rules" description="Requests currently route using the original model name." />}
             >
               <DataGrid
-                rows={rules() ?? []}
+                rows={currentRules() ?? []}
                 columns={[
                   { id: 'source_model', header: 'Source', cell: (rule) => <span>{rule.source_model}</span> },
                   { id: 'target_model', header: 'Target', cell: (rule) => <span>{rule.target_model}</span> },
@@ -242,7 +245,7 @@ export const Models: Component = () => {
                   { id: 'note', header: 'Note', cell: (rule) => <span title={rule.note ?? '-'}>{rule.note ?? '-'}</span> },
                 ]}
                 getRowKey={(rule) => rule.id}
-                loading={rules.loading}
+                loading={rules.loading && (currentRules()?.length ?? 0) === 0}
                 rowActions={(rule) => (
                   <div class="ui-row-actions">
                     <IconButton icon={<Pencil />} label="Edit" onClick={() => openEditDialog(rule)} />
@@ -266,7 +269,7 @@ export const Models: Component = () => {
           open={dialogOpen()}
           onOpenChange={setDialogOpen}
           title={editingRule() ? 'Edit Model Rule' : 'Add Model Rule'}
-          description="Choose whether the target model should always replace the source, or only act as a fallback when the source is unavailable."
+          description="Choose whether the target model should always replace the source, or only continue the chain when the current model is unavailable."
           footer={
             <>
               <Button onClick={() => setDialogOpen(false)} disabled={submitting()}>Cancel</Button>
@@ -282,7 +285,7 @@ export const Models: Component = () => {
             <TextField label="Note" value={form().note} onInput={(event) => setForm((current) => ({ ...current, note: event.currentTarget.value }))} />
             <Checkbox
               label="Always force rewrite"
-              description="When enabled, requests always route to the target model. When disabled, the target model is only used as a fallback."
+              description="When enabled, requests always continue to the target model. When disabled, the target is used only if the current model has no backend."
               checked={form().force}
               onChange={(checked) => setForm((current) => ({ ...current, force: checked }))}
             />

client/src/routes/Scripts.tsx

@@ -107,6 +107,9 @@ export const Scripts: Component = () => {
   const [scripts, { refetch: refetchScripts }] = createResource(() => api.scripts.getAll());
   const [users, { refetch: refetchUsers }] = createResource(() => api.users.getAll());
   const [backends, { refetch: refetchBackends }] = createResource(() => api.backends.getAll());
+  const currentScripts = createMemo(() => scripts.state === 'ready' || scripts.state === 'refreshing' ? scripts.latest : undefined);
+  const currentUsers = createMemo(() => users.state === 'ready' || users.state === 'refreshing' ? users.latest : undefined);
+  const currentBackends = createMemo(() => backends.state === 'ready' || backends.state === 'refreshing' ? backends.latest : undefined);
   const [form, setForm] = createSignal<ScriptFormState>(emptyForm());
   const [selectedScriptId, setSelectedScriptId] = createSignal<number | null>(null);
   const [pendingDeleteScript, setPendingDeleteScript] = createSignal<UserScript | null>(null);
@@ -116,11 +119,11 @@ export const Scripts: Component = () => {
   const [testResult, setTestResult] = createSignal<{ success: boolean; error?: string; executionTime?: number } | null>(null);
   const [testing, setTesting] = createSignal(false);
 
-  const userOptions = createMemo(() => (users() ?? []).map((user) => ({ value: String(user.id), label: user.name })));
-  const backendOptions = createMemo(() => (backends() ?? []).map((backend) => ({ value: String(backend.id), label: backend.name })));
+  const userOptions = createMemo(() => (currentUsers() ?? []).map((user) => ({ value: String(user.id), label: user.name })));
+  const backendOptions = createMemo(() => (currentBackends() ?? []).map((backend) => ({ value: String(backend.id), label: backend.name })));
 
-  const activeCount = createMemo(() => (scripts() ?? []).filter((script) => script.is_active).length);
-  const selectedScript = createMemo(() => (scripts() ?? []).find((script) => script.id === selectedScriptId()) ?? null);
+  const activeCount = createMemo(() => (currentScripts() ?? []).filter((script) => script.is_active).length);
+  const selectedScript = createMemo(() => (currentScripts() ?? []).find((script) => script.id === selectedScriptId()) ?? null);
 
   const syncForm = (script?: UserScript | null) => {
     if (!script) {
@@ -144,8 +147,8 @@ export const Scripts: Component = () => {
   };
 
   const getTargetLabel = (script: Pick<UserScript, 'script_type' | 'target_user_id' | 'target_backend_id'>) => {
-    const user = (users() ?? []).find((item) => item.id === script.target_user_id);
-    const backend = (backends() ?? []).find((item) => item.id === script.target_backend_id);
+    const user = (currentUsers() ?? []).find((item) => item.id === script.target_user_id);
+    const backend = (currentBackends() ?? []).find((item) => item.id === script.target_backend_id);
 
     if (script.script_type === 'per-user-backend') {
       return {
@@ -275,8 +278,8 @@ export const Scripts: Component = () => {
     setTestResult(null);
     try {
       const result = await api.scripts.test(current.id, {
-        user: users()?.[0] || undefined,
-        backend: backends()?.[0] || undefined,
+        user: currentUsers()?.[0] || undefined,
+        backend: currentBackends()?.[0] || undefined,
         request: {
           method: 'POST',
           path: '/v1/chat/completions',
@@ -329,11 +332,11 @@ export const Scripts: Component = () => {
             bodyClass="ui-stack ui-stack--tight"
           >
             <Show
-              when={!scripts.loading || (scripts()?.length ?? 0) > 0}
+              when={!scripts.loading || (currentScripts()?.length ?? 0) > 0}
               fallback={<EmptyState title="Loading scripts" description="Reading middleware definitions and target mappings." />}
             >
               <Show
-                when={(scripts()?.length ?? 0) > 0}
+                when={(currentScripts()?.length ?? 0) > 0}
                 fallback={
                   <EmptyState
                     title="No scripts yet"
@@ -345,7 +348,7 @@ export const Scripts: Component = () => {
                 }
               >
                 <DataGrid
-                  rows={scripts() ?? []}
+                  rows={currentScripts() ?? []}
                   columns={[
                     {
                       id: 'name',
@@ -377,7 +380,7 @@ export const Scripts: Component = () => {
                     },
                   ]}
                   getRowKey={(script) => script.id}
-                  loading={scripts.loading}
+                  loading={scripts.loading && (currentScripts()?.length ?? 0) === 0}
                   onRowClick={(script) => syncForm(script)}
                   rowActions={(script) => (
                     <div class="ui-row-actions">

client/src/routes/Users.tsx

@@ -38,6 +38,7 @@ interface UserFormState {
   api_key: string;
   is_active: boolean;
   detail_logging: boolean;
+  copy_reasoning_to_reasoning_content: boolean;
 }
 
 const emptyForm = (): UserFormState => ({
@@ -46,6 +47,7 @@ const emptyForm = (): UserFormState => ({
   api_key: '',
   is_active: true,
   detail_logging: false,
+  copy_reasoning_to_reasoning_content: false,
 });
 
 const maskApiKey = (apiKey: string) => `${apiKey.slice(0, 5)}...`;
@@ -54,6 +56,9 @@ export const Users: Component = () => {
   const [users, { refetch: refetchUsers }] = createResource(() => api.users.getAll());
   const [backends] = createResource(() => api.backends.getAll());
   const [permissions, { refetch: refetchPermissions }] = createResource(() => api.permissions.getAll());
+  const currentUsers = createMemo(() => users.state === 'ready' || users.state === 'refreshing' ? users.latest : undefined);
+  const currentBackends = createMemo(() => backends.state === 'ready' || backends.state === 'refreshing' ? backends.latest : undefined);
+  const currentPermissions = createMemo(() => permissions.state === 'ready' || permissions.state === 'refreshing' ? permissions.latest : undefined);
   const [query, setQuery] = createSignal('');
   const [dialogOpen, setDialogOpen] = createSignal(false);
   const [userDeleteConfirmOpen, setUserDeleteConfirmOpen] = createSignal(false);
@@ -70,7 +75,7 @@ export const Users: Component = () => {
 
   const filteredUsers = createMemo(() => {
     const value = query().trim().toLowerCase();
-    const list = users() ?? [];
+    const list = currentUsers() ?? [];
     if (!value) return list;
     return list.filter((user) => {
       const haystack = [user.name, user.email ?? '', user.api_key].join(' ').toLowerCase();
@@ -78,29 +83,29 @@ export const Users: Component = () => {
     });
   });
 
-  const activeCount = createMemo(() => (users() ?? []).filter((user) => user.is_active).length);
-  const selectedUser = createMemo(() => (users() ?? []).find((user) => user.id === selectedUserId()) ?? null);
+  const activeCount = createMemo(() => (currentUsers() ?? []).filter((user) => user.is_active).length);
+  const selectedUser = createMemo(() => (currentUsers() ?? []).find((user) => user.id === selectedUserId()) ?? null);
   const permissionsForSelectedUser = createMemo(() => {
     const currentUserId = selectedUserId();
     if (!currentUserId) return [];
-    return (permissions() ?? []).filter((permission) => permission.user_id === currentUserId);
+    return (currentPermissions() ?? []).filter((permission) => permission.user_id === currentUserId);
   });
   const assignedBackendIds = createMemo(() => new Set(permissionsForSelectedUser().map((permission) => permission.backend_id)));
   const availableBackendOptions = createMemo(() =>
-    (backends() ?? [])
+    (currentBackends() ?? [])
       .filter((backend) => !assignedBackendIds().has(backend.id))
       .map((backend) => ({ value: String(backend.id), label: backend.name }))
   );
   const backendNameById = createMemo(() => {
     const names = new Map<number, string>();
-    for (const backend of backends() ?? []) {
+    for (const backend of currentBackends() ?? []) {
       names.set(backend.id, backend.name);
     }
     return names;
   });
 
   createEffect(() => {
-    const list = users() ?? [];
+    const list = currentUsers() ?? [];
     const currentSelectedUserId = selectedUserId();
 
     if (list.length === 0) {
@@ -129,6 +134,7 @@ export const Users: Component = () => {
       api_key: user.api_key,
       is_active: user.is_active,
       detail_logging: user.detail_logging,
+      copy_reasoning_to_reasoning_content: user.copy_reasoning_to_reasoning_content,
     });
     setDialogOpen(true);
   };
@@ -151,6 +157,7 @@ export const Users: Component = () => {
           api_key: current.api_key.trim() || undefined,
           is_active: current.is_active,
           detail_logging: current.detail_logging,
+          copy_reasoning_to_reasoning_content: current.copy_reasoning_to_reasoning_content,
         });
         setNotice({ tone: 'success', message: 'User updated.' });
       } else {
@@ -159,6 +166,7 @@ export const Users: Component = () => {
           email: current.email.trim() || undefined,
           api_key: current.api_key.trim() || undefined,
           detail_logging: current.detail_logging,
+          copy_reasoning_to_reasoning_content: current.copy_reasoning_to_reasoning_content,
         });
         setNotice({ tone: 'success', message: 'User created.' });
       }
@@ -362,6 +370,11 @@ export const Users: Component = () => {
                       header: 'Detail Log',
                       cell: (user) => <StatusBadge tone={user.detail_logging ? 'warning' : 'neutral'}>{user.detail_logging ? 'On' : 'Off'}</StatusBadge>,
                     },
+                    {
+                      id: 'reasoning_compat',
+                      header: 'Reasoning Compat',
+                      cell: (user) => <StatusBadge tone={user.copy_reasoning_to_reasoning_content ? 'success' : 'neutral'}>{user.copy_reasoning_to_reasoning_content ? 'On' : 'Off'}</StatusBadge>,
+                    },
                     {
                       id: 'status',
                       header: 'Status',
@@ -369,7 +382,7 @@ export const Users: Component = () => {
                     },
                   ]}
                   getRowKey={(user) => user.id}
-                  loading={users.loading}
+                  loading={users.loading && filteredUsers().length === 0}
                   emptyMessage="No users match the current search."
                   onRowClick={(user) => setSelectedUserId(user.id)}
                   rowActions={(user) => (
@@ -455,7 +468,7 @@ export const Users: Component = () => {
                           },
                         ]}
                         getRowKey={(permission) => `${permission.user_id}-${permission.backend_id}`}
-                        loading={permissions.loading || backends.loading}
+                        loading={(permissions.loading || backends.loading) && permissionsForSelectedUser().length === 0}
                         rowActions={(permission) => (
                           <IconButton
                             variant="danger"
@@ -526,6 +539,12 @@ export const Users: Component = () => {
               checked={form().detail_logging}
               onChange={(checked) => setForm((current) => ({ ...current, detail_logging: checked }))}
             />
+            <Checkbox
+              label="Copy reasoning to reasoning_content"
+              description="Enable for clients that only display thinking from reasoning_content."
+              checked={form().copy_reasoning_to_reasoning_content}
+              onChange={(checked) => setForm((current) => ({ ...current, copy_reasoning_to_reasoning_content: checked }))}
+            />
           </form>
         </FormDialog>
 

client/src/types/index.ts

@@ -5,6 +5,7 @@ export type User = {
   email?: string;
   is_active: boolean;
   detail_logging: boolean;
+  copy_reasoning_to_reasoning_content: boolean;
   created_at: string;
   updated_at: string;
 };

client/src/ui/lib/format.ts

@@ -0,0 +1,21 @@
+const durationFormatters = {
+  seconds: new Intl.NumberFormat('en-US', { maximumFractionDigits: 1 }),
+  minutes: new Intl.NumberFormat('en-US', { maximumFractionDigits: 1 }),
+};
+
+export function formatDurationMs(value: number): string {
+  if (!Number.isFinite(value)) {
+    return '0ms';
+  }
+
+  const absoluteValue = Math.abs(value);
+  if (absoluteValue < 1000) {
+    return `${Math.round(value)}ms`;
+  }
+
+  if (absoluteValue < 60_000) {
+    return `${durationFormatters.seconds.format(value / 1000)}s`;
+  }
+
+  return `${durationFormatters.minutes.format(value / 60_000)}m`;
+}

client/src/ui/patterns/Charts.tsx

@@ -99,10 +99,71 @@ function formatCompactNumber(value: number): string {
   return new Intl.NumberFormat('en-US', { notation: 'compact', maximumFractionDigits: 1 }).format(value);
 }
 
+function formatNumberWithUnit(value: number, unit?: string): string {
+  const formatted = formatCompactNumber(value);
+  return unit ? `${formatted} ${unit}` : formatted;
+}
+
 function formatPercent(value: number): string {
   return `${value.toFixed(1)}%`;
 }
 
+function getSymlogMax(value: number): number {
+  return value <= 0 ? 1 : value * 1.1;
+}
+
+function roundSymlogTick(value: number, max: number): number {
+  if (value <= 0) {
+    return 0;
+  }
+
+  if (value < 10) {
+    return Math.min(max, Math.max(1, Math.round(value)));
+  }
+
+  const power = 10 ** Math.floor(Math.log10(value));
+  const normalized = value / power;
+  const niceNormalized = normalized <= 1 ? 1 : normalized <= 2 ? 2 : normalized <= 5 ? 5 : 10;
+  return Math.min(max, niceNormalized * power);
+}
+
+function getSymlogTicks(maxValue: number, intervals: number = 4): number[] {
+  const max = Math.max(1, Math.round(maxValue));
+  const transformedMax = Math.log1p(max);
+  const safeIntervals = Math.max(2, intervals);
+  const ticks = new Set<number>();
+
+  for (let index = 0; index <= safeIntervals; index += 1) {
+    const rawTick = index === safeIntervals ? max : Math.expm1((transformedMax / safeIntervals) * index);
+    ticks.add(roundSymlogTick(rawTick, max));
+  }
+
+  return Array.from(ticks).sort((left, right) => left - right);
+}
+
+function getSymlogTicksInRange(minValue: number, maxValue: number, intervals: number = 4): number[] {
+  const min = Math.max(0, minValue);
+  const max = Math.max(min + 1, Math.round(maxValue));
+  const transformedMin = Math.log1p(min);
+  const transformedMax = Math.log1p(max);
+  const safeIntervals = Math.max(2, intervals);
+  const ticks = new Set<number>();
+
+  for (let index = 0; index <= safeIntervals; index += 1) {
+    const rawTick = index === 0
+      ? min
+      : index === safeIntervals
+        ? max
+        : Math.expm1(transformedMin + ((transformedMax - transformedMin) / safeIntervals) * index);
+    const tick = roundSymlogTick(rawTick, max);
+    if (tick >= min && tick <= max) {
+      ticks.add(tick);
+    }
+  }
+
+  return Array.from(ticks).sort((left, right) => left - right);
+}
+
 function getDateTicks(values: Date[], width: number): Date[] {
   if (values.length <= 7) {
     return values;
@@ -115,6 +176,13 @@ function getDateTicks(values: Date[], width: number): Date[] {
   return scale.ticks(Math.max(2, Math.floor(width / 120)));
 }
 
+function getSvgPointerX(event: PointerEvent): number {
+  const target = event.currentTarget as SVGGraphicsElement;
+  const container = target.ownerSVGElement ?? target;
+  const [x] = d3.pointer(event, container);
+  return x;
+}
+
 interface TimeSeriesChartSeries {
   key: string;
   label: string;
@@ -146,6 +214,7 @@ interface TimeSeriesChartProps {
   formatLeftValue?: (value: number) => string;
   formatRightValue?: (value: number) => string;
   tooltipTitle?: string;
+  yScaleType?: 'linear' | 'log';
 }
 
 interface ChartLegendProps {
@@ -183,6 +252,7 @@ export function ChartLegend(props: ChartLegendProps) {
 type ParsedTimeSeriesDatum = TimeSeriesDatum & { parsedDate: Date };
 type ParsedComboDatum = { date: string; lineValue: number; barValue: number; parsedDate: Date };
 type ParsedBoxPlotDatum = { date: string; min: number; q1: number; median: number; q3: number; max: number; parsedDate: Date };
+type HistogramBin = { bin_start: number; bin_end: number; count: number };
 
 export function TimeSeriesChart(props: TimeSeriesChartProps) {
   const env = createChartEnvironment();
@@ -218,32 +288,82 @@ export function TimeSeriesChart(props: TimeSeriesChartProps) {
   const leftSeries = createMemo(() => visibleSeries().filter((series) => series.axis !== 'right'));
   const rightSeries = createMemo(() => visibleSeries().filter((series) => series.axis === 'right'));
 
+  const isLogScale = () => props.yScaleType === 'log';
+
   const leftScale = createMemo(() => {
-    const maxValue =
-      d3.max(points(), (point: ParsedTimeSeriesDatum) =>
-        d3.max(leftSeries(), (series: TimeSeriesChartSeries) => Number(point[series.key] ?? 0))
-      ) ?? 0;
-    return d3.scaleLinear().domain([0, maxValue === 0 ? 1 : maxValue * 1.1]).nice().range([dimensions().marginTop + getInnerHeight(dimensions()), dimensions().marginTop]);
+    const values = points().flatMap((point: ParsedTimeSeriesDatum) =>
+      leftSeries().map((series: TimeSeriesChartSeries) => Number(point[series.key] ?? 0))
+    );
+    const maxValue = d3.max(values) ?? 0;
+    const range = [dimensions().marginTop + getInnerHeight(dimensions()), dimensions().marginTop];
+
+    if (isLogScale()) {
+      const positiveValues = values.filter((v) => v > 0);
+      const minLog = positiveValues.length > 0 ? d3.min(positiveValues) ?? 1 : 1;
+      const maxLog = maxValue === 0 ? 10 : maxValue * 1.1;
+      return d3.scaleLog().domain([minLog, maxLog]).range(range);
+    }
+
+    return d3.scaleLinear().domain([0, maxValue === 0 ? 1 : maxValue * 1.1]).nice().range(range);
   });
 
   const rightScale = createMemo(() => {
-    const maxValue =
-      d3.max(points(), (point: ParsedTimeSeriesDatum) =>
-        d3.max(rightSeries(), (series: TimeSeriesChartSeries) => Number(point[series.key] ?? 0))
-      ) ?? 0;
-    return d3.scaleLinear().domain([0, maxValue === 0 ? 1 : maxValue * 1.1]).nice().range([dimensions().marginTop + getInnerHeight(dimensions()), dimensions().marginTop]);
+    const values = points().flatMap((point: ParsedTimeSeriesDatum) =>
+      rightSeries().map((series: TimeSeriesChartSeries) => Number(point[series.key] ?? 0))
+    );
+    const maxValue = d3.max(values) ?? 0;
+    const range = [dimensions().marginTop + getInnerHeight(dimensions()), dimensions().marginTop];
+
+    if (isLogScale()) {
+      const positiveValues = values.filter((v) => v > 0);
+      const minLog = positiveValues.length > 0 ? d3.min(positiveValues) ?? 1 : 1;
+      const maxLog = maxValue === 0 ? 10 : maxValue * 1.1;
+      return d3.scaleLog().domain([minLog, maxLog]).range(range);
+    }
+
+    return d3.scaleLinear().domain([0, maxValue === 0 ? 1 : maxValue * 1.1]).nice().range(range);
   });
 
-  const leftTicks = createMemo(() => leftScale().ticks(4));
-  const rightTicks = createMemo(() => rightSeries().length > 0 ? rightScale().ticks(4) : []);
+  const leftTicks = createMemo(() => {
+    if (isLogScale()) {
+      const maxValue = d3.max(points(), (point: ParsedTimeSeriesDatum) =>
+        d3.max(leftSeries(), (series: TimeSeriesChartSeries) => Number(point[series.key] ?? 0))
+      ) ?? 0;
+      return getSymlogTicks(maxValue, 8).filter((tick) => tick > 0);
+    }
+    return leftScale().ticks(4);
+  });
+  const rightTicks = createMemo(() => {
+    if (rightSeries().length === 0) {
+      return [];
+    }
+    if (isLogScale()) {
+      const maxValue = d3.max(points(), (point: ParsedTimeSeriesDatum) =>
+        d3.max(rightSeries(), (series: TimeSeriesChartSeries) => Number(point[series.key] ?? 0))
+      ) ?? 0;
+      return getSymlogTicks(maxValue, 8).filter((tick) => tick > 0);
+    }
+    return rightScale().ticks(4);
+  });
   const xTicks = createMemo(() => xScale().ticks(Math.max(2, Math.floor(getInnerWidth(dimensions()) / 120))));
   const dateTicks = createMemo(() => getDateTicks(points().map((point) => point.parsedDate), getInnerWidth(dimensions())));
 
-  const linePath = (series: TimeSeriesChartSeries) =>
-    d3.line()
-      .defined((point: ParsedTimeSeriesDatum) => typeof point[series.key] === 'number')
+  const linePath = (series: TimeSeriesChartSeries) => {
+    const scale = series.axis === 'right' ? rightScale() : leftScale();
+    return d3.line()
+      .defined((point: ParsedTimeSeriesDatum) => {
+        const value = Number(point[series.key] ?? 0);
+        if (!Number.isFinite(value)) {
+          return false;
+        }
+        if (isLogScale() && value <= 0) {
+          return false;
+        }
+        return true;
+      })
       .x((point: ParsedTimeSeriesDatum) => xScale()(point.parsedDate))
-      .y((point: ParsedTimeSeriesDatum) => (series.axis === 'right' ? rightScale() : leftScale())(Number(point[series.key] ?? 0)))(points()) ?? '';
+      .y((point: ParsedTimeSeriesDatum) => scale(Number(point[series.key] ?? 0)))(points()) ?? '';
+  };
 
   const hoveredPoint = createMemo(() => {
     const index = hoverIndex();
@@ -257,7 +377,15 @@ export function TimeSeriesChart(props: TimeSeriesChartProps) {
             ...series,
             value: Number(hoveredPoint()?.[series.key] ?? 0),
           }))
-          .filter((series) => Number.isFinite(series.value))
+          .filter((series) => {
+            if (!Number.isFinite(series.value)) {
+              return false;
+            }
+            if (isLogScale() && series.value <= 0) {
+              return false;
+            }
+            return true;
+          })
       : []
   );
 
@@ -279,9 +407,7 @@ export function TimeSeriesChart(props: TimeSeriesChartProps) {
   };
 
   const handlePointerMove = (event: PointerEvent) => {
-    const rect = (event.currentTarget as SVGRectElement).getBoundingClientRect();
-    const offsetX = event.clientX - rect.left;
-    const hoveredDate = xScale().invert(offsetX);
+    const hoveredDate = xScale().invert(getSvgPointerX(event));
     const nearestIndex = d3.leastIndex(points(), (point: ParsedTimeSeriesDatum) => Math.abs(point.parsedDate.getTime() - hoveredDate.getTime()));
     setHoverIndex(nearestIndex ?? null);
   };
@@ -486,9 +612,7 @@ export function ComboChart(props: ComboChartProps) {
   });
 
   const handlePointerMove = (event: PointerEvent) => {
-    const rect = (event.currentTarget as SVGRectElement).getBoundingClientRect();
-    const offsetX = event.clientX - rect.left;
-    const hoveredDate = xScale().invert(offsetX);
+    const hoveredDate = xScale().invert(getSvgPointerX(event));
     const nearestIndex = d3.leastIndex(points(), (point: ParsedComboDatum) => Math.abs(point.parsedDate.getTime() - hoveredDate.getTime()));
     setHoverIndex(nearestIndex ?? null);
   };
@@ -628,34 +752,69 @@ export function ComboChart(props: ComboChartProps) {
 }
 
 interface HistogramChartProps {
-  data: Array<{ bin_start: number; bin_end: number; count: number }>;
+  data: HistogramBin[];
   height?: number;
+  xTickUnit?: string;
+  yTickUnit?: string;
 }
 
 export function HistogramChart(props: HistogramChartProps) {
   const env = createChartEnvironment();
+  const [hoverIndex, setHoverIndex] = createSignal<number | null>(null);
   const theme = createMemo(() => {
     env.themeVersion();
     return readChartTheme();
   });
   const dimensions = createMemo(() => buildChartDimensions(env.width(), props.height ?? 200, 20));
+  const xDomain = createMemo(() => {
+    const min = d3.min(props.data, (bin: HistogramBin) => bin.bin_start) ?? 0;
+    const max = d3.max(props.data, (bin: HistogramBin) => bin.bin_end) ?? 1;
+    return {
+      min: Math.max(0, min),
+      max,
+    };
+  });
   const xScale = createMemo(() => {
-    const min = d3.min(props.data, (bin: { bin_start: number; bin_end: number; count: number }) => bin.bin_start) ?? 0;
-    const max = d3.max(props.data, (bin: { bin_start: number; bin_end: number; count: number }) => bin.bin_end) ?? 1;
-    return d3.scaleLinear().domain([min, max]).range([dimensions().marginLeft, dimensions().marginLeft + getInnerWidth(dimensions())]);
+    const domain = xDomain();
+    return d3.scaleSymlog().domain([domain.min, getSymlogMax(domain.max)]).range([dimensions().marginLeft, dimensions().marginLeft + getInnerWidth(dimensions())]);
   });
   const yScale = createMemo(() => {
-    const max = d3.max(props.data, (bin: { bin_start: number; bin_end: number; count: number }) => bin.count) ?? 0;
-    return d3.scaleLinear().domain([0, max === 0 ? 1 : max * 1.1]).range([dimensions().marginTop + getInnerHeight(dimensions()), dimensions().marginTop]);
+    const max = d3.max(props.data, (bin: HistogramBin) => bin.count) ?? 0;
+    return d3.scaleSymlog().domain([0, getSymlogMax(max)]).range([dimensions().marginTop + getInnerHeight(dimensions()), dimensions().marginTop]);
   });
-  const xTicks = createMemo(() => xScale().ticks(Math.max(2, Math.floor(getInnerWidth(dimensions()) / 100))));
+  const yTicks = createMemo(() => getSymlogTicks(d3.max(props.data, (bin: HistogramBin) => bin.count) ?? 0));
+  const xTicks = createMemo(() => {
+    const domain = xDomain();
+    return getSymlogTicksInRange(domain.min, domain.max);
+  });
+  const hoveredBin = createMemo(() => {
+    const index = hoverIndex();
+    return index === null ? null : props.data[index] ?? null;
+  });
+
+  const handlePointerMove = (event: PointerEvent) => {
+    const x = getSvgPointerX(event);
+    const hoveredValue = xScale().invert(x);
+    const index = props.data.findIndex((bin) => hoveredValue >= bin.bin_start && hoveredValue <= bin.bin_end);
+
+    if (index >= 0) {
+      setHoverIndex(index);
+      return;
+    }
+
+    const nearestIndex = d3.leastIndex(props.data, (bin: HistogramBin) => {
+      const center = (xScale()(bin.bin_start) + xScale()(bin.bin_end)) / 2;
+      return Math.abs(center - x);
+    });
+    setHoverIndex(nearestIndex ?? null);
+  };
 
   return (
     <div class="ui-chart">
       <div class="ui-chart__frame" ref={env.rootRef}>
         <Show when={env.width() > 0 && props.data.length > 0} fallback={<div class="ui-chart__empty">No histogram data available.</div>}>
           <svg viewBox={`0 0 ${dimensions().width} ${dimensions().height}`} class="ui-chart__svg" role="img" aria-label="Histogram">
-            <For each={yScale().ticks(4)}>
+            <For each={yTicks()}>
               {(tick) => (
                 <>
                   <line
@@ -667,7 +826,7 @@ export function HistogramChart(props: HistogramChartProps) {
                     stroke-dasharray="3 4"
                   />
                   <text x={dimensions().marginLeft - 8} y={yScale()(tick)} fill={theme().textMuted} text-anchor="end" dominant-baseline="middle" class="ui-chart__tick">
-                    {formatCompactNumber(tick)}
+                    {formatNumberWithUnit(tick, props.yTickUnit)}
                   </text>
                 </>
               )}
@@ -681,7 +840,9 @@ export function HistogramChart(props: HistogramChartProps) {
                   width={Math.max(2, xScale()(bin.bin_end) - xScale()(bin.bin_start) - 2)}
                   height={dimensions().marginTop + getInnerHeight(dimensions()) - yScale()(bin.count)}
                   fill={theme().warning}
-                  opacity="0.8"
+                  opacity={hoveredBin() === bin ? '0.95' : '0.8'}
+                  stroke={hoveredBin() === bin ? theme().text : 'transparent'}
+                  stroke-width={hoveredBin() === bin ? '1.5' : '0'}
                   rx="2"
                 />
               )}
@@ -696,11 +857,40 @@ export function HistogramChart(props: HistogramChartProps) {
                   text-anchor="middle"
                   class="ui-chart__tick"
                 >
-                  {formatCompactNumber(tick)}
+                  {formatNumberWithUnit(tick, props.xTickUnit)}
                 </text>
               )}
             </For>
+            <rect
+              x={dimensions().marginLeft}
+              y={dimensions().marginTop}
+              width={getInnerWidth(dimensions())}
+              height={getInnerHeight(dimensions())}
+              fill="transparent"
+              onPointerMove={handlePointerMove}
+              onPointerLeave={() => setHoverIndex(null)}
+            />
           </svg>
+
+          <Show when={hoveredBin()}>
+            {(bin) => (
+              <div class="ui-chart__tooltip">
+                <div class="ui-chart__tooltip-title">Completion tokens</div>
+                <div class="ui-chart__tooltip-row">
+                  <span class="ui-chart__legend-swatch" style={{ background: theme().warning }} />
+                  <span>Range</span>
+                  <strong>
+                    {formatNumberWithUnit(bin().bin_start, props.xTickUnit)} - {formatNumberWithUnit(bin().bin_end, props.xTickUnit)}
+                  </strong>
+                </div>
+                <div class="ui-chart__tooltip-row">
+                  <span class="ui-chart__legend-swatch" style={{ background: theme().accent }} />
+                  <span>Requests</span>
+                  <strong>{formatNumberWithUnit(bin().count, props.yTickUnit)}</strong>
+                </div>
+              </div>
+            )}
+          </Show>
         </Show>
       </div>
     </div>
@@ -714,6 +904,7 @@ interface BoxPlotChartProps {
 
 export function BoxPlotChart(props: BoxPlotChartProps) {
   const env = createChartEnvironment();
+  const [hoverIndex, setHoverIndex] = createSignal<number | null>(null);
   const theme = createMemo(() => {
     env.themeVersion();
     return readChartTheme();
@@ -731,15 +922,29 @@ export function BoxPlotChart(props: BoxPlotChartProps) {
   });
   const yScale = createMemo(() => {
     const max = d3.max(points(), (point: ParsedBoxPlotDatum) => point.max) ?? 0;
-    return d3.scaleLinear().domain([0, max === 0 ? 1 : max * 1.1]).range([dimensions().marginTop + getInnerHeight(dimensions()), dimensions().marginTop]);
+    return d3.scaleSymlog().domain([0, getSymlogMax(max)]).range([dimensions().marginTop + getInnerHeight(dimensions()), dimensions().marginTop]);
   });
+  const yTicks = createMemo(() => getSymlogTicks(d3.max(points(), (point: ParsedBoxPlotDatum) => point.max) ?? 0));
+  const hoveredPoint = createMemo(() => {
+    const index = hoverIndex();
+    return index === null ? null : points()[index] ?? null;
+  });
+
+  const handlePointerMove = (event: PointerEvent) => {
+    const x = getSvgPointerX(event);
+    const nearestIndex = d3.leastIndex(points(), (point: ParsedBoxPlotDatum) => {
+      const center = (xScale()(point.date) ?? dimensions().marginLeft) + xScale().bandwidth() / 2;
+      return Math.abs(center - x);
+    });
+    setHoverIndex(nearestIndex ?? null);
+  };
 
   return (
     <div class="ui-chart">
       <div class="ui-chart__frame" ref={env.rootRef}>
         <Show when={env.width() > 0 && points().length > 0} fallback={<div class="ui-chart__empty">No box plot data available.</div>}>
           <svg viewBox={`0 0 ${dimensions().width} ${dimensions().height}`} class="ui-chart__svg" role="img" aria-label="Box plot chart">
-            <For each={yScale().ticks(4)}>
+            <For each={yTicks()}>
               {(tick) => (
                 <>
                   <line
@@ -769,8 +974,9 @@ export function BoxPlotChart(props: BoxPlotChartProps) {
                       width={xScale().bandwidth()}
                       height={Math.max(2, yScale()(point.q1) - yScale()(point.q3))}
                       fill={theme().accent}
-                      opacity="0.25"
+                      opacity={hoveredPoint() === point ? '0.4' : '0.25'}
                       stroke={theme().accent}
+                      stroke-width={hoveredPoint() === point ? '2' : '1'}
                     />
                     <line x1={xScale()(point.date)} x2={(xScale()(point.date) ?? 0) + xScale().bandwidth()} y1={yScale()(point.median)} y2={yScale()(point.median)} stroke={theme().accent} stroke-width="2" />
                     <text
@@ -786,7 +992,61 @@ export function BoxPlotChart(props: BoxPlotChartProps) {
                 );
               }}
             </For>
+            <Show when={hoveredPoint()}>
+              {(point) => (
+                <line
+                  x1={(xScale()(point().date) ?? dimensions().marginLeft) + xScale().bandwidth() / 2}
+                  x2={(xScale()(point().date) ?? dimensions().marginLeft) + xScale().bandwidth() / 2}
+                  y1={dimensions().marginTop}
+                  y2={dimensions().marginTop + getInnerHeight(dimensions())}
+                  stroke={theme().textMuted}
+                  stroke-dasharray="4 4"
+                />
+              )}
+            </Show>
+            <rect
+              x={dimensions().marginLeft}
+              y={dimensions().marginTop}
+              width={getInnerWidth(dimensions())}
+              height={getInnerHeight(dimensions())}
+              fill="transparent"
+              onPointerMove={handlePointerMove}
+              onPointerLeave={() => setHoverIndex(null)}
+            />
           </svg>
+
+          <Show when={hoveredPoint()}>
+            {(point) => (
+              <div class="ui-chart__tooltip">
+                <div class="ui-chart__tooltip-title">{formatDate(point().parsedDate)}</div>
+                <div class="ui-chart__tooltip-row">
+                  <span class="ui-chart__legend-swatch" style={{ background: theme().textMuted }} />
+                  <span>Max</span>
+                  <strong>{formatNumberWithUnit(point().max, 'tok')}</strong>
+                </div>
+                <div class="ui-chart__tooltip-row">
+                  <span class="ui-chart__legend-swatch" style={{ background: theme().accent }} />
+                  <span>Q3</span>
+                  <strong>{formatNumberWithUnit(point().q3, 'tok')}</strong>
+                </div>
+                <div class="ui-chart__tooltip-row">
+                  <span class="ui-chart__legend-swatch" style={{ background: theme().accent }} />
+                  <span>Median</span>
+                  <strong>{formatNumberWithUnit(point().median, 'tok')}</strong>
+                </div>
+                <div class="ui-chart__tooltip-row">
+                  <span class="ui-chart__legend-swatch" style={{ background: theme().accent }} />
+                  <span>Q1</span>
+                  <strong>{formatNumberWithUnit(point().q1, 'tok')}</strong>
+                </div>
+                <div class="ui-chart__tooltip-row">
+                  <span class="ui-chart__legend-swatch" style={{ background: theme().textMuted }} />
+                  <span>Min</span>
+                  <strong>{formatNumberWithUnit(point().min, 'tok')}</strong>
+                </div>
+              </div>
+            )}
+          </Show>
         </Show>
       </div>
     </div>

client/src/ui/patterns/ConversationTimeline.tsx

@@ -3,10 +3,14 @@ import { MetaCluster } from './MetaCluster';
 import { StatusBadge, type StatusTone } from './StatusBadge';
 
 type KnownChatRole = 'system' | 'user' | 'assistant';
+const COMPACT_CHAT_STREAM_FORMAT = 'kyush.chat_stream.compact.v1';
+const RAW_CHAT_STREAM_FORMAT = 'kyush.chat_stream.raw.v1';
 
 interface ParsedMessage {
   role: string;
   content: string;
+  reasoning?: string;
+  toolCalls?: string;
   metadata?: Array<{ key: string; value: string }>;
 }
 
@@ -16,6 +20,74 @@ interface ConversationTimelineProps {
   emptyMessage?: string;
 }
 
+interface ParsedStreamResponse {
+  messages: ParsedMessage[];
+  model?: string;
+  created?: number;
+  usage?: Record<string, unknown>;
+}
+
+interface StreamChoiceState {
+  index: number;
+  role?: string;
+  content: string[];
+  reasoning: string[];
+  toolCalls: Map<number, StreamToolCallState>;
+  finishReason?: string;
+  stopReason?: string;
+}
+
+interface StreamToolCallState {
+  index: number;
+  id?: string;
+  type?: string;
+  function?: {
+    name?: string;
+    arguments?: string;
+  };
+}
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return Boolean(value) && typeof value === 'object' && !Array.isArray(value);
+}
+
+function stringifyValue(value: unknown): string {
+  if (typeof value === 'string') return value;
+  if (value === undefined || value === null) return '';
+
+  try {
+    return JSON.stringify(value);
+  } catch {
+    return String(value);
+  }
+}
+
+function prettyJson(value: unknown): string {
+  try {
+    return JSON.stringify(value, null, 2);
+  } catch {
+    return String(value);
+  }
+}
+
+function hasMeaningfulToolCall(value: unknown): boolean {
+  if (value === undefined || value === null) return false;
+  if (typeof value === 'string') return value.trim().length > 0;
+  if (Array.isArray(value)) return value.some((item) => hasMeaningfulToolCall(item));
+  if (isRecord(value)) {
+    return Object.entries(value).some(([key, item]) => key !== 'index' && hasMeaningfulToolCall(item));
+  }
+  return true;
+}
+
+function normalizeToolCalls(value: unknown): string | undefined {
+  if (!hasMeaningfulToolCall(value)) return undefined;
+
+  const rendered = prettyJson(value).trim();
+  if (!rendered || rendered === '[]' || rendered === '{}') return undefined;
+  return rendered;
+}
+
 function normalizePayload(value: unknown): Record<string, unknown> | null {
   if (!value) return null;
 
@@ -31,6 +103,49 @@ function normalizePayload(value: unknown): Record<string, unknown> | null {
   return typeof value === 'object' ? (value as Record<string, unknown>) : null;
 }
 
+function normalizeCompactStreamResponse(payload: Record<string, unknown> | null): ParsedStreamResponse | null {
+  const compactPayload = payload?.format === RAW_CHAT_STREAM_FORMAT && isRecord(payload.compact)
+    ? payload.compact
+    : payload;
+
+  if (compactPayload?.format !== COMPACT_CHAT_STREAM_FORMAT) return null;
+
+  const rawChoices = compactPayload.choices;
+  const messages = Array.isArray(rawChoices)
+    ? rawChoices
+        .filter((choice): choice is Record<string, unknown> => isRecord(choice))
+        .map((choice) => {
+          const metadata = [
+            choice.finish_reason !== undefined && choice.finish_reason !== null
+              ? { key: 'Finish', value: String(choice.finish_reason) }
+              : null,
+            choice.stop_reason !== undefined && choice.stop_reason !== null
+              ? { key: 'Stop Reason', value: String(choice.stop_reason) }
+              : null,
+            choice.matched_stop !== undefined && choice.matched_stop !== null
+              ? { key: 'Matched Stop', value: String(choice.matched_stop) }
+              : null,
+          ].filter((item): item is { key: string; value: string } => Boolean(item));
+
+          return {
+            role: typeof choice.role === 'string' ? choice.role : 'assistant',
+            content: stringifyValue(choice.content),
+            reasoning: stringifyValue(choice.reasoning).trim() || undefined,
+            toolCalls: normalizeToolCalls(choice.tool_calls),
+            metadata,
+          };
+        })
+        .filter((message) => message.content || message.reasoning || message.toolCalls || (message.metadata?.length ?? 0) > 0)
+    : [];
+
+  return {
+    messages,
+    model: typeof compactPayload.model === 'string' ? compactPayload.model : undefined,
+    created: typeof compactPayload.created === 'number' ? compactPayload.created : undefined,
+    usage: isRecord(compactPayload.usage) ? compactPayload.usage : undefined,
+  };
+}
+
 function normalizeMessages(payload: Record<string, unknown> | null): ParsedMessage[] {
   const messages = payload?.messages;
   if (!Array.isArray(messages)) return [];
@@ -39,7 +154,7 @@ function normalizeMessages(payload: Record<string, unknown> | null): ParsedMessa
     .filter((item): item is Record<string, unknown> => Boolean(item) && typeof item === 'object')
     .map((item) => ({
       role: typeof item.role === 'string' ? item.role : 'unknown',
-      content: typeof item.content === 'string' ? item.content : JSON.stringify(item.content ?? ''),
+      content: stringifyValue(item.content),
     }));
 }
 
@@ -52,25 +167,11 @@ function normalizeAssistantMessages(payload: Record<string, unknown> | null): Pa
       const message = (choice as Record<string, unknown>).message;
       if (!message || typeof message !== 'object') return null;
 
-      const content = typeof (message as Record<string, unknown>).content === 'string'
-        ? String((message as Record<string, unknown>).content)
-        : JSON.stringify((message as Record<string, unknown>).content ?? '');
+      const messageRecord = message as Record<string, unknown>;
+      const content = stringifyValue(messageRecord.content);
+      const reasoning = stringifyValue(messageRecord.reasoning_content ?? messageRecord.reasoning).trim();
+      const toolCalls = normalizeToolCalls(messageRecord.tool_calls);
       const metadata = [
-        (message as Record<string, unknown>).reasoning_content !== undefined
-          ? {
-              key: 'Reasoning',
-              value:
-                typeof (message as Record<string, unknown>).reasoning_content === 'string'
-                  ? String((message as Record<string, unknown>).reasoning_content)
-                  : JSON.stringify((message as Record<string, unknown>).reasoning_content),
-            }
-          : null,
-        (message as Record<string, unknown>).tool_calls !== undefined
-          ? {
-              key: 'Tool Calls',
-              value: JSON.stringify((message as Record<string, unknown>).tool_calls),
-            }
-          : null,
         (choice as Record<string, unknown>).finish_reason !== undefined
           ? { key: 'Finish', value: String((choice as Record<string, unknown>).finish_reason) }
           : null,
@@ -85,6 +186,8 @@ function normalizeAssistantMessages(payload: Record<string, unknown> | null): Pa
       return {
         role: 'assistant' as const,
         content,
+        reasoning: reasoning || undefined,
+        toolCalls,
         metadata,
       };
     });
@@ -92,9 +195,187 @@ function normalizeAssistantMessages(payload: Record<string, unknown> | null): Pa
   return messages.filter((message): message is ParsedMessage => message !== null);
 }
 
+function extractSseJsonPayloads(value: string): Record<string, unknown>[] {
+  const payloads: Record<string, unknown>[] = [];
+  const lines = value.replace(/\r\n/g, '\n').replace(/\r/g, '\n').split('\n');
+  let dataLines: string[] = [];
+
+  const flush = () => {
+    if (dataLines.length === 0) return;
+    const data = dataLines.join('\n');
+    dataLines = [];
+
+    if (data.trim() === '[DONE]') return;
+
+    try {
+      const parsed = JSON.parse(data);
+      if (isRecord(parsed)) payloads.push(parsed);
+    } catch {
+      // Ignore non-JSON SSE data frames.
+    }
+  };
+
+  for (const line of lines) {
+    if (line === '') {
+      flush();
+      continue;
+    }
+
+    if (line.startsWith('data:')) {
+      dataLines.push(line.slice(5).replace(/^ /, ''));
+    }
+  }
+
+  flush();
+  return payloads;
+}
+
+function mergeToolCall(target: StreamToolCallState, chunk: Record<string, unknown>): void {
+  if (typeof chunk.id === 'string') target.id = chunk.id;
+  if (typeof chunk.type === 'string') target.type = chunk.type;
+
+  const functionChunk = chunk.function;
+  if (!isRecord(functionChunk)) return;
+
+  target.function = target.function ?? {};
+  if (typeof functionChunk.name === 'string') {
+    target.function.name = `${target.function.name ?? ''}${functionChunk.name}`;
+  }
+  if (typeof functionChunk.arguments === 'string') {
+    target.function.arguments = `${target.function.arguments ?? ''}${functionChunk.arguments}`;
+  }
+}
+
+function parseStreamResponse(value: unknown): ParsedStreamResponse | null {
+  if (typeof value !== 'string' || !value.includes('data:')) return null;
+
+  const payloads = extractSseJsonPayloads(value);
+  if (payloads.length === 0) return null;
+
+  const choices = new Map<number, StreamChoiceState>();
+  let model: string | undefined;
+  let created: number | undefined;
+  let usage: Record<string, unknown> | undefined;
+
+  const getChoice = (index: number) => {
+    const existing = choices.get(index);
+    if (existing) return existing;
+
+    const createdChoice: StreamChoiceState = {
+      index,
+      content: [],
+      reasoning: [],
+      toolCalls: new Map(),
+    };
+    choices.set(index, createdChoice);
+    return createdChoice;
+  };
+
+  for (const payload of payloads) {
+    if (typeof payload.model === 'string' && !model) model = payload.model;
+    if (typeof payload.created === 'number' && created === undefined) created = payload.created;
+    if (isRecord(payload.usage)) usage = payload.usage;
+
+    if (!Array.isArray(payload.choices)) continue;
+
+    for (const rawChoice of payload.choices) {
+      if (!isRecord(rawChoice)) continue;
+
+      const index = typeof rawChoice.index === 'number' ? rawChoice.index : 0;
+      const choice = getChoice(index);
+      const delta = rawChoice.delta;
+
+      if (isRecord(delta)) {
+        if (typeof delta.role === 'string') choice.role = delta.role;
+        if (typeof delta.content === 'string') choice.content.push(delta.content);
+        const reasoning = typeof delta.reasoning_content === 'string'
+          ? delta.reasoning_content
+          : typeof delta.reasoning === 'string'
+            ? delta.reasoning
+            : undefined;
+        if (reasoning) choice.reasoning.push(reasoning);
+
+        if (Array.isArray(delta.tool_calls)) {
+          for (const rawToolCall of delta.tool_calls) {
+            if (!isRecord(rawToolCall)) continue;
+            const toolIndex = typeof rawToolCall.index === 'number' ? rawToolCall.index : choice.toolCalls.size;
+            const existing = choice.toolCalls.get(toolIndex) ?? { index: toolIndex };
+            mergeToolCall(existing, rawToolCall);
+            choice.toolCalls.set(toolIndex, existing);
+          }
+        }
+      }
+
+      if (rawChoice.finish_reason !== undefined && rawChoice.finish_reason !== null) {
+        choice.finishReason = String(rawChoice.finish_reason);
+      }
+      if (rawChoice.stop_reason !== undefined && rawChoice.stop_reason !== null) {
+        choice.stopReason = String(rawChoice.stop_reason);
+      }
+    }
+  }
+
+  const messages = [...choices.values()]
+    .sort((left, right) => left.index - right.index)
+    .map((choice) => {
+      const toolCalls = [...choice.toolCalls.values()].sort((left, right) => left.index - right.index);
+      const metadata = [
+        choice.finishReason ? { key: 'Finish', value: choice.finishReason } : null,
+        choice.stopReason ? { key: 'Stop Reason', value: choice.stopReason } : null,
+      ].filter((item): item is { key: string; value: string } => Boolean(item));
+
+      return {
+        role: choice.role ?? 'assistant',
+        content: choice.content.join(''),
+        reasoning: choice.reasoning.join('') || undefined,
+        toolCalls: normalizeToolCalls(toolCalls),
+        metadata,
+      };
+    })
+    .filter((message) => message.content || message.reasoning || message.toolCalls || (message.metadata?.length ?? 0) > 0);
+
+  return {
+    messages,
+    model,
+    created,
+    usage,
+  };
+}
+
+function getAssistantMessages(responseBody?: unknown): ParsedMessage[] {
+  const payload = normalizePayload(responseBody);
+  const compactStream = normalizeCompactStreamResponse(payload);
+  if (compactStream) return compactStream.messages;
+
+  const stream = parseStreamResponse(responseBody);
+  if (stream) return stream.messages;
+  return normalizeAssistantMessages(payload);
+}
+
+export function extractAssistantConversationPreview(responseBody?: unknown): string {
+  const assistantMessage = getAssistantMessages(responseBody).find((message) => (
+    message.content.trim() || message.reasoning?.trim() || message.toolCalls?.trim()
+  ));
+
+  if (!assistantMessage) return '-';
+
+  const source = assistantMessage.content.trim()
+    ? assistantMessage.content
+    : assistantMessage.reasoning
+      ? `Thinking: ${assistantMessage.reasoning}`
+      : `Tool Calls: ${assistantMessage.toolCalls ?? ''}`;
+  const normalized = source
+    .replace(/\r/g, '')
+    .replace(/\n+/g, ' ')
+    .trim();
+
+  if (!normalized) return '-';
+  return normalized.length > 50 ? `${normalized.slice(0, 50)}...` : normalized;
+}
+
 export function hasRenderableConversation(requestBody?: unknown, responseBody?: unknown): boolean {
   const requestMessages = normalizeMessages(normalizePayload(requestBody));
-  const responseMessages = normalizeAssistantMessages(normalizePayload(responseBody));
+  const responseMessages = getAssistantMessages(responseBody);
   return requestMessages.length > 0 || responseMessages.length > 0;
 }
 
@@ -121,23 +402,29 @@ function getRoleClass(role: string): string {
 export function ConversationTimeline(props: ConversationTimelineProps) {
   const parsedRequest = createMemo(() => normalizePayload(props.requestBody));
   const parsedResponse = createMemo(() => normalizePayload(props.responseBody));
+  const parsedCompactStreamResponse = createMemo(() => normalizeCompactStreamResponse(parsedResponse()));
+  const parsedStreamResponse = createMemo(() => parseStreamResponse(props.responseBody));
 
   const requestMessages = createMemo(() => normalizeMessages(parsedRequest()));
-  const responseMessages = createMemo(() => normalizeAssistantMessages(parsedResponse()));
+  const responseMessages = createMemo(() => parsedCompactStreamResponse()?.messages ?? parsedStreamResponse()?.messages ?? normalizeAssistantMessages(parsedResponse()));
   const messages = createMemo(() => [...requestMessages(), ...responseMessages()]);
 
   const summaryItems = createMemo(() => {
     const request = parsedRequest();
     const response = parsedResponse();
+    const stream = parsedCompactStreamResponse() ?? parsedStreamResponse();
     const usage = response?.usage && typeof response.usage === 'object' ? response.usage as Record<string, unknown> : null;
+    const responseUsage = usage ?? stream?.usage ?? null;
 
     return [
       typeof request?.model === 'string' ? { key: 'Model', value: request.model } : null,
       request?.temperature !== undefined ? { key: 'Temp', value: String(request.temperature) } : null,
-      typeof response?.created === 'number' ? { key: 'Created', value: String(response.created) } : null,
-      usage?.prompt_tokens !== undefined ? { key: 'Prompt', value: String(usage.prompt_tokens) } : null,
-      usage?.completion_tokens !== undefined ? { key: 'Completion', value: String(usage.completion_tokens) } : null,
-      usage?.total_tokens !== undefined ? { key: 'Total', value: String(usage.total_tokens) } : null,
+      typeof stream?.model === 'string' && stream.model !== request?.model ? { key: 'Response Model', value: stream.model } : null,
+      typeof response?.created === 'number' ? { key: 'Res. Created', value: String(response.created) } : null,
+      // typeof stream?.created === 'number' ? { key: 'Stream Created', value: String(stream.created) } : null,
+      responseUsage?.prompt_tokens !== undefined ? { key: 'Prompt', value: String(responseUsage.prompt_tokens) } : null,
+      responseUsage?.completion_tokens !== undefined ? { key: 'Completion', value: String(responseUsage.completion_tokens) } : null,
+      responseUsage?.total_tokens !== undefined ? { key: 'Total', value: String(responseUsage.total_tokens) } : null,
     ].filter((item): item is { key: string; value: string } => Boolean(item));
   });
 
@@ -160,7 +447,26 @@ export function ConversationTimeline(props: ConversationTimelineProps) {
                   <span class="ui-conversation__turn-index">Turn {index() + 1}</span>
                 </header>
                 <div class="ui-conversation__bubble">
-                  <pre class="ui-conversation__content">{message.content}</pre>
+                  <Show when={message.reasoning}>
+                    <section class="ui-conversation__block ui-conversation__block--reasoning">
+                      <div class="ui-conversation__block-label">Thinking</div>
+                      <pre class="ui-conversation__content">{message.reasoning}</pre>
+                    </section>
+                  </Show>
+                  <Show when={message.content.trim().length > 0 || (!message.reasoning && !message.toolCalls)}>
+                    <section class="ui-conversation__block">
+                      <Show when={message.reasoning}>
+                        <div class="ui-conversation__block-label">Response</div>
+                      </Show>
+                      <pre class="ui-conversation__content">{message.content}</pre>
+                    </section>
+                  </Show>
+                  <Show when={message.toolCalls}>
+                    <section class="ui-conversation__block ui-conversation__block--tool-calls">
+                      <div class="ui-conversation__block-label">Tool Calls</div>
+                      <pre class="ui-conversation__content">{message.toolCalls}</pre>
+                    </section>
+                  </Show>
                   <Show when={message.metadata && message.metadata.length > 0}>
                     <div class="ui-conversation__meta">
                       <MetaCluster items={message.metadata!} />

client/src/ui/styles/base.css

@@ -86,3 +86,10 @@ select {
   height: 1px;
   background: var(--color-border);
 }
+
+.ui-divider--vertical {
+  width: 1px;
+  height: 1.5em;
+  background: var(--color-border);
+  align-self: center;
+}

client/src/ui/styles/patterns.css

@@ -464,6 +464,8 @@
 }
 
 .ui-conversation__bubble {
+  display: grid;
+  gap: var(--space-4);
   padding: var(--space-4);
   border: 1px solid var(--color-border);
   border-radius: var(--radius-2);
@@ -498,8 +500,29 @@
   color: var(--color-text);
 }
 
+.ui-conversation__block {
+  display: grid;
+  gap: var(--space-2);
+  min-width: 0;
+}
+
+.ui-conversation__block--reasoning,
+.ui-conversation__block--tool-calls {
+  padding: var(--space-3);
+  border: 1px solid var(--color-border);
+  border-radius: var(--radius-2);
+  background: var(--color-bg-inset);
+}
+
+.ui-conversation__block-label {
+  color: var(--color-text-soft);
+  font-size: var(--text-1);
+  font-weight: 600;
+  text-transform: uppercase;
+  letter-spacing: 0.06em;
+}
+
 .ui-conversation__meta {
-  margin-top: var(--space-4);
   padding-top: var(--space-4);
   border-top: 1px solid var(--color-border);
 }
@@ -623,6 +646,18 @@
   min-width: 180px;
 }
 
+.dashboard__refresh-button {
+  min-width: 116px;
+}
+
+.detail-logs__refresh-button {
+  min-width: 116px;
+}
+
+.analytics__refresh-button {
+  min-width: 116px;
+}
+
 .analytics__grid--wide {
   grid-template-columns: 1.5fr minmax(340px, 1fr);
 }

client/src/ui/styles/primitives.css

@@ -66,6 +66,10 @@
   cursor: not-allowed;
 }
 
+.ui-button--loading svg {
+  animation: ui-button-spin 0.8s linear infinite;
+}
+
 .ui-field {
   display: grid;
   gap: var(--space-2);
@@ -445,3 +449,15 @@
     padding: var(--space-4);
   }
 }
+
+@media (prefers-reduced-motion: reduce) {
+  .ui-button--loading svg {
+    animation: none;
+  }
+}
+
+@keyframes ui-button-spin {
+  to {
+    transform: rotate(360deg);
+  }
+}

database/analytics-schema.sql

@@ -28,5 +28,8 @@ CREATE TABLE IF NOT EXISTS backend_metrics (
 -- Indexes for performance
 CREATE INDEX IF NOT EXISTS idx_usage_stats_user ON usage_stats(user_id);
 CREATE INDEX IF NOT EXISTS idx_usage_stats_date ON usage_stats(date);
+CREATE INDEX IF NOT EXISTS idx_usage_stats_backend_date ON usage_stats(backend_id, date);
+CREATE INDEX IF NOT EXISTS idx_usage_stats_user_backend_date ON usage_stats(user_id, backend_id, date);
 CREATE INDEX IF NOT EXISTS idx_backend_metrics_backend ON backend_metrics(backend_id);
 CREATE INDEX IF NOT EXISTS idx_backend_metrics_date ON backend_metrics(date);
+CREATE INDEX IF NOT EXISTS idx_backend_metrics_backend_date ON backend_metrics(backend_id, date);

database/request-logs-schema.sql

@@ -23,7 +23,9 @@ CREATE TABLE IF NOT EXISTS request_logs (
 
 CREATE INDEX IF NOT EXISTS idx_request_logs_created_at ON request_logs(created_at);
 CREATE INDEX IF NOT EXISTS idx_request_logs_local_date ON request_logs(local_date);
+CREATE INDEX IF NOT EXISTS idx_request_logs_local_date_backend ON request_logs(local_date, backend_id);
 CREATE INDEX IF NOT EXISTS idx_request_logs_user ON request_logs(user_id);
 CREATE INDEX IF NOT EXISTS idx_request_logs_backend ON request_logs(backend_id);
 CREATE INDEX IF NOT EXISTS idx_request_logs_endpoint ON request_logs(endpoint);
 CREATE INDEX IF NOT EXISTS idx_request_logs_detail_logged ON request_logs(detail_logged);
+CREATE INDEX IF NOT EXISTS idx_request_logs_completion_tokens ON request_logs(completion_tokens);

database/schema.sql

@@ -9,6 +9,7 @@ CREATE TABLE IF NOT EXISTS users (
     email TEXT,
     is_active BOOLEAN DEFAULT 1,
     detail_logging INTEGER NOT NULL DEFAULT 0,
+    copy_reasoning_to_reasoning_content INTEGER NOT NULL DEFAULT 0,
     created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
     updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
 );

docs/analytics.md

@@ -27,4 +27,5 @@
 
 - 모델 추이의 모델 키는 `response_model -> routed_model -> request_model -> unknown` 순서로 결정한다.
 - response length 계열 시각화는 `completion_tokens` 값이 있는 요청만 집계한다.
+- response length 계열 시각화는 긴 꼬리 분포를 읽기 쉽도록 로그 계열 스케일을 사용한다.
 - 상세 요청 단위의 latency/body 확인은 계속 `DetailLogs` 화면에서 담당한다.

docs/api.md

@@ -23,9 +23,12 @@
 `/v1/**`는 기존 사용자 API 키 인증을 유지하며 관리자 인증과 분리된다.
 
 추가 동작:
-- `/v1/chat/completions` 는 요청 모델명을 먼저 전역 rewrite 규칙으로 해석한 뒤, 최종 모델을 서빙하는 허용 가능한 활성 백엔드만 후보로 사용한다
-- `force=true` rewrite 는 항상 적용된다
-- `force=false` rewrite 는 원본 모델을 서빙하는 허용 가능한 활성 백엔드가 없을 때만 fallback 으로 적용된다
+- `/v1/chat/completions` 는 요청 모델명을 먼저 전역 rewrite 체인으로 해석한 뒤, 최종 모델을 서빙하는 허용 가능한 활성 백엔드만 후보로 사용한다
+- 사용자 옵션 `copy_reasoning_to_reasoning_content` 가 켜져 있으면 chat completion 응답의 `reasoning` 필드를 `reasoning_content` 로 추가 복제한다. streaming/non-stream 모두 적용되며 기존 `reasoning_content` 는 덮어쓰지 않는다
+- `force=true` rewrite 는 항상 적용되고 target 모델의 다음 규칙까지 계속 평가한다
+- `force=false` rewrite 는 현재 모델을 서빙하는 허용 가능한 활성 백엔드가 없을 때만 fallback 으로 적용되고 target 모델의 다음 규칙까지 계속 평가한다
+- `/v1/models` 는 native backend 모델뿐 아니라 현재 사용자 권한에서 최종 후보가 있는 rewrite source alias도 함께 반환한다
+- `MODEL_LIST_INCLUDE_ROUTING_METADATA=1|true|yes|on` 이면 `/v1/models` 의 각 model object에 비표준 `kyush_router` metadata를 추가한다. 이 metadata는 `requested_model`, `routed_model`, `was_rewritten`, `rule_type`, `rewrite_path` 를 포함한다
 - 최종 후보가 없으면 모델 미지원 오류를 반환하고 `request_model`, `routed_model` 을 함께 내려준다
 
 ## Admin API
@@ -52,9 +55,9 @@
 | Method | Path | Description |
 |--------|------|-------------|
 | GET | `/admin/users` | 전체 사용자 목록 |
-| POST | `/admin/users` | 사용자 생성 (`api_key` 생략 시 자동 발급, 지정 시 수동 등록) |
+| POST | `/admin/users` | 사용자 생성 (`api_key` 생략 시 자동 발급, 지정 시 수동 등록, copy_reasoning_to_reasoning_content 선택 가능) |
 | GET | `/admin/users/:id` | 사용자 조회 |
-| PUT | `/admin/users/:id` | 사용자 수정 (name, email, api_key, is_active, detail_logging) |
+| PUT | `/admin/users/:id` | 사용자 수정 (name, email, api_key, is_active, detail_logging, copy_reasoning_to_reasoning_content) |
 | DELETE | `/admin/users/:id` | 사용자 삭제 |
 | POST | `/admin/users/:id/regenerate-api-key` | API 키 재발급 |
 
@@ -80,6 +83,8 @@
 | PUT | `/admin/model-rewrites/:id` | 전역 모델 rewrite 규칙 수정 |
 | DELETE | `/admin/model-rewrites/:id` | 전역 모델 rewrite 규칙 삭제 |
 
+활성 rewrite 그래프에 cycle을 만드는 생성/수정 요청은 `409 { error, cycle }` 로 거부된다. 비활성 규칙끼리의 cycle은 저장할 수 있지만 활성화 시점에는 같은 검사를 통과해야 한다.
+
 `GET /admin/backends/:id/models` 응답에는 아래가 함께 포함된다.
 - `backend`: 백엔드 기본 정보 + 캐시 요약
 - `cache`: 메모리 캐시 상태 (`ready`, `uninitialized`, `error`, `inactive`)
@@ -128,6 +133,8 @@
 
 - `model-trends` 는 `response_model -> routed_model -> request_model -> unknown` 순서로 모델 키를 결정한다.
 - response length 계열 endpoint는 `completion_tokens` 가 있는 요청만 집계한다.
+- `response-length-histogram` 은 긴 꼬리 분포를 읽기 쉽도록 로그 간격 bin을 반환한다.
+- stream response body 저장 방식은 `DETAIL_STREAM_LOG_MODE=compact|raw|both|off` 로 제어한다. 기본값 `compact` 는 raw SSE를 저장하지 않고 누적된 thinking/content/tool call/usage JSON을 저장하며, 기존 raw SSE 로그는 관리자 UI에서 계속 파싱된다.
 - 자세한 내용은 [docs/analytics.md](./analytics.md) 참고.
 
 ### Dashboard Summary

docs/client.md

@@ -70,6 +70,7 @@ SPA는 `/dashboard`를 라우터 base로 사용하고, 관리자 API는 계속 `
 - 공통 필터는 기간(`7`, `30`, `90`일)과 backend 선택이다.
 - 상단 summary strip 뒤에 일별 volume, reliability, response time, model trends, response length 분포 패널이 배치된다.
 - 상세 raw request 확인은 계속 `DetailLogs` 화면이 담당한다.
+- `DetailLogs` 의 Conversation 탭은 non-stream completion JSON, 기존 raw SSE stream 문자열, 신규 compact stream JSON(`kyush.chat_stream.compact.v1`)을 모두 파싱한다.
 - 자세한 내용은 [docs/analytics.md](./analytics.md) 참고.
 
 ## Model Management UI
@@ -77,7 +78,13 @@ SPA는 `/dashboard`를 라우터 base로 사용하고, 관리자 API는 계속 `
 - `Backends` 화면은 백엔드별 모델 캐시 상태, 모델 수, 마지막 sync 상태를 표시한다
 - `Backends` 화면에서 활성 백엔드는 수동 refresh 와 캐시된 모델 목록 확인이 가능하다
 - 비활성 백엔드는 모델 조회를 시도하지 않으며 UI에서도 `Skipped` 상태로 표시된다
-- `Models` 화면은 전체 메모리 모델 카탈로그와 전역 모델 rewrite 규칙을 관리한다
+- `Models` 화면은 전체 메모리 모델 카탈로그와 전역 모델 rewrite 체인을 관리한다
 - rewrite 규칙은 2가지 모드를 가진다
-  - `Force`: 원본 모델 사용 가능 여부와 관계없이 항상 target model 로 rewrite
-  - `Fallback`: 원본 모델을 서빙하는 허용 가능한 활성 백엔드가 없을 때만 target model 로 rewrite
+  - `Force`: 현재 모델 사용 가능 여부와 관계없이 항상 target model 로 이동하고 다음 규칙을 계속 평가
+  - `Fallback`: 현재 모델을 서빙하는 허용 가능한 활성 백엔드가 없을 때만 target model 로 이동하고 다음 규칙을 계속 평가
+- 활성 rewrite cycle은 저장 시점에 거부되며, `/v1/models` 는 실제 요청 가능한 rewrite alias를 함께 반환한다
+
+## User Reasoning Compatibility
+
+- `Users` 화면은 API 키별 `Copy reasoning to reasoning_content` 옵션을 표시하고 편집한다
+- 이 옵션은 같은 백엔드를 공유하는 사용자라도 downstream 클라이언트 호환성에 맞춰 독립적으로 켜거나 끌 수 있다

docs/database.md

@@ -18,6 +18,7 @@ DB는 `DB_DIR` 하위에 분리 저장된다.
 | email | TEXT | |
 | is_active | BOOLEAN | DEFAULT 1 |
 | detail_logging | INTEGER | NOT NULL DEFAULT 0 |
+| copy_reasoning_to_reasoning_content | INTEGER | NOT NULL DEFAULT 0 |
 | created_at | TIMESTAMP | DEFAULT CURRENT_TIMESTAMP |
 | updated_at | TIMESTAMP | DEFAULT CURRENT_TIMESTAMP |
 
@@ -215,7 +216,16 @@ Indexes: `idx_backend_metrics_backend`, `idx_backend_metrics_date`
 | request_headers | TEXT | JSON 문자열 |
 | request_body | TEXT | JSON 또는 raw 문자열 |
 | response_headers | TEXT | JSON 문자열 |
-| response_body | TEXT | JSON 또는 raw 문자열 |
+| response_body | TEXT | JSON 또는 raw 문자열. stream 상세 로그는 기본적으로 `kyush.chat_stream.compact.v1` JSON으로 저장되며, `DETAIL_STREAM_LOG_MODE=raw` 인 경우 기존 raw SSE 문자열로 저장된다 |
 | created_at | TEXT | UTC ISO timestamp |
 
 Indexes: `idx_request_logs_created_at`, `idx_request_logs_local_date`, `idx_request_logs_user`, `idx_request_logs_backend`, `idx_request_logs_endpoint`, `idx_request_logs_detail_logged`
+
+### Stream response body formats
+
+기존 월별 DB의 raw SSE 문자열은 계속 유효하다. 새 stream 로그는 `DETAIL_STREAM_LOG_MODE` 값에 따라 아래 형식 중 하나로 저장된다.
+
+- `compact`(기본): `response_body` 가 `{"format":"kyush.chat_stream.compact.v1", ...}` JSON 문자열이다. 반복되는 chunk 공통 필드(`id`, `object`, `created`, `model`)는 한 번만 저장하고, `choices[].reasoning`, `choices[].content`, `choices[].tool_calls[].function.arguments` 는 누적 문자열로 저장한다.
+- `raw`: `response_body` 가 기존처럼 `data: ...\n\n` 형태의 raw SSE 문자열이다.
+- `both`: `response_body` 가 `{"format":"kyush.chat_stream.raw.v1","compact":...,"raw_sse":"..."}` JSON 문자열이다.
+- `off`: stream `response_body` 를 저장하지 않는다.

docs/model-routing.md

@@ -9,7 +9,7 @@
 1. 사용자 API 키 인증
 2. 사용자가 접근 가능한 backend id 목록 로드
 3. 접근 가능한 활성 백엔드 중 아직 메모리 카탈로그가 초기화되지 않은 백엔드만 `/v1/models` 로 lazy fetch
-4. 요청 `model` 에 대해 전역 `model_rewrites` 규칙 평가
+4. 요청 `model` 에 대해 전역 `model_rewrites` 체인을 끝까지 평가
 5. 최종 모델을 서빙하는 허용 가능한 활성 백엔드만 후보로 선택
 6. 후보 중 1개를 랜덤 선택 후 업스트림으로 포워딩
 
@@ -17,7 +17,9 @@
 
 1. 사용자 API 키 인증
 2. 접근 가능한 활성 백엔드의 메모리 카탈로그를 확인
-3. 모델 ID 합집합을 반환
+3. native backend 모델과 rewrite `source_model` alias를 같은 체인 해석기로 평가
+4. 최종 모델 후보가 있는 requestable 모델 ID 합집합을 반환
+5. `MODEL_LIST_INCLUDE_ROUTING_METADATA` 가 켜져 있으면 각 model object에 비표준 `kyush_router` routing metadata를 추가한다
 
 ## Caching Rules
 
@@ -37,13 +39,24 @@
 
 | Mode | Condition | Result |
 |------|-----------|--------|
-| `force=true` | 항상 | `source_model` 을 즉시 `target_model` 로 치환 |
-| `force=false` | 원본 모델 후보가 없을 때만 | `target_model` 을 fallback 으로 사용 |
+| `force=true` | 항상 | `source_model` 을 `target_model` 로 치환하고 다음 규칙을 계속 평가 |
+| `force=false` | 현재 모델 후보가 없을 때만 | `target_model` 을 fallback 으로 사용하고 다음 규칙을 계속 평가 |
 
 해석 기준:
-- “원본 모델 후보가 있다”는 것은 사용자가 접근 가능하고 활성 상태이며, 메모리 카탈로그상 해당 모델을 서빙하는 백엔드가 하나 이상 있다는 뜻이다
-- 원본 모델 후보가 있으면 fallback 규칙은 무시된다
+- “현재 모델 후보가 있다”는 것은 사용자가 접근 가능하고 활성 상태이며, 메모리 카탈로그상 해당 모델을 서빙하는 백엔드가 하나 이상 있다는 뜻이다
+- 현재 모델 후보가 있으면 fallback 규칙은 무시되고 체인 평가가 멈춘다
+- force 규칙은 현재 모델 후보 존재 여부와 관계없이 target으로 이동한다
 - 최종 모델 후보가 없으면 라우터는 포워딩하지 않고 모델 미지원 오류를 반환한다
+- 활성 rewrite 그래프에 cycle이 생기는 관리자 생성/수정은 거부된다
+- 직접 DB 조작 등으로 runtime cycle이 발견되면 라우터는 설정 오류를 반환한다
+- 체인 평가는 요청별 allowed backend set과 candidate memo를 사용해 반복 DB 조회를 피한다
+- `/v1/models` 의 `kyush_router` metadata는 적용된 rewrite hop만 `rewrite_path` 에 담는다. 후보가 있어 fallback이 중단된 규칙은 path에 포함하지 않는다
+- `kyush_router` 는 public routing 설명용이며 backend id/name 같은 내부 라우팅 대상 정보는 포함하지 않는다
+
+예시:
+`AutoModelTranslate -(Force)-> Qwen3.5 -(Force)-> Qwen/Qwen3.5-397B-A17B-FP8 -(Fallback)-> Gemma4 -(Force)-> cyankiwi/gemma-4-26B-A4B-it-AWQ-4bit`
+
+위 예시에서 `Qwen/Qwen3.5-397B-A17B-FP8` 후보가 있으면 fallback이 적용되지 않고 그 모델로 라우팅된다. 후보가 없으면 `Gemma4`로 이동한 뒤 force 규칙을 이어서 적용한다.
 
 ## Admin Surface
 

docs/scripts.md

@@ -20,6 +20,8 @@ isolated-vm 기반 JavaScript 샌드박스에서 요청/응답을 조작하는
 
 백엔드 응답 수신 후 실행. 현재 구현에서는 응답 컨텍스트를 검사하거나 로그/부가 처리를 수행하는 용도로 실행되며, 훅이 반환한 변경 내용이 최종 HTTP 응답에 다시 반영되지는 않는다.
 
+참고: `reasoning` 을 `reasoning_content` 로 복제하는 OpenAI-compatible 호환 처리는 script hook이 아니라 사용자별 라우터 내장 옵션 `copy_reasoning_to_reasoning_content` 로 수행한다.
+
 ## Script Context
 
 스크립트에서 접근 가능한 데이터:

docs/server.md

@@ -54,14 +54,28 @@ server/src/
 - `core.db` 에는 `admin_sessions`, `admin_api_tokens` 도 함께 저장된다
 - `core.db` 에는 `backend_models`, `model_rewrites` 도 저장된다
 - 시간 경계 계산은 `TZ` 기준이다
+- 상세 로그가 켜진 stream 응답은 `DETAIL_STREAM_LOG_MODE` 에 따라 저장된다
+  - `compact`(기본): SSE chunk를 전달하면서 동시에 누적 파싱해 반복 필드를 제거한 `kyush.chat_stream.compact.v1` JSON을 `response_body`에 저장한다
+  - `raw`: 기존 동작처럼 raw SSE 문자열 전체를 저장한다
+  - `both`: compact JSON과 raw SSE를 함께 담은 `kyush.chat_stream.raw.v1` JSON을 저장한다
+  - `off`: stream `response_body` 저장을 생략한다. request/response headers와 request body는 detail logging 설정에 따라 계속 저장된다
 
 ## Model Routing
 
-- 요청 모델명은 먼저 전역 `model_rewrites` 규칙을 확인한다
-- `force=1` 규칙은 항상 `source_model -> target_model` 로 변환한다
-- `force=0` 규칙은 원본 모델을 서빙하는 허용 가능한 활성 백엔드가 없을 때만 fallback 으로 적용한다
+- 요청 모델명은 먼저 전역 `model_rewrites` 체인을 확인한다
+- `force=1` 규칙은 항상 `source_model -> target_model` 로 변환하고 다음 규칙을 계속 확인한다
+- `force=0` 규칙은 현재 모델을 서빙하는 허용 가능한 활성 백엔드가 없을 때만 fallback 으로 적용하고 다음 규칙을 계속 확인한다
+- 활성 rewrite cycle은 관리자 생성/수정 시 거부하고, runtime에서도 방어한다
 - 최종 모델을 서빙하는 허용 가능한 활성 백엔드가 없으면 `/v1/chat/completions` 는 모델 미지원 오류를 반환한다
-- `/v1/models` 는 허용 가능한 활성 백엔드들의 캐시된 모델 목록 합집합을 반환한다
+- `/v1/models` 는 허용 가능한 활성 백엔드들의 native 모델과 실제 요청 가능한 rewrite alias 합집합을 반환한다
+- `MODEL_LIST_INCLUDE_ROUTING_METADATA` 가 켜져 있으면 `/v1/models` 는 비표준 `kyush_router` metadata를 추가해 요청 모델, 최종 라우팅 모델, 적용된 rewrite path를 노출한다.
+
+## Reasoning Compatibility
+
+- 사용자별 `copy_reasoning_to_reasoning_content` 옵션이 켜져 있으면 `/v1/chat/completions` 응답에서 `reasoning` 을 `reasoning_content` 로 추가 복제한다
+- 같은 백엔드라도 API 키별로 옵션을 다르게 둘 수 있다
+- streaming 응답은 옵션이 켜진 경우에만 SSE JSON frame을 변환하고, 옵션이 꺼진 경우 기존처럼 원본 바이트를 전달한다
+- 이미 `reasoning_content` 가 있으면 덮어쓰지 않고 `reasoning` 원본도 유지한다
 
 참고:
 - 세부 라우팅 규칙과 캐시 트리거는 [docs/model-routing.md](./model-routing.md) 참고
@@ -73,6 +87,7 @@ server/src/
 - `AnalyticsService` 는 `analytics.db` 의 일별 집계와 `request_logs_YYYY-MM.db` 의 범위 조회를 함께 사용해 시계열/분포 데이터를 만든다.
 - 모델 추이 키는 `response_model -> routed_model -> request_model -> unknown` 순서로 결정한다.
 - response length 계열 집계는 `completion_tokens` 가 있는 요청만 포함한다.
+- response length histogram은 긴 꼬리 분포를 위해 로그 간격 bin을 사용한다.
 - 자세한 화면/API 설명은 [docs/analytics.md](./analytics.md) 참고.
 
 ## Deployment Notes

package.json

@@ -1,6 +1,6 @@
 {
-  "name": "kyush-llm-router",
-  "version": "1.0",
+  "name": "kyush-llm-router-express",
+  "version": "1.0.10-express",
   "description": "LLM routing server with multi-user API key management",
   "scripts": {
     "dev": "pnpm --parallel dev",

pnpm-workspace.yaml

@@ -2,6 +2,11 @@ packages:
   - server
   - client
 
+allowBuilds:
+  better-sqlite3: true
+  esbuild: true
+  isolated-vm: true
+
 onlyBuiltDependencies:
   - better-sqlite3
   - esbuild

server/src/config/analytics-db.ts

@@ -3,6 +3,25 @@ import path from 'path';
 import fs from 'fs';
 import { ensureDir, getAnalyticsDbPath } from './db-paths';
 
+function hasIndex(database: Database.Database, tableName: string, indexName: string): boolean {
+  const result = database.prepare(`SELECT name FROM sqlite_master WHERE type = 'index' AND tbl_name = ? AND name = ?`).get(tableName, indexName) as { name: string } | undefined;
+  return Boolean(result);
+}
+
+function ensureAnalyticsIndexes(db: Database.Database): void {
+  const indexes: Array<{ name: string; table: string; sql: string }> = [
+    { name: 'idx_usage_stats_backend_date', table: 'usage_stats', sql: 'CREATE INDEX IF NOT EXISTS idx_usage_stats_backend_date ON usage_stats(backend_id, date)' },
+    { name: 'idx_usage_stats_user_backend_date', table: 'usage_stats', sql: 'CREATE INDEX IF NOT EXISTS idx_usage_stats_user_backend_date ON usage_stats(user_id, backend_id, date)' },
+    { name: 'idx_backend_metrics_backend_date', table: 'backend_metrics', sql: 'CREATE INDEX IF NOT EXISTS idx_backend_metrics_backend_date ON backend_metrics(backend_id, date)' },
+  ];
+
+  for (const { name, table, sql } of indexes) {
+    if (!hasIndex(db, table, name)) {
+      db.exec(sql);
+    }
+  }
+}
+
 let db: Database.Database;
 
 export function getAnalyticsDb(): Database.Database {
@@ -16,6 +35,7 @@ export function getAnalyticsDb(): Database.Database {
     const schemaPath = path.join(__dirname, '..', '..', '..', 'database', 'analytics-schema.sql');
     const schema = fs.readFileSync(schemaPath, 'utf-8');
     db.exec(schema);
+    ensureAnalyticsIndexes(db);
   }
   return db;
 }

server/src/config/database.ts

@@ -14,6 +14,9 @@ function runCoreMigrations(database: Database.Database): void {
   if (hasColumn(database, 'model_rewrites', 'force') === false) {
     database.exec('ALTER TABLE model_rewrites ADD COLUMN force BOOLEAN DEFAULT 0');
   }
+  if (hasColumn(database, 'users', 'copy_reasoning_to_reasoning_content') === false) {
+    database.exec('ALTER TABLE users ADD COLUMN copy_reasoning_to_reasoning_content INTEGER NOT NULL DEFAULT 0');
+  }
 }
 
 export function getDb(): Database.Database {

server/src/config/model-list-metadata.ts

@@ -0,0 +1,4 @@
+export function shouldIncludeModelListRoutingMetadata(): boolean {
+  const value = process.env.MODEL_LIST_INCLUDE_ROUTING_METADATA?.trim().toLowerCase();
+  return value === '1' || value === 'true' || value === 'yes' || value === 'on';
+}

server/src/config/request-logs-db.ts

@@ -20,6 +20,22 @@ function initRequestLogsSchema(db: Database.Database): void {
   }
 }
 
+function ensureRequestLogsIndexes(db: Database.Database): void {
+  const existingIndexes = db.prepare("SELECT name FROM sqlite_master WHERE type = 'index' AND tbl_name = 'request_logs'").all() as Array<{ name: string }>;
+  const indexNames = new Set(existingIndexes.map((idx) => idx.name));
+
+  const indexes = [
+    ['idx_request_logs_local_date_backend', 'CREATE INDEX IF NOT EXISTS idx_request_logs_local_date_backend ON request_logs(local_date, backend_id)'],
+    ['idx_request_logs_completion_tokens', 'CREATE INDEX IF NOT EXISTS idx_request_logs_completion_tokens ON request_logs(completion_tokens)'],
+  ];
+
+  for (const [name, sql] of indexes) {
+    if (!indexNames.has(name)) {
+      db.exec(sql);
+    }
+  }
+}
+
 export function getRequestLogsDb(monthKey: string = getLocalMonthKey()): Database.Database {
   const existing = connections.get(monthKey);
   if (existing) {
@@ -31,6 +47,7 @@ export function getRequestLogsDb(monthKey: string = getLocalMonthKey()): Databas
 
   const db = new Database(dbPath);
   initRequestLogsSchema(db);
+  ensureRequestLogsIndexes(db);
   connections.set(monthKey, db);
   return db;
 }

server/src/config/stream-logging.ts

@@ -0,0 +1,11 @@
+export type DetailStreamLogMode = 'compact' | 'raw' | 'both' | 'off';
+
+export function getDetailStreamLogMode(): DetailStreamLogMode {
+  const value = process.env.DETAIL_STREAM_LOG_MODE?.trim().toLowerCase();
+
+  if (value === 'compact' || value === 'raw' || value === 'both' || value === 'off') {
+    return value;
+  }
+
+  return 'compact';
+}

server/src/index.ts

@@ -12,6 +12,7 @@ import { requireAdminAccess, requireSessionCsrf } from './utils/adminAuth';
 import { logger } from './utils/logger';
 import { getUtcTimestamp } from './utils/time';
 import { ModelCatalogService } from './services/ModelCatalogService';
+import { createJsonBodyParser, JSON_BODY_LIMIT, requestBodyErrorHandler } from './utils/requestBody';
 
 const envPathCandidates = [
   path.resolve(__dirname, '..', '..', '.env'),
@@ -43,9 +44,8 @@ export function createServer(): Application {
     origin: corsOrigins,
     credentials: true,
   }));
-  app.use(express.json({
-    limit: '30mb',
-  }));
+  app.use(createJsonBodyParser());
+  app.use(requestBodyErrorHandler);
 
   app.use('/admin/auth', adminAuthRoutes);
   app.use('/admin/analytics', requireAdminAccess, requireSessionCsrf, analyticsRoutes);

server/src/models/User.ts

@@ -7,6 +7,7 @@ export class UserModel {
   static asUser(row: any): User {
     row.is_active = !!row.is_active;
     row.detail_logging = !!row.detail_logging;
+    row.copy_reasoning_to_reasoning_content = !!row.copy_reasoning_to_reasoning_content;
     return row as User;
   }
 
@@ -31,10 +32,11 @@ export class UserModel {
     const apiKey = data.api_key ?? generateApiKey();
     const timestamp = getUtcTimestamp();
     const detailLogging = data.detail_logging ?? false;
+    const copyReasoning = data.copy_reasoning_to_reasoning_content ?? false;
     const stmt = getDb().prepare(
-      'INSERT INTO users (api_key, name, email, detail_logging, created_at, updated_at) VALUES (?, ?, ?, ?, ?, ?)'
+      'INSERT INTO users (api_key, name, email, detail_logging, copy_reasoning_to_reasoning_content, created_at, updated_at) VALUES (?, ?, ?, ?, ?, ?, ?)'
     );
-    const result = stmt.run(apiKey, data.name, data.email || null, detailLogging ? 1 : 0, timestamp, timestamp);
+    const result = stmt.run(apiKey, data.name, data.email || null, detailLogging ? 1 : 0, copyReasoning ? 1 : 0, timestamp, timestamp);
     
     return {
       id: result.lastInsertRowid as number,
@@ -43,6 +45,7 @@ export class UserModel {
       email: data.email,
       is_active: true,
       detail_logging: detailLogging,
+      copy_reasoning_to_reasoning_content: copyReasoning,
       created_at: timestamp,
       updated_at: timestamp,
     };
@@ -72,6 +75,10 @@ export class UserModel {
       updates.push('detail_logging = ?');
       values.push(data.detail_logging ? 1 : 0);
     }
+    if (data.copy_reasoning_to_reasoning_content !== undefined) {
+      updates.push('copy_reasoning_to_reasoning_content = ?');
+      values.push(data.copy_reasoning_to_reasoning_content ? 1 : 0);
+    }
 
     if (updates.length === 0) {
       return this.findById(id);

server/src/routes/admin.ts

@@ -21,6 +21,19 @@ const router: Router = Router();
 
 router.use('/scripts', scriptRoutes);
 
+function sendRewriteCycleError(res: Response, rules: ReturnType<typeof ModelRewriteModel.findAll>): boolean {
+  const cycle = ModelCatalogService.detectRewriteCycle(rules);
+  if (!cycle) {
+    return false;
+  }
+
+  res.status(409).json({
+    error: 'Model rewrite cycle detected',
+    cycle,
+  });
+  return true;
+}
+
 router.get('/dashboard/summary', (req: Request, res: Response) => {
   const days = req.query.days ? Number(req.query.days) : 30;
   res.json(AnalyticsService.getDashboardSummary(days));
@@ -34,7 +47,7 @@ router.get('/users', (req: Request, res: Response) => {
 });
 
 router.post('/users', (req: Request, res: Response) => {
-  const { name, email, api_key, detail_logging } = req.body as CreateUserData;
+  const { name, email, api_key, detail_logging, copy_reasoning_to_reasoning_content } = req.body as CreateUserData;
 
   if (!name?.trim()) {
     res.status(400).json({ error: 'Name is required' });
@@ -47,6 +60,7 @@ router.post('/users', (req: Request, res: Response) => {
       email: email?.trim() || undefined,
       api_key: api_key?.trim() || undefined,
       detail_logging,
+      copy_reasoning_to_reasoning_content,
     });
 
     res.status(201).json(user);
@@ -80,7 +94,7 @@ router.put('/users/:id', (req: Request, res: Response) => {
     return;
   }
 
-  const { name, email, api_key, is_active, detail_logging } = req.body as UpdateUserData;
+  const { name, email, api_key, is_active, detail_logging, copy_reasoning_to_reasoning_content } = req.body as UpdateUserData;
 
   if (typeof name === 'string' && !name.trim()) {
     res.status(400).json({ error: 'Name cannot be empty' });
@@ -94,6 +108,7 @@ router.put('/users/:id', (req: Request, res: Response) => {
       api_key: typeof api_key === 'string' ? api_key.trim() || undefined : undefined,
       is_active,
       detail_logging,
+      copy_reasoning_to_reasoning_content,
     });
 
     res.json(updatedUser);
@@ -303,10 +318,30 @@ router.post('/model-rewrites', (req: Request, res: Response) => {
     return;
   }
 
+  const sourceModel = source_model.trim();
+  const targetModel = target_model.trim();
+  const timestamp = getUtcTimestamp();
+  const candidateRules = [
+    ...ModelRewriteModel.findAll(),
+    {
+      id: 0,
+      source_model: sourceModel,
+      target_model: targetModel,
+      is_active: is_active === false ? false : true,
+      force: !!force,
+      note,
+      created_at: timestamp,
+      updated_at: timestamp,
+    },
+  ];
+  if (sendRewriteCycleError(res, candidateRules)) {
+    return;
+  }
+
   try {
     const rule = ModelRewriteModel.create({
-      source_model: source_model.trim(),
-      target_model: target_model.trim(),
+      source_model: sourceModel,
+      target_model: targetModel,
       is_active,
       force,
       note,
@@ -330,8 +365,40 @@ router.put('/model-rewrites/:id', (req: Request, res: Response) => {
     return;
   }
 
+  const data = req.body as UpdateModelRewriteData;
+  if (typeof data.source_model === 'string' && !data.source_model.trim()) {
+    res.status(400).json({ error: 'source_model cannot be empty' });
+    return;
+  }
+  if (typeof data.target_model === 'string' && !data.target_model.trim()) {
+    res.status(400).json({ error: 'target_model cannot be empty' });
+    return;
+  }
+
+  const candidateRules = ModelRewriteModel.findAll().map((rule) => {
+    if (rule.id !== id) {
+      return rule;
+    }
+
+    return {
+      ...rule,
+      source_model: typeof data.source_model === 'string' ? data.source_model.trim() : rule.source_model,
+      target_model: typeof data.target_model === 'string' ? data.target_model.trim() : rule.target_model,
+      is_active: data.is_active !== undefined ? data.is_active : rule.is_active,
+      force: data.force !== undefined ? data.force : rule.force,
+      note: data.note !== undefined ? data.note : rule.note,
+    };
+  });
+  if (sendRewriteCycleError(res, candidateRules)) {
+    return;
+  }
+
   try {
-    const updated = ModelRewriteModel.update(id, req.body as UpdateModelRewriteData);
+    const updated = ModelRewriteModel.update(id, {
+      ...data,
+      source_model: typeof data.source_model === 'string' ? data.source_model.trim() : undefined,
+      target_model: typeof data.target_model === 'string' ? data.target_model.trim() : undefined,
+    });
     ModelCatalogService.loadRewriteMap();
     res.json(updated);
   } catch (error) {

server/src/routes/analytics.ts

@@ -4,85 +4,117 @@ import { AnalyticsService } from '../services/AnalyticsService';
 const router: Router = Router();
 
 router.get('/usage', (req: Request, res: Response) => {
-  const { userId, backendId, days } = req.query;
-  const result = AnalyticsService.getUsageStats(
-    userId ? Number(userId) : undefined,
-    backendId ? Number(backendId) : undefined,
-    days ? Number(days) : 30
-  );
-  res.json(result);
+  try {
+    const { userId, backendId, days } = req.query;
+    const result = AnalyticsService.getUsageStats(
+      userId ? Number(userId) : undefined,
+      backendId ? Number(backendId) : undefined,
+      days ? Number(days) : 30
+    );
+    res.json(result);
+  } catch (error) {
+    res.status(500).json({ error: 'Failed to fetch usage stats', details: error instanceof Error ? error.message : String(error) });
+  }
 });
 
 router.get('/requests', (req: Request, res: Response) => {
-  const { month, date, limit, offset, q, userId, backendId, endpoint, detailLogged } = req.query;
-  const result = AnalyticsService.getRequestLogs({
-    month: typeof month === 'string' ? month : undefined,
-    date: typeof date === 'string' ? date : undefined,
-    limit: limit ? Number(limit) : 100,
-    offset: offset ? Number(offset) : 0,
-    q: typeof q === 'string' ? q : undefined,
-    userId: userId ? Number(userId) : undefined,
-    backendId: backendId ? Number(backendId) : undefined,
-    endpoint: typeof endpoint === 'string' ? endpoint : undefined,
-    detailLogged: detailLogged === undefined ? undefined : detailLogged === '1' || detailLogged === 'true',
-  });
-  res.json(result);
+  try {
+    const { month, date, limit, offset, q, userId, backendId, endpoint, detailLogged } = req.query;
+    const result = AnalyticsService.getRequestLogs({
+      month: typeof month === 'string' ? month : undefined,
+      date: typeof date === 'string' ? date : undefined,
+      limit: limit ? Number(limit) : 100,
+      offset: offset ? Number(offset) : 0,
+      q: typeof q === 'string' ? q : undefined,
+      userId: userId ? Number(userId) : undefined,
+      backendId: backendId ? Number(backendId) : undefined,
+      endpoint: typeof endpoint === 'string' ? endpoint : undefined,
+      detailLogged: detailLogged === undefined ? undefined : detailLogged === '1' || detailLogged === 'true',
+    });
+    res.json(result);
+  } catch (error) {
+    res.status(500).json({ error: 'Failed to fetch request logs', details: error instanceof Error ? error.message : String(error) });
+  }
 });
 
 router.get('/metrics', (req: Request, res: Response) => {
-  const { backendId, days } = req.query;
-  const result = AnalyticsService.getBackendMetrics(
-    backendId ? Number(backendId) : undefined,
-    days ? Number(days) : 30
-  );
-  res.json(result);
+  try {
+    const { backendId, days } = req.query;
+    const result = AnalyticsService.getBackendMetrics(
+      backendId ? Number(backendId) : undefined,
+      days ? Number(days) : 30
+    );
+    res.json(result);
+  } catch (error) {
+    res.status(500).json({ error: 'Failed to fetch backend metrics', details: error instanceof Error ? error.message : String(error) });
+  }
 });
 
 router.get('/daily-totals', (req: Request, res: Response) => {
-  const { backendId, days } = req.query;
-  const result = AnalyticsService.getDailyTotals(
-    backendId ? Number(backendId) : undefined,
-    days ? Number(days) : 30
-  );
-  res.json(result);
+  try {
+    const { backendId, days } = req.query;
+    const result = AnalyticsService.getDailyTotals(
+      backendId ? Number(backendId) : undefined,
+      days ? Number(days) : 30
+    );
+    res.json(result);
+  } catch (error) {
+    res.status(500).json({ error: 'Failed to fetch daily totals', details: error instanceof Error ? error.message : String(error) });
+  }
 });
 
 router.get('/backend-quality', (req: Request, res: Response) => {
-  const { backendId, days } = req.query;
-  const result = AnalyticsService.getBackendQuality(
-    backendId ? Number(backendId) : undefined,
-    days ? Number(days) : 30
-  );
-  res.json(result);
+  try {
+    const { backendId, days } = req.query;
+    const result = AnalyticsService.getBackendQuality(
+      backendId ? Number(backendId) : undefined,
+      days ? Number(days) : 30
+    );
+    res.json(result);
+  } catch (error) {
+    res.status(500).json({ error: 'Failed to fetch backend quality', details: error instanceof Error ? error.message : String(error) });
+  }
 });
 
 router.get('/model-trends', (req: Request, res: Response) => {
-  const { backendId, days, limit } = req.query;
-  const result = AnalyticsService.getModelTrends(
-    backendId ? Number(backendId) : undefined,
-    days ? Number(days) : 30,
-    limit ? Number(limit) : 8
-  );
-  res.json(result);
+  try {
+    const { backendId, days, limit } = req.query;
+    const result = AnalyticsService.getModelTrends(
+      backendId ? Number(backendId) : undefined,
+      days ? Number(days) : 30,
+      limit ? Number(limit) : 8
+    );
+    res.json(result);
+  } catch (error) {
+    res.status(500).json({ error: 'Failed to fetch model trends', details: error instanceof Error ? error.message : String(error) });
+  }
 });
 
 router.get('/response-length-histogram', (req: Request, res: Response) => {
-  const { backendId, days, bins } = req.query;
-  const result = AnalyticsService.getResponseLengthHistogram(
-    backendId ? Number(backendId) : undefined,
-    days ? Number(days) : 30,
-    bins ? Number(bins) : 20
-  );
-  res.json(result);
+  try {
+    const { backendId, days, bins } = req.query;
+    const result = AnalyticsService.getResponseLengthHistogram(
+      backendId ? Number(backendId) : undefined,
+      days ? Number(days) : 30,
+      bins ? Number(bins) : 20
+    );
+    res.json(result);
+  } catch (error) {
+    res.status(500).json({ error: 'Failed to fetch response length histogram', details: error instanceof Error ? error.message : String(error) });
+  }
 });
 
 router.get('/response-length-box-plot', (req: Request, res: Response) => {
-  const { backendId, days } = req.query;
-  const result = AnalyticsService.getResponseLengthBoxPlot(
-    backendId ? Number(backendId) : undefined,
-    days ? Number(days) : 30
-  );
-  res.json(result);
+  try {
+    const { backendId, days } = req.query;
+    const result = AnalyticsService.getResponseLengthBoxPlot(
+      backendId ? Number(backendId) : undefined,
+      days ? Number(days) : 30
+    );
+    res.json(result);
+  } catch (error) {
+    res.status(500).json({ error: 'Failed to fetch response length box plot', details: error instanceof Error ? error.message : String(error) });
+  }
 });
 
 export default router;

server/src/routes/api.ts

@@ -1,11 +1,14 @@
 import { Router, Request, Response } from 'express';
 import { authenticate, AuthenticatedRequest } from './auth';
-import { BackendModel } from '../models/Backend';
 import { RouterService } from '../services/RouterService';
 import { AnalyticsService } from '../services/AnalyticsService';
 import { ScriptEngine } from '../services/ScriptEngine';
 import { logger } from '../utils/logger';
-import { ModelCatalogService } from '../services/ModelCatalogService';
+import { ModelCatalogService, ModelRewriteCycleError } from '../services/ModelCatalogService';
+import { getDetailStreamLogMode } from '../config/stream-logging';
+import { shouldIncludeModelListRoutingMetadata } from '../config/model-list-metadata';
+import { ChatStreamLogAccumulator } from '../utils/streamLog';
+import { ReasoningCompatSseTransformer, copyReasoningToReasoningContentInChatCompletion } from '../utils/reasoningCompat';
 
 const router: Router = Router();
 
@@ -34,17 +37,14 @@ router.post('/chat/completions', async (req: AuthenticatedRequest, res: Response
 
   const requestedModel = typeof req.body?.model === 'string' ? req.body.model : '';
   await ModelCatalogService.ensureInitializedForBackends(allowedBackendIds);
-  const resolution = ModelCatalogService.resolveRequestedModel(requestedModel, allowedBackendIds);
-  const activeAllowedBackendIds = BackendModel.findActive()
-    .map((item) => item.id)
-    .filter((backendId) => allowedBackendIds.includes(backendId));
+  const activeAllowedBackendIds = ModelCatalogService.getActiveAllowedBackendIds(allowedBackendIds);
   if (activeAllowedBackendIds.length === 0) {
     AnalyticsService.logRequest({
       user_id: user.id,
       backend_id: 0,
       endpoint: '/v1/chat/completions',
       request_model: requestedModel,
-      routed_model: resolution.routedModel,
+      routed_model: requestedModel,
       status_code: 403,
       error_message: 'No active backends available',
       detail_logged: user.detail_logging,
@@ -54,7 +54,34 @@ router.post('/chat/completions', async (req: AuthenticatedRequest, res: Response
     res.status(403).json({ error: 'No active backends available' });
     return;
   }
-  const candidateBackendIds = ModelCatalogService.getCandidateBackendIds(resolution.routedModel, allowedBackendIds);
+
+  let resolution: ReturnType<typeof ModelCatalogService.resolveRequestedModel>;
+  try {
+    resolution = ModelCatalogService.resolveRequestedModel(requestedModel, activeAllowedBackendIds);
+  } catch (error) {
+    const errorMsg = error instanceof ModelRewriteCycleError
+      ? error.message
+      : error instanceof Error
+        ? error.message
+        : 'Model rewrite resolution failed';
+    AnalyticsService.logRequest({
+      user_id: user.id,
+      backend_id: 0,
+      endpoint: '/v1/chat/completions',
+      request_model: requestedModel,
+      routed_model: requestedModel,
+      status_code: 500,
+      error_message: errorMsg,
+      detail_logged: user.detail_logging,
+      request_headers: user.detail_logging ? normalizeHeaders(req.headers) : undefined,
+      request_body: user.detail_logging ? req.body : undefined,
+    });
+    logger.error(`Model rewrite resolution failed for user ${user.id}: ${errorMsg}`);
+    res.status(500).json({ error: 'Model rewrite configuration error' });
+    return;
+  }
+
+  const candidateBackendIds = ModelCatalogService.getCandidateBackendIds(resolution.routedModel, activeAllowedBackendIds);
   const backend = RouterService.selectBackend(candidateBackendIds);
   if (!backend) {
     AnalyticsService.logRequest({
@@ -187,37 +214,49 @@ router.post('/chat/completions', async (req: AuthenticatedRequest, res: Response
 
       const reader = backendResponse.body!.getReader();
       const decoder = new TextDecoder();
+      const encoder = new TextEncoder();
+      const streamTransformer = user.copy_reasoning_to_reasoning_content ? new ReasoningCompatSseTransformer() : null;
       req.on('close', () => reader.cancel());
 
-      let responseModel: string | undefined;
-      let usage: { prompt_tokens?: number; completion_tokens?: number; total_tokens?: number } | undefined;
-      const collectedChunks: string[] = [];
+      const detailStreamLogMode = getDetailStreamLogMode();
+      const streamLog = new ChatStreamLogAccumulator(detailLoggingEnabled && (detailStreamLogMode === 'raw' || detailStreamLogMode === 'both'));
 
       try {
         while (true) {
           const { done, value } = await reader.read();
           if (done) break;
-          res.write(value);
 
-          // Parse SSE chunks for model and usage metadata
           const text = decoder.decode(value, { stream: true });
-          if (detailLoggingEnabled) collectedChunks.push(text);
-
-          for (const line of text.split('\n')) {
-            if (!line.startsWith('data: ') || line === 'data: [DONE]') continue;
-            try {
-              const parsed = JSON.parse(line.slice(6));
-              if (parsed.model && !responseModel) responseModel = parsed.model;
-              if (parsed.usage) usage = parsed.usage;
-            } catch { /* non-JSON data line, skip */ }
+          if (streamTransformer) {
+            const transformedText = streamTransformer.append(text);
+            if (transformedText) {
+              res.write(encoder.encode(transformedText));
+              streamLog.append(transformedText);
+            }
+          } else {
+            res.write(value);
+            streamLog.append(text);
           }
         }
+
+        const remainingText = decoder.decode();
+        if (streamTransformer) {
+          const transformedText = streamTransformer.append(remainingText) + streamTransformer.flush();
+          if (transformedText) {
+            res.write(encoder.encode(transformedText));
+            streamLog.append(transformedText, false);
+          }
+        } else if (remainingText) {
+          streamLog.append(remainingText, false);
+        }
       } catch (err) {
         logger.error(`Stream interrupted for user ${user.id}: ${err instanceof Error ? err.message : String(err)}`);
       } finally {
         res.end();
       }
 
+      streamLog.flush();
+
       const responseTime = Date.now() - startTime;
       AnalyticsService.logRequest({
         user_id: user.id,
@@ -225,17 +264,17 @@ router.post('/chat/completions', async (req: AuthenticatedRequest, res: Response
         endpoint: '/v1/chat/completions',
         request_model: model,
         routed_model: resolution.routedModel,
-        response_model: responseModel,
-        prompt_tokens: usage?.prompt_tokens,
-        completion_tokens: usage?.completion_tokens,
-        total_tokens: usage?.total_tokens,
+        response_model: streamLog.getResponseModel(),
+        prompt_tokens: streamLog.getUsage()?.prompt_tokens,
+        completion_tokens: streamLog.getUsage()?.completion_tokens,
+        total_tokens: streamLog.getUsage()?.total_tokens,
         status_code: backendResponse.status,
         response_time_ms: responseTime,
         detail_logged: detailLoggingEnabled,
         request_headers: detailLoggingEnabled ? modifiedContext.request.headers : undefined,
         request_body: detailLoggingEnabled ? modifiedContext.request.body : undefined,
         response_headers: detailLoggingEnabled ? backendResponseHeaders : undefined,
-        response_body: detailLoggingEnabled ? collectedChunks.join('') : undefined,
+        response_body: detailLoggingEnabled ? streamLog.toLogBody(detailStreamLogMode) : undefined,
         local_date: undefined,
       });
 
@@ -255,11 +294,14 @@ router.post('/chat/completions', async (req: AuthenticatedRequest, res: Response
     );
 
     const responseTime = Date.now() - startTime;
+    const responseData = user.copy_reasoning_to_reasoning_content
+      ? copyReasoningToReasoningContentInChatCompletion(response.data, false)
+      : response.data;
 
     const responseContext = {
       status: response.status,
       headers: response.headers,
-      body: response.data,
+      body: responseData,
       isStream: false,
     };
 
@@ -276,23 +318,23 @@ router.post('/chat/completions', async (req: AuthenticatedRequest, res: Response
       endpoint: '/v1/chat/completions',
       request_model: model,
       routed_model: resolution.routedModel,
-      response_model: response.data && typeof response.data === 'object' && 'model' in response.data ? String(response.data.model) : undefined,
-      prompt_tokens: response.data && typeof response.data === 'object' && 'usage' in response.data && typeof (response.data as { usage?: { prompt_tokens?: number } }).usage === 'object' ? (response.data as { usage: { prompt_tokens: number } }).usage?.prompt_tokens : undefined,
-      completion_tokens: response.data && typeof response.data === 'object' && 'usage' in response.data && typeof (response.data as { usage?: { completion_tokens?: number } }).usage === 'object' ? (response.data as { usage: { completion_tokens: number } }).usage?.completion_tokens : undefined,
-      total_tokens: response.data && typeof response.data === 'object' && 'usage' in response.data && typeof (response.data as { usage?: { total_tokens?: number } }).usage === 'object' ? (response.data as { usage: { total_tokens: number } }).usage?.total_tokens : undefined,
+      response_model: responseData && typeof responseData === 'object' && 'model' in responseData ? String(responseData.model) : undefined,
+      prompt_tokens: responseData && typeof responseData === 'object' && 'usage' in responseData && typeof (responseData as { usage?: { prompt_tokens?: number } }).usage === 'object' ? (responseData as { usage: { prompt_tokens: number } }).usage?.prompt_tokens : undefined,
+      completion_tokens: responseData && typeof responseData === 'object' && 'usage' in responseData && typeof (responseData as { usage?: { completion_tokens?: number } }).usage === 'object' ? (responseData as { usage: { completion_tokens: number } }).usage?.completion_tokens : undefined,
+      total_tokens: responseData && typeof responseData === 'object' && 'usage' in responseData && typeof (responseData as { usage?: { total_tokens?: number } }).usage === 'object' ? (responseData as { usage: { total_tokens: number } }).usage?.total_tokens : undefined,
       status_code: response.status,
       response_time_ms: responseTime,
-      error_message: response.status >= 400 ? JSON.stringify(response.data) : undefined,
+      error_message: response.status >= 400 ? JSON.stringify(responseData) : undefined,
       detail_logged: detailLoggingEnabled,
       request_headers: detailLoggingEnabled ? modifiedContext.request.headers : undefined,
       request_body: detailLoggingEnabled ? modifiedContext.request.body : undefined,
       response_headers: detailLoggingEnabled ? response.headers : undefined,
-      response_body: detailLoggingEnabled ? response.data : undefined,
+      response_body: detailLoggingEnabled ? responseData : undefined,
       local_date: undefined,
     });
 
     if (response.status >= 400) {
-      const errorDetails = response.data as any;
+      const errorDetails = responseData as any;
       const errorInfo = errorDetails.error || 'Unknown error';
       const causeInfo = errorDetails.cause ? ` (Cause: ${errorDetails.cause})` : '';
       const backendInfo = errorDetails.backend ? ` [Backend: ${errorDetails.backend}]` : '';
@@ -300,7 +342,7 @@ router.post('/chat/completions', async (req: AuthenticatedRequest, res: Response
       void ModelCatalogService.refreshBackendAfterFailure(backend.id);
     }
 
-    res.status(response.status).json(response.data);
+    res.status(response.status).json(responseData);
   } catch (error) {
     const responseTime = Date.now() - startTime;
 
@@ -338,17 +380,51 @@ router.get('/models', async (req: AuthenticatedRequest, res: Response) => {
   }
 
   await ModelCatalogService.ensureInitializedForBackends(allowedBackendIds);
-  const activeAllowedBackendIds = BackendModel.findActive()
-    .map((item) => item.id)
-    .filter((backendId) => allowedBackendIds.includes(backendId));
+  const activeAllowedBackendIds = ModelCatalogService.getActiveAllowedBackendIds(allowedBackendIds);
   if (activeAllowedBackendIds.length === 0) {
     res.status(403).json({ error: 'No active backends available' });
     return;
   }
-  const models = ModelCatalogService.getModelsForAllowedBackends(activeAllowedBackendIds).map((entry) => ({
-    id: entry.model_id,
-    object: 'model',
-  }));
+  let models: Array<{
+    id: string;
+    object: string;
+    kyush_router?: {
+      requested_model: string;
+      routed_model: string;
+      was_rewritten: boolean;
+      rule_type: string;
+      rewrite_path: Array<{ source_model: string; target_model: string; mode: string }>;
+    };
+  }>;
+  try {
+    const includeRoutingMetadata = shouldIncludeModelListRoutingMetadata();
+    models = ModelCatalogService.getRequestableModelsForAllowedBackends(activeAllowedBackendIds).map((entry) => {
+      const model = {
+        id: entry.model_id,
+        object: 'model',
+      };
+
+      if (!includeRoutingMetadata) {
+        return model;
+      }
+
+      return {
+        ...model,
+        kyush_router: {
+          requested_model: entry.routing.requestedModel,
+          routed_model: entry.routing.routedModel,
+          was_rewritten: entry.routing.wasRewritten,
+          rule_type: entry.routing.ruleType,
+          rewrite_path: entry.routing.rewritePath,
+        },
+      };
+    });
+  } catch (error) {
+    const errorMsg = error instanceof Error ? error.message : 'Model rewrite resolution failed';
+    logger.error(`Model list resolution failed: ${errorMsg}`);
+    res.status(500).json({ error: 'Model rewrite configuration error' });
+    return;
+  }
   res.json({ object: 'list', data: models });
 });
 

server/src/services/AnalyticsService.ts

@@ -21,15 +21,6 @@ type DailyTotalsRow = {
   total_tokens: number;
 };
 
-type RequestLogRangeRow = {
-  local_date: string;
-  backend_id: number;
-  request_model: string | null;
-  routed_model: string | null;
-  response_model: string | null;
-  completion_tokens: number | null;
-};
-
 function getDateRange(days: number): { startDate: string; endDate: string } {
   const normalizedDays = Math.max(1, days);
   const endDate = getLocalDateKey();
@@ -37,17 +28,17 @@ function getDateRange(days: number): { startDate: string; endDate: string } {
   return { startDate, endDate };
 }
 
-function buildRequestLogRangeWhere(filter: RequestLogFilter): { whereClause: string; params: unknown[] } {
+function buildWhereClause(startDate: string, endDate: string, backendId: number | undefined): { whereClause: string; params: unknown[] } {
   const clauses = ['local_date >= ?', 'local_date <= ?'];
-  const params: unknown[] = [filter.startDate, filter.endDate];
+  const params: unknown[] = [startDate, endDate];
 
-  if (filter.backendId) {
+  if (backendId) {
     clauses.push('backend_id = ?');
-    params.push(filter.backendId);
+    params.push(backendId);
   }
 
   return {
-    whereClause: `WHERE ${clauses.join(' AND ')}`,
+    whereClause: clauses.length > 0 ? `WHERE ${clauses.join(' AND ')}` : '',
     params,
   };
 }
@@ -63,10 +54,10 @@ function groupByDate(rows: DailyTotalsRow[]): DailyTotalsRow[] {
   for (const row of rows) {
     const existing = grouped.get(row.date);
     if (existing) {
-      existing.total_requests += row.total_requests;
-      existing.total_tokens += row.total_tokens;
+        existing.total_requests += row.total_requests;
+        existing.total_tokens += row.total_tokens;
     } else {
-      grouped.set(row.date, { ...row });
+        grouped.set(row.date, { ...row });
     }
   }
 
@@ -131,49 +122,22 @@ export class AnalyticsService {
     const db = getAnalyticsDb();
     const today = getLocalDateKey();
     const isSuccess = logData.status_code >= 200 && logData.status_code < 300;
+    const tokens = logData.total_tokens || 0;
+    const responseTime = logData.response_time_ms || 0;
+    const errorIncrement = isSuccess ? 0 : 1;
+    const initialSuccessRate = isSuccess ? 1.0 : 0.0;
 
-    const existing = db.prepare(
-      'SELECT * FROM backend_metrics WHERE backend_id = ? AND date = ?'
-    ).get(backendId, today) as {
-      total_requests: number;
-      total_tokens: number;
-      avg_response_time_ms: number;
-      error_count: number;
-    } | undefined;
-
-    if (existing) {
-      const newTotalRequests = existing.total_requests + 1;
-      const newTotalTokens = existing.total_tokens + (logData.total_tokens || 0);
-      const newErrorCount = existing.error_count + (isSuccess ? 0 : 1);
-      const newAvgResponseTime = logData.response_time_ms
-        ? (existing.avg_response_time_ms * existing.total_requests + logData.response_time_ms) / newTotalRequests
-        : existing.avg_response_time_ms;
-      const newSuccessRate = (newTotalRequests - newErrorCount) / newTotalRequests;
-
-      db.prepare(`
-        UPDATE backend_metrics SET
-          total_requests = ?,
-          total_tokens = ?,
-          avg_response_time_ms = ?,
-          error_count = ?,
-          success_rate = ?
-        WHERE backend_id = ? AND date = ?
-      `).run(newTotalRequests, newTotalTokens, newAvgResponseTime, newErrorCount, newSuccessRate, backendId, today);
-    } else {
-      db.prepare(`
-        INSERT INTO backend_metrics (
-          backend_id, date, total_requests, total_tokens,
-          avg_response_time_ms, error_count, success_rate
-        ) VALUES (?, ?, 1, ?, ?, ?, ?)
-      `).run(
-        backendId,
-        today,
-        logData.total_tokens || 0,
-        logData.response_time_ms || 0,
-        isSuccess ? 0 : 1,
-        isSuccess ? 1.0 : 0.0
-      );
-    }
+    db.prepare(`
+      INSERT INTO backend_metrics (backend_id, date, total_requests, total_tokens, avg_response_time_ms, error_count, success_rate)
+      VALUES (?, ?, 1, ?, ?, ?, ?)
+      ON CONFLICT(backend_id, date)
+      DO UPDATE SET
+        total_requests = total_requests + 1,
+        total_tokens = total_tokens + excluded.total_tokens,
+        avg_response_time_ms = (avg_response_time_ms * total_requests + excluded.avg_response_time_ms) / (total_requests + 1),
+        error_count = error_count + excluded.error_count,
+        success_rate = (total_requests + 1 - (error_count + excluded.error_count)) / (total_requests + 1)
+    `).run(backendId, today, tokens, responseTime, errorIncrement, initialSuccessRate);
   }
 
   static getRequestLogs(query: RequestLogQuery = {}): RequestLogPage {
@@ -269,50 +233,74 @@ export class AnalyticsService {
     return db.prepare(query).all(...params);
   }
 
-  private static collectRequestLogRangeRows(filter: RequestLogFilter): RequestLogRangeRow[] {
-    const { whereClause, params } = buildRequestLogRangeWhere(filter);
-    const rows: RequestLogRangeRow[] = [];
-
-    for (const month of getRequestLogMonthsForRange(filter.startDate, filter.endDate)) {
-      const db = getRequestLogsDb(month);
-      const monthRows = db.prepare(`
-        SELECT local_date, backend_id, request_model, routed_model, response_model, completion_tokens
-        FROM request_logs
-        ${whereClause}
-      `).all(...params) as RequestLogRangeRow[];
-      rows.push(...monthRows);
-    }
-
-    return rows;
-  }
-
+  // SQL-level aggregation: first find top models, then get per-date counts
   static getModelTrends(backendId?: number, days: number = 30, limit: number = 8): unknown[] {
     const { startDate, endDate } = getDateRange(days);
-    const rows = this.collectRequestLogRangeRows({ backendId, startDate, endDate });
-    const countsByModel = new Map<string, number>();
-    const countsByDateAndModel = new Map<string, number>();
+    const months = getRequestLogMonthsForRange(startDate, endDate);
 
-    for (const row of rows) {
-      const model = row.response_model || row.routed_model || row.request_model || 'unknown';
-      countsByModel.set(model, (countsByModel.get(model) ?? 0) + 1);
-      const key = `${row.local_date}::${model}`;
-      countsByDateAndModel.set(key, (countsByDateAndModel.get(key) ?? 0) + 1);
+    const modelCounts = new Map<string, number>();
+
+    for (const month of months) {
+      const db = getRequestLogsDb(month);
+      const { whereClause, params } = buildWhereClause(startDate, endDate, backendId);
+      const rows = db.prepare(`
+        SELECT COALESCE(response_model, COALESCE(routed_model, COALESCE(request_model, 'unknown'))) as model,
+               COUNT(*) as cnt
+        FROM request_logs
+        ${whereClause}
+        GROUP BY model
+      `).all(...params) as Array<{ model: string; cnt: number }>;
+
+      for (const row of rows) {
+        modelCounts.set(row.model, (modelCounts.get(row.model) ?? 0) + row.cnt);
+      }
     }
 
-    const topModels = Array.from(countsByModel.entries())
+    const topModels = Array.from(modelCounts.entries())
       .sort((left, right) => right[1] - left[1] || left[0].localeCompare(right[0]))
       .slice(0, Math.max(1, limit))
       .map(([model]) => model);
 
-    const result: Array<{ date: string; model: string; request_count: number }> = [];
-    const seenDates = new Set(rows.map((row) => row.local_date));
+    if (topModels.length === 0) {
+      return [];
+    }
 
-    for (const date of Array.from(seenDates).sort((left, right) => left.localeCompare(right))) {
+    const topModelSet = new Set(topModels);
+    const dateCounts = new Map<string, Map<string, number>>();
+
+    for (const month of months) {
+      const db = getRequestLogsDb(month);
+      const { whereClause, params } = buildWhereClause(startDate, endDate, backendId);
+      const rows = db.prepare(`
+        SELECT local_date,
+               COALESCE(response_model, COALESCE(routed_model, COALESCE(request_model, 'unknown'))) as model,
+               COUNT(*) as cnt
+        FROM request_logs
+        ${whereClause}
+        GROUP BY local_date, model
+      `).all(...params) as Array<{ local_date: string; model: string; cnt: number }>;
+
+      for (const row of rows) {
+        if (!topModelSet.has(row.model)) continue;
+        let dateMap = dateCounts.get(row.local_date);
+        if (!dateMap) {
+          dateMap = new Map();
+          dateCounts.set(row.local_date, dateMap);
+        }
+        dateMap.set(row.model, row.cnt);
+      }
+    }
+
+    const result: Array<{ date: string; model: string; request_count: number }> = [];
+    const sortedDates = Array.from(dateCounts.keys()).sort((left, right) => left.localeCompare(right));
+
+    for (const date of sortedDates) {
+      const dateMap = dateCounts.get(date)!;
       for (const model of topModels) {
         result.push({
           date,
           model,
-          request_count: countsByDateAndModel.get(`${date}::${model}`) ?? 0,
+          request_count: dateMap.get(model) ?? 0,
         });
       }
     }
@@ -320,65 +308,158 @@ export class AnalyticsService {
     return result;
   }
 
+  // SQL-level histogram: use CASE-based binning with log-transformed values
   static getResponseLengthHistogram(backendId?: number, days: number = 30, bins: number = 20): unknown[] {
     const { startDate, endDate } = getDateRange(days);
-    const values = this.collectRequestLogRangeRows({ backendId, startDate, endDate })
-      .map((row) => row.completion_tokens)
-      .filter((value): value is number => typeof value === 'number' && Number.isFinite(value) && value >= 0);
+    const months = getRequestLogMonthsForRange(startDate, endDate);
+    const safeBinCount = Math.max(1, bins);
 
-    if (values.length === 0) {
+    // First pass: find min/max across all months (aggregated, not row-level)
+    let globalMin = Infinity;
+    let globalMax = -Infinity;
+    let totalCount = 0;
+
+    for (const month of months) {
+      const db = getRequestLogsDb(month);
+      const { whereClause, params } = buildWhereClause(startDate, endDate, backendId);
+      const row = db.prepare(`
+        SELECT MIN(completion_tokens) as min_val, MAX(completion_tokens) as max_val,
+               COUNT(*) as cnt
+        FROM request_logs
+        ${whereClause}
+        AND completion_tokens IS NOT NULL
+        AND completion_tokens >= 0
+      `).get(...params) as { min_val: number | null; max_val: number | null; cnt: number } | undefined;
+
+      if (row && row.cnt > 0) {
+        if (typeof row.min_val === 'number') globalMin = Math.min(globalMin, row.min_val);
+        if (typeof row.max_val === 'number') globalMax = Math.max(globalMax, row.max_val);
+        totalCount += row.cnt;
+      }
+    }
+
+    if (totalCount === 0 || globalMin === Infinity) {
       return [];
     }
 
-    const safeBinCount = Math.max(1, bins);
-    const min = Math.min(...values);
-    const max = Math.max(...values);
-
-    if (min === max) {
-      return [{ bin_start: min, bin_end: max, count: values.length }];
+    if (globalMin === globalMax) {
+      return [{ bin_start: globalMin, bin_end: globalMax, count: totalCount }];
+    }
+
+    const transformedMin = Math.log1p(globalMin);
+    const transformedMax = Math.log1p(globalMax);
+    const width = (transformedMax - transformedMin) / safeBinCount;
+
+    // Build bin boundaries for SQL CASE expression
+    const binBoundaries: number[] = [];
+    for (let i = 0; i < safeBinCount - 1; i++) {
+      binBoundaries.push(Math.expm1(transformedMin + width * (i + 1)));
+    }
+
+    // Build SQL CASE expression for bin assignment
+    const caseParts: string[] = [];
+    for (let i = 0; i < safeBinCount - 1; i++) {
+      caseParts.push(`WHEN completion_tokens < ${binBoundaries[i]} THEN ${i}`);
+    }
+    caseParts.push(`ELSE ${safeBinCount - 1}`);
+    const caseExpr = `CASE ${caseParts.join(' ')} END`;
+
+    // Second pass: count per bin using SQL aggregation
+    const binCounts = new Array(safeBinCount).fill(0);
+
+    for (const month of months) {
+      const db = getRequestLogsDb(month);
+      const { whereClause, params } = buildWhereClause(startDate, endDate, backendId);
+      const rows = db.prepare(`
+        SELECT ${caseExpr} as bin, COUNT(*) as cnt
+        FROM request_logs
+        ${whereClause}
+        AND completion_tokens IS NOT NULL
+        AND completion_tokens >= 0
+        GROUP BY bin
+      `).all(...params) as Array<{ bin: number; cnt: number }>;
+
+      for (const row of rows) {
+        const binIndex = Math.min(safeBinCount - 1, Math.max(0, row.bin));
+        binCounts[binIndex] += row.cnt;
+      }
     }
 
-    const width = (max - min) / safeBinCount;
     const histogram = Array.from({ length: safeBinCount }, (_, index) => ({
-      bin_start: min + width * index,
-      bin_end: index === safeBinCount - 1 ? max : min + width * (index + 1),
-      count: 0,
+      bin_start: index === 0 ? globalMin : Math.expm1(transformedMin + width * index),
+      bin_end: index === safeBinCount - 1 ? globalMax : Math.expm1(transformedMin + width * (index + 1)),
+      count: binCounts[index],
     }));
 
-    for (const value of values) {
-      const index = Math.min(safeBinCount - 1, Math.floor((value - min) / width));
-      histogram[index].count += 1;
-    }
-
     return histogram;
   }
 
+  // SQL-level box plot: fetch per-date aggregates, compute quantiles from sampled data
   static getResponseLengthBoxPlot(backendId?: number, days: number = 30): unknown[] {
     const { startDate, endDate } = getDateRange(days);
-    const rows = this.collectRequestLogRangeRows({ backendId, startDate, endDate });
-    const valuesByDate = new Map<string, number[]>();
+    const months = getRequestLogMonthsForRange(startDate, endDate);
 
-    for (const row of rows) {
-      if (typeof row.completion_tokens !== 'number' || !Number.isFinite(row.completion_tokens) || row.completion_tokens < 0) {
-        continue;
+    const dailyStats = new Map<string, { min: number; max: number; count: number; values: number[] }>();
+
+    for (const month of months) {
+      const db = getRequestLogsDb(month);
+      const { whereClause, params } = buildWhereClause(startDate, endDate, backendId);
+
+      // Get per-date min/max/count via SQL aggregation
+      const summaryRows = db.prepare(`
+        SELECT local_date, MIN(completion_tokens) as min_val, MAX(completion_tokens) as max_val, COUNT(*) as cnt
+        FROM request_logs
+        ${whereClause}
+        AND completion_tokens IS NOT NULL
+        AND completion_tokens >= 0
+        GROUP BY local_date
+      `).all(...params) as Array<{ local_date: string; min_val: number; max_val: number; cnt: number }>;
+
+      for (const row of summaryRows) {
+        const entry = dailyStats.get(row.local_date);
+        if (entry) {
+          entry.count += row.cnt;
+          entry.min = Math.min(entry.min, row.min_val);
+          entry.max = Math.max(entry.max, row.max_val);
+        } else {
+          dailyStats.set(row.local_date, {
+            min: row.min_val,
+            max: row.max_val,
+            count: row.cnt,
+            values: [],
+          });
+        }
       }
 
-      const values = valuesByDate.get(row.local_date) ?? [];
-      values.push(row.completion_tokens);
-      valuesByDate.set(row.local_date, values);
+      // For quantiles, fetch values per date (only completion_tokens column, limited)
+      const dateRows = db.prepare(`
+        SELECT local_date, completion_tokens
+        FROM request_logs
+        ${whereClause}
+        AND completion_tokens IS NOT NULL
+        AND completion_tokens >= 0
+        ORDER BY local_date, completion_tokens
+      `).all(...params) as Array<{ local_date: string; completion_tokens: number }>;
+
+      for (const row of dateRows) {
+        const entry = dailyStats.get(row.local_date);
+        if (entry) {
+          entry.values.push(row.completion_tokens);
+        }
+      }
     }
 
-    return Array.from(valuesByDate.entries())
+    return Array.from(dailyStats.entries())
       .sort((left, right) => left[0].localeCompare(right[0]))
-      .map(([date, values]) => {
-        const sortedValues = [...values].sort((left, right) => left - right);
+      .map(([date, stats]) => {
+        const sortedValues = stats.values.sort((left, right) => left - right);
         return {
           date,
-          min: sortedValues[0],
+          min: sortedValues.length > 0 ? sortedValues[0] : stats.min,
           q1: calculateQuantile(sortedValues, 0.25),
           median: calculateQuantile(sortedValues, 0.5),
           q3: calculateQuantile(sortedValues, 0.75),
-          max: sortedValues[sortedValues.length - 1],
+          max: sortedValues.length > 0 ? sortedValues[sortedValues.length - 1] : stats.max,
           count: sortedValues.length,
         };
       });
@@ -433,6 +514,11 @@ export class AnalyticsService {
       }))
       .sort((left, right) => left.name.localeCompare(right.name));
 
+    // Parallel execution for series data (better-sqlite3 is synchronous, but this makes it explicit)
+    const dailyTotals = this.getDailyTotals(undefined, normalizedDays);
+    const backendQuality = this.getBackendQuality(undefined, normalizedDays);
+    const modelTrends = this.getModelTrends(undefined, normalizedDays, 6);
+
     return {
       window_days: normalizedDays,
       generated_at: now,
@@ -470,9 +556,9 @@ export class AnalyticsService {
         users_without_permissions: users.filter((user) => !permissionsByUserId.has(user.id)).length,
       },
       series: {
-        daily_totals: this.getDailyTotals(undefined, normalizedDays),
-        backend_quality: this.getBackendQuality(undefined, normalizedDays) as DashboardSummaryResponse['series']['backend_quality'],
-        model_trends: this.getModelTrends(undefined, normalizedDays, 6) as DashboardSummaryResponse['series']['model_trends'],
+        daily_totals: dailyTotals,
+        backend_quality: backendQuality as DashboardSummaryResponse['series']['backend_quality'],
+        model_trends: modelTrends as DashboardSummaryResponse['series']['model_trends'],
       },
     };
   }

server/src/services/ModelCatalogService.ts

@@ -31,20 +31,49 @@ interface FetchModelsResponse {
   rawModels: Array<{ model_id: string; raw_json?: string }>;
 }
 
-interface RewriteResolution {
+export interface RewritePathHop {
+  source_model: string;
+  target_model: string;
+  mode: 'force' | 'fallback';
+}
+
+export interface RewriteResolution {
   requestedModel: string;
   routedModel: string;
   wasRewritten: boolean;
-  ruleType: 'none' | 'force' | 'fallback';
+  ruleType: 'none' | 'force' | 'fallback' | 'chain';
+  rewritePath: RewritePathHop[];
 }
 
 interface RewriteConfig {
+  id: number;
+  sourceModel: string;
   targetModel: string;
   force: boolean;
 }
 
+interface ResolutionContext {
+  allowedActiveBackendIds: number[];
+  allowedActiveBackendIdSet: Set<number>;
+  candidateMemo: Map<string, number[]>;
+}
+
+export interface RequestableModelCatalogEntry extends BackendModelCatalogEntry {
+  routing: RewriteResolution;
+}
+
 const DEFAULT_REFRESH_MIN_MS = 5 * 60 * 1000;
 
+export class ModelRewriteCycleError extends Error {
+  cycle: string[];
+
+  constructor(cycle: string[]) {
+    super(`Model rewrite cycle detected: ${cycle.join(' -> ')}`);
+    this.name = 'ModelRewriteCycleError';
+    this.cycle = cycle;
+  }
+}
+
 export class ModelCatalogService {
   private static backendModelsByBackendId = new Map<number, BackendCacheEntry>();
   private static backendIdsByModel = new Map<string, Set<number>>();
@@ -188,14 +217,60 @@ export class ModelCatalogService {
     this.modelRewriteMap.clear();
     for (const rule of ModelRewriteModel.findAll()) {
       if (rule.is_active) {
-        this.modelRewriteMap.set(rule.source_model, {
-          targetModel: rule.target_model,
+        const sourceModel = this.normalizeModelId(rule.source_model);
+        const targetModel = this.normalizeModelId(rule.target_model);
+        this.modelRewriteMap.set(sourceModel, {
+          id: rule.id,
+          sourceModel,
+          targetModel,
           force: rule.force,
         });
       }
     }
   }
 
+  private static createResolutionContext(allowedBackendIds: number[]): ResolutionContext {
+    const allowed = new Set(allowedBackendIds);
+    const allowedActiveBackendIds = BackendModel.findActive()
+      .map((backend) => backend.id)
+      .filter((backendId) => allowed.has(backendId));
+
+    return {
+      allowedActiveBackendIds,
+      allowedActiveBackendIdSet: new Set(allowedActiveBackendIds),
+      candidateMemo: new Map<string, number[]>(),
+    };
+  }
+
+  static getActiveAllowedBackendIds(allowedBackendIds: number[]): number[] {
+    return this.createResolutionContext(allowedBackendIds).allowedActiveBackendIds;
+  }
+
+  private static getCandidateBackendIdsWithContext(modelId: string, context: ResolutionContext): number[] {
+    const normalized = this.normalizeModelId(modelId);
+    const memoized = context.candidateMemo.get(normalized);
+    if (memoized) return memoized;
+
+    const backendIds = this.backendIdsByModel.get(normalized);
+    const candidates = backendIds
+      ? Array.from(backendIds).filter((backendId) => context.allowedActiveBackendIdSet.has(backendId))
+      : [];
+    const sorted = candidates.sort((a, b) => a - b);
+
+    context.candidateMemo.set(normalized, sorted);
+    return sorted;
+  }
+
+  private static getRuleTypeFromAppliedRules(appliedRules: RewriteConfig[]): RewriteResolution['ruleType'] {
+    if (appliedRules.length === 0) {
+      return 'none';
+    }
+    if (appliedRules.length === 1) {
+      return appliedRules[0].force ? 'force' : 'fallback';
+    }
+    return 'chain';
+  }
+
   static syncActiveBackendCacheState(): void {
     const backends = BackendModel.findAll();
     const backendIds = new Set(backends.map((backend) => backend.id));
@@ -216,44 +291,139 @@ export class ModelCatalogService {
     this.rebuildModelIndex();
   }
 
-  static resolveRequestedModel(modelId: string, allowedBackendIds: number[]): RewriteResolution {
+  private static resolveRequestedModelWithContext(modelId: string, context: ResolutionContext): RewriteResolution {
     const requestedModel = this.normalizeModelId(modelId);
-    const rewrite = this.modelRewriteMap.get(requestedModel);
-    if (!rewrite) {
-      return {
-        requestedModel,
-        routedModel: requestedModel,
-        wasRewritten: false,
-        ruleType: 'none',
-      };
+    const visitedModels = new Map<string, number>();
+    const path: string[] = [];
+    const appliedRules: RewriteConfig[] = [];
+    let currentModel = requestedModel;
+    const maxSteps = this.modelRewriteMap.size + 1;
+
+    for (let step = 0; step <= maxSteps; step += 1) {
+      const firstSeenAt = visitedModels.get(currentModel);
+      if (firstSeenAt !== undefined) {
+        throw new ModelRewriteCycleError([...path.slice(firstSeenAt), currentModel]);
+      }
+      visitedModels.set(currentModel, path.length);
+      path.push(currentModel);
+
+      const rewrite = this.modelRewriteMap.get(currentModel);
+      if (!rewrite) {
+        return {
+          requestedModel,
+          routedModel: currentModel,
+          wasRewritten: currentModel !== requestedModel,
+          ruleType: this.getRuleTypeFromAppliedRules(appliedRules),
+          rewritePath: appliedRules.map((rule) => ({
+            source_model: rule.sourceModel,
+            target_model: rule.targetModel,
+            mode: rule.force ? 'force' : 'fallback',
+          })),
+        };
+      }
+
+      if (!rewrite.force) {
+        const originalCandidates = this.getCandidateBackendIdsWithContext(currentModel, context);
+        if (originalCandidates.length > 0) {
+          return {
+            requestedModel,
+            routedModel: currentModel,
+            wasRewritten: currentModel !== requestedModel,
+            ruleType: this.getRuleTypeFromAppliedRules(appliedRules),
+            rewritePath: appliedRules.map((rule) => ({
+              source_model: rule.sourceModel,
+              target_model: rule.targetModel,
+              mode: rule.force ? 'force' : 'fallback',
+            })),
+          };
+        }
+      }
+
+      appliedRules.push(rewrite);
+      currentModel = this.normalizeModelId(rewrite.targetModel);
     }
 
-    if (rewrite.force) {
-      return {
-        requestedModel,
-        routedModel: rewrite.targetModel,
-        wasRewritten: rewrite.targetModel !== requestedModel,
-        ruleType: 'force',
-      };
+    throw new ModelRewriteCycleError([...path, currentModel]);
+  }
+
+  static resolveRequestedModel(modelId: string, allowedBackendIds: number[]): RewriteResolution {
+    return this.resolveRequestedModelWithContext(modelId, this.createResolutionContext(allowedBackendIds));
+  }
+
+  static detectRewriteCycle(rules: ModelRewriteRule[]): string[] | null {
+    const activeRules = new Map<string, string>();
+    for (const rule of rules) {
+      if (rule.is_active) {
+        activeRules.set(this.normalizeModelId(rule.source_model), this.normalizeModelId(rule.target_model));
+      }
     }
 
-    const originalCandidates = this.getCandidateBackendIds(requestedModel, allowedBackendIds);
-    if (originalCandidates.length > 0) {
-      return {
-        requestedModel,
-        routedModel: requestedModel,
-        wasRewritten: false,
-        ruleType: 'none',
-      };
-    }
+    const visited = new Set<string>();
+    const visiting = new Map<string, number>();
+    const path: string[] = [];
 
-    const routedModel = rewrite.targetModel;
-    return {
-      requestedModel,
-      routedModel,
-      wasRewritten: routedModel !== requestedModel,
-      ruleType: 'fallback',
+    const visit = (modelId: string): string[] | null => {
+      const firstSeenAt = visiting.get(modelId);
+      if (firstSeenAt !== undefined) {
+        return [...path.slice(firstSeenAt), modelId];
+      }
+      if (visited.has(modelId)) {
+        return null;
+      }
+
+      visiting.set(modelId, path.length);
+      path.push(modelId);
+
+      const targetModel = activeRules.get(modelId);
+      if (targetModel) {
+        const cycle = visit(targetModel);
+        if (cycle) {
+          return cycle;
+        }
+      }
+
+      path.pop();
+      visiting.delete(modelId);
+      visited.add(modelId);
+      return null;
     };
+
+    for (const sourceModel of activeRules.keys()) {
+      const cycle = visit(sourceModel);
+      if (cycle) {
+        return cycle;
+      }
+    }
+
+    return null;
+  }
+
+  static getRequestableModelsForAllowedBackends(allowedBackendIds: number[]): RequestableModelCatalogEntry[] {
+    const context = this.createResolutionContext(allowedBackendIds);
+    const requestableModelIds = new Set<string>();
+    const candidateModelIds = new Set<string>([
+      ...this.backendIdsByModel.keys(),
+      ...this.modelRewriteMap.keys(),
+    ]);
+
+    for (const modelId of candidateModelIds) {
+      const resolution = this.resolveRequestedModelWithContext(modelId, context);
+      const routedBackendIds = this.getCandidateBackendIdsWithContext(resolution.routedModel, context);
+      if (routedBackendIds.length > 0) {
+        requestableModelIds.add(this.normalizeModelId(modelId));
+      }
+    }
+
+    return Array.from(requestableModelIds)
+      .sort((a, b) => a.localeCompare(b))
+      .map((modelId) => {
+        const resolution = this.resolveRequestedModelWithContext(modelId, context);
+        return {
+          model_id: modelId,
+          backend_ids: this.getCandidateBackendIdsWithContext(resolution.routedModel, context),
+          routing: resolution,
+        };
+      });
   }
 
   static getBackendCacheStatus(backendId: number): BackendModelCacheStatus {
@@ -378,13 +548,7 @@ export class ModelCatalogService {
   }
 
   static getCandidateBackendIds(modelId: string, allowedBackendIds: number[]): number[] {
-    const normalized = this.normalizeModelId(modelId);
-    const backendIds = this.backendIdsByModel.get(normalized);
-    if (!backendIds) return [];
-
-    const allowed = new Set(allowedBackendIds);
-    const active = new Set(BackendModel.findActive().map((backend) => backend.id));
-    return Array.from(backendIds).filter((backendId) => allowed.has(backendId) && active.has(backendId));
+    return this.getCandidateBackendIdsWithContext(modelId, this.createResolutionContext(allowedBackendIds));
   }
 
   static getModelsForAllowedBackends(allowedBackendIds: number[]): BackendModelCatalogEntry[] {

server/src/utils/reasoningCompat.ts

@@ -0,0 +1,91 @@
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return Boolean(value) && typeof value === 'object' && !Array.isArray(value);
+}
+
+function copyReasoningField(target: Record<string, unknown>): boolean {
+  if (typeof target.reasoning !== 'string' || target.reasoning_content !== undefined) {
+    return false;
+  }
+
+  target.reasoning_content = target.reasoning;
+  return true;
+}
+
+export function copyReasoningToReasoningContentInChatCompletion(payload: unknown, stream: boolean): unknown {
+  if (!isRecord(payload) || !Array.isArray(payload.choices)) {
+    return payload;
+  }
+
+  for (const choice of payload.choices) {
+    if (!isRecord(choice)) continue;
+
+    const target = stream ? choice.delta : choice.message;
+    if (isRecord(target)) {
+      copyReasoningField(target);
+    }
+  }
+
+  return payload;
+}
+
+export class ReasoningCompatSseTransformer {
+  private buffer = '';
+
+  append(text: string): string {
+    this.buffer = `${this.buffer}${text}`.replace(/\r\n/g, '\n').replace(/\r/g, '\n');
+    const transformedEvents: string[] = [];
+
+    while (true) {
+      const delimiterIndex = this.buffer.indexOf('\n\n');
+      if (delimiterIndex < 0) break;
+
+      const eventText = this.buffer.slice(0, delimiterIndex);
+      this.buffer = this.buffer.slice(delimiterIndex + 2);
+      transformedEvents.push(this.transformEvent(eventText));
+    }
+
+    return transformedEvents.join('');
+  }
+
+  flush(): string {
+    if (!this.buffer) {
+      return '';
+    }
+
+    const remaining = this.buffer;
+    this.buffer = '';
+    return remaining;
+  }
+
+  private transformEvent(eventText: string): string {
+    const lines = eventText.split('\n');
+    const dataLines = lines
+      .filter((line) => line.startsWith('data:'))
+      .map((line) => line.slice(5).replace(/^ /, ''));
+
+    if (dataLines.length === 0) {
+      return `${eventText}\n\n`;
+    }
+
+    const data = dataLines.join('\n');
+    if (data.trim() === '[DONE]') {
+      return `${eventText}\n\n`;
+    }
+
+    try {
+      const parsed = JSON.parse(data);
+      const before = JSON.stringify(parsed);
+      copyReasoningToReasoningContentInChatCompletion(parsed, true);
+      const after = JSON.stringify(parsed);
+
+      if (before === after) {
+        return `${eventText}\n\n`;
+      }
+
+      const nonDataLines = lines.filter((line) => !line.startsWith('data:'));
+      return `${[...nonDataLines, `data: ${after}`].join('\n')}\n\n`;
+    } catch {
+      return `${eventText}\n\n`;
+    }
+  }
+}

server/src/utils/requestBody.ts

@@ -0,0 +1,82 @@
+import express, { ErrorRequestHandler, RequestHandler } from 'express';
+import { logger } from './logger';
+
+export const JSON_BODY_LIMIT = '30mb';
+
+interface BodyParserError extends Error {
+  type?: string;
+  status?: number;
+  statusCode?: number;
+  limit?: number;
+  length?: number;
+  expected?: number;
+  received?: number;
+}
+
+function parseContentLength(value: string | undefined): number | undefined {
+  if (!value) return undefined;
+  const parsed = Number(value);
+  return Number.isFinite(parsed) && parsed >= 0 ? parsed : undefined;
+}
+
+function firstNumber(...values: Array<number | undefined>): number | undefined {
+  return values.find((value): value is number => typeof value === 'number' && Number.isFinite(value));
+}
+
+export function formatByteSize(bytes: number | undefined): string {
+  if (bytes === undefined) return 'unknown';
+  if (bytes < 1024) return `${bytes} B`;
+
+  const units = ['KB', 'MB', 'GB'];
+  let value = bytes / 1024;
+  let unitIndex = 0;
+  while (value >= 1024 && unitIndex < units.length - 1) {
+    value /= 1024;
+    unitIndex += 1;
+  }
+
+  return `${value.toFixed(value >= 10 ? 1 : 2)} ${units[unitIndex]} (${bytes} B)`;
+}
+
+export function createJsonBodyParser(limit: string = JSON_BODY_LIMIT): RequestHandler {
+  return express.json({ limit });
+}
+
+export const requestBodyErrorHandler: ErrorRequestHandler = (err: BodyParserError, req, res, next) => {
+  const status = err.status ?? err.statusCode;
+  const isPayloadTooLarge = err.type === 'entity.too.large' || status === 413;
+  const isBodyParserClientError = typeof err.type === 'string' && status !== undefined && status >= 400 && status < 500;
+
+  if (isPayloadTooLarge) {
+    const contentLengthBytes = parseContentLength(req.get('content-length'));
+    const payloadBytes = firstNumber(err.length, err.received, err.expected, contentLengthBytes);
+    const limitBytes = firstNumber(err.limit);
+
+    logger.warn(
+      `Request body too large: ${req.method} ${req.originalUrl || req.url} ` +
+      `payload=${formatByteSize(payloadBytes)} limit=${formatByteSize(limitBytes)} ` +
+      `content-length=${formatByteSize(contentLengthBytes)}`,
+      {
+        content_length_bytes: contentLengthBytes,
+        payload_size_bytes: payloadBytes,
+        parser_limit_bytes: limitBytes,
+        parser_error_type: err.type,
+      }
+    );
+
+    res.status(413).json({
+      error: 'Request body too large',
+      payload_size_bytes: payloadBytes,
+      limit_bytes: limitBytes,
+    });
+    return;
+  }
+
+  if (isBodyParserClientError) {
+    logger.warn(`Invalid request body: ${req.method} ${req.originalUrl || req.url} (${err.type})`);
+    res.status(status).json({ error: err.message || 'Invalid request body' });
+    return;
+  }
+
+  next(err);
+};

server/src/utils/streamLog.ts

@@ -0,0 +1,301 @@
+import { DetailStreamLogMode } from '../config/stream-logging';
+
+export const COMPACT_CHAT_STREAM_FORMAT = 'kyush.chat_stream.compact.v1';
+export const RAW_CHAT_STREAM_FORMAT = 'kyush.chat_stream.raw.v1';
+
+type UsageSnapshot = {
+  prompt_tokens?: number;
+  completion_tokens?: number;
+  total_tokens?: number;
+};
+
+type CompactToolCall = {
+  index: number;
+  id?: string;
+  type?: string;
+  function?: {
+    name?: string;
+    arguments?: string;
+  };
+};
+
+type CompactChoice = {
+  index: number;
+  role?: string;
+  reasoning?: string;
+  content?: string;
+  tool_calls?: CompactToolCall[];
+  finish_reason?: string;
+  stop_reason?: string;
+  matched_stop?: string;
+};
+
+type CompactChatStreamLog = {
+  format: typeof COMPACT_CHAT_STREAM_FORMAT;
+  id?: string;
+  object?: string;
+  created?: number;
+  model?: string;
+  choices: CompactChoice[];
+  usage?: UsageSnapshot;
+  stream: {
+    network_chunk_count: number;
+    data_frame_count: number;
+    done: boolean;
+  };
+};
+
+type RawChatStreamLog = {
+  format: typeof RAW_CHAT_STREAM_FORMAT;
+  compact: CompactChatStreamLog;
+  raw_sse: string;
+};
+
+type ChoiceState = {
+  index: number;
+  role?: string;
+  reasoning: string[];
+  content: string[];
+  toolCalls: Map<number, CompactToolCall>;
+  finishReason?: string;
+  stopReason?: string;
+  matchedStop?: string;
+};
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return Boolean(value) && typeof value === 'object' && !Array.isArray(value);
+}
+
+function mergeToolCall(target: CompactToolCall, chunk: Record<string, unknown>): void {
+  if (typeof chunk.id === 'string') target.id = chunk.id;
+  if (typeof chunk.type === 'string') target.type = chunk.type;
+
+  const functionChunk = chunk.function;
+  if (!isRecord(functionChunk)) return;
+
+  target.function = target.function ?? {};
+  if (typeof functionChunk.name === 'string') {
+    target.function.name = `${target.function.name ?? ''}${functionChunk.name}`;
+  }
+  if (typeof functionChunk.arguments === 'string') {
+    target.function.arguments = `${target.function.arguments ?? ''}${functionChunk.arguments}`;
+  }
+}
+
+function getUsage(value: unknown): UsageSnapshot | undefined {
+  if (!isRecord(value)) return undefined;
+
+  return {
+    prompt_tokens: typeof value.prompt_tokens === 'number' ? value.prompt_tokens : undefined,
+    completion_tokens: typeof value.completion_tokens === 'number' ? value.completion_tokens : undefined,
+    total_tokens: typeof value.total_tokens === 'number' ? value.total_tokens : undefined,
+  };
+}
+
+export class ChatStreamLogAccumulator {
+  private id: string | undefined;
+  private object: string | undefined;
+  private created: number | undefined;
+  private model: string | undefined;
+  private usage: UsageSnapshot | undefined;
+  private networkChunkCount = 0;
+  private dataFrameCount = 0;
+  private done = false;
+  private buffer = '';
+  private rawChunks: string[] = [];
+  private choices = new Map<number, ChoiceState>();
+
+  constructor(private readonly collectRaw: boolean) {}
+
+  append(text: string, countNetworkChunk = true): void {
+    if (countNetworkChunk) {
+      this.networkChunkCount += 1;
+    }
+
+    if (this.collectRaw) {
+      this.rawChunks.push(text);
+    }
+
+    this.buffer = `${this.buffer}${text}`.replace(/\r\n/g, '\n').replace(/\r/g, '\n');
+
+    while (true) {
+      const delimiterIndex = this.buffer.indexOf('\n\n');
+      if (delimiterIndex < 0) break;
+
+      const eventText = this.buffer.slice(0, delimiterIndex);
+      this.buffer = this.buffer.slice(delimiterIndex + 2);
+      this.processEvent(eventText);
+    }
+  }
+
+  flush(): void {
+    const remaining = this.buffer.trim();
+    if (!remaining) {
+      this.buffer = '';
+      return;
+    }
+
+    this.processEvent(this.buffer);
+    this.buffer = '';
+  }
+
+  getResponseModel(): string | undefined {
+    return this.model;
+  }
+
+  getUsage(): UsageSnapshot | undefined {
+    return this.usage;
+  }
+
+  toLogBody(mode: DetailStreamLogMode): string | CompactChatStreamLog | RawChatStreamLog | undefined {
+    this.flush();
+
+    if (mode === 'off') {
+      return undefined;
+    }
+
+    if (mode === 'raw') {
+      return this.rawChunks.join('');
+    }
+
+    const compact = this.toCompactLog();
+    if (mode === 'both') {
+      return {
+        format: RAW_CHAT_STREAM_FORMAT,
+        compact,
+        raw_sse: this.rawChunks.join(''),
+      };
+    }
+
+    return compact;
+  }
+
+  private getChoice(index: number): ChoiceState {
+    const existing = this.choices.get(index);
+    if (existing) return existing;
+
+    const created: ChoiceState = {
+      index,
+      reasoning: [],
+      content: [],
+      toolCalls: new Map(),
+    };
+    this.choices.set(index, created);
+    return created;
+  }
+
+  private processEvent(eventText: string): void {
+    const dataLines = eventText
+      .split('\n')
+      .filter((line) => line.startsWith('data:'))
+      .map((line) => line.slice(5).replace(/^ /, ''));
+
+    if (dataLines.length === 0) return;
+
+    const data = dataLines.join('\n');
+    if (data.trim() === '[DONE]') {
+      this.done = true;
+      return;
+    }
+
+    this.dataFrameCount += 1;
+
+    try {
+      const parsed = JSON.parse(data);
+      if (isRecord(parsed)) {
+        this.processPayload(parsed);
+      }
+    } catch {
+      // Non-JSON data frames are forwarded to clients, but not represented in compact logs.
+    }
+  }
+
+  private processPayload(payload: Record<string, unknown>): void {
+    if (typeof payload.id === 'string' && !this.id) this.id = payload.id;
+    if (typeof payload.object === 'string' && !this.object) this.object = payload.object;
+    if (typeof payload.created === 'number' && this.created === undefined) this.created = payload.created;
+    if (typeof payload.model === 'string' && !this.model) this.model = payload.model;
+
+    const usage = getUsage(payload.usage);
+    if (usage) this.usage = usage;
+
+    if (!Array.isArray(payload.choices)) return;
+
+    for (const rawChoice of payload.choices) {
+      if (!isRecord(rawChoice)) continue;
+
+      const index = typeof rawChoice.index === 'number' ? rawChoice.index : 0;
+      const choice = this.getChoice(index);
+      const delta = rawChoice.delta;
+
+      if (isRecord(delta)) {
+        if (typeof delta.role === 'string') choice.role = delta.role;
+        const reasoning = typeof delta.reasoning_content === 'string'
+          ? delta.reasoning_content
+          : typeof delta.reasoning === 'string'
+            ? delta.reasoning
+            : undefined;
+        if (reasoning) choice.reasoning.push(reasoning);
+        if (typeof delta.content === 'string') choice.content.push(delta.content);
+
+        if (Array.isArray(delta.tool_calls)) {
+          for (const rawToolCall of delta.tool_calls) {
+            if (!isRecord(rawToolCall)) continue;
+            const toolIndex = typeof rawToolCall.index === 'number' ? rawToolCall.index : choice.toolCalls.size;
+            const toolCall = choice.toolCalls.get(toolIndex) ?? { index: toolIndex };
+            mergeToolCall(toolCall, rawToolCall);
+            choice.toolCalls.set(toolIndex, toolCall);
+          }
+        }
+      }
+
+      if (rawChoice.finish_reason !== undefined && rawChoice.finish_reason !== null) {
+        choice.finishReason = String(rawChoice.finish_reason);
+      }
+      if (rawChoice.stop_reason !== undefined && rawChoice.stop_reason !== null) {
+        choice.stopReason = String(rawChoice.stop_reason);
+      }
+      if (rawChoice.matched_stop !== undefined && rawChoice.matched_stop !== null) {
+        choice.matchedStop = String(rawChoice.matched_stop);
+      }
+    }
+  }
+
+  private toCompactLog(): CompactChatStreamLog {
+    const choices = [...this.choices.values()]
+      .sort((left, right) => left.index - right.index)
+      .map((choice) => {
+        const compactChoice: CompactChoice = {
+          index: choice.index,
+        };
+        const reasoning = choice.reasoning.join('');
+        const content = choice.content.join('');
+        const toolCalls = [...choice.toolCalls.values()].sort((left, right) => left.index - right.index);
+
+        if (choice.role) compactChoice.role = choice.role;
+        if (reasoning) compactChoice.reasoning = reasoning;
+        if (content) compactChoice.content = content;
+        if (toolCalls.length > 0) compactChoice.tool_calls = toolCalls;
+        if (choice.finishReason) compactChoice.finish_reason = choice.finishReason;
+        if (choice.stopReason) compactChoice.stop_reason = choice.stopReason;
+        if (choice.matchedStop) compactChoice.matched_stop = choice.matchedStop;
+
+        return compactChoice;
+      });
+
+    return {
+      format: COMPACT_CHAT_STREAM_FORMAT,
+      id: this.id,
+      object: this.object,
+      created: this.created,
+      model: this.model,
+      choices,
+      usage: this.usage,
+      stream: {
+        network_chunk_count: this.networkChunkCount,
+        data_frame_count: this.dataFrameCount,
+        done: this.done,
+      },
+    };
+  }
+}

server/tests/integration/admin.test.ts

@@ -34,6 +34,17 @@ describe('Admin API - User Management', () => {
       expect(response.body.email).toBe(userData.email);
       expect(response.body).toHaveProperty('api_key');
       expect(response.body.api_key).toMatch(/^sk-/);
+      expect(response.body.copy_reasoning_to_reasoning_content).toBe(false);
+    });
+
+    it('should create a user with reasoning compatibility enabled', async () => {
+      const response = await admin.post('/admin/users').send({
+        name: 'Reasoning Compat User',
+        copy_reasoning_to_reasoning_content: true,
+      });
+
+      expect(response.status).toBe(201);
+      expect(response.body.copy_reasoning_to_reasoning_content).toBe(true);
     });
 
     it('should create a user with a manually supplied api key', async () => {
@@ -111,6 +122,15 @@ describe('Admin API - User Management', () => {
       expect(response.body.api_key).toBe('legacy-updated-key-001');
     });
 
+    it('should update user reasoning compatibility setting', async () => {
+      const response = await admin
+        .put(`/admin/users/${userId}`)
+        .send({ copy_reasoning_to_reasoning_content: true });
+
+      expect(response.status).toBe(200);
+      expect(response.body.copy_reasoning_to_reasoning_content).toBe(true);
+    });
+
     it('should return 404 for non-existent user', async () => {
       const response = await admin.put('/admin/users/99999').send({ name: 'Test' });
       

server/tests/integration/api.test.ts

@@ -5,6 +5,7 @@ import { initDb } from '../../src/config/database';
 import { RequestLogService } from '../../src/services/RequestLogService';
 import { AnalyticsService } from '../../src/services/AnalyticsService';
 import { createAdminClient } from '../utils/adminClient';
+import { getLocalDateKey, getUtcTimestamp } from '../../src/utils/time';
 
 describe('Auth & Proxy API', () => {
   let app: ReturnType<typeof createTestApp>;
@@ -164,6 +165,10 @@ describe('Auth & Proxy API', () => {
     });
 
     it('should expose chart-friendly analytics endpoints', async () => {
+      const primaryDate = getLocalDateKey();
+      const secondarySourceDate = new Date(Date.now() - 24 * 60 * 60 * 1000);
+      const secondaryDate = getLocalDateKey(secondarySourceDate);
+
       AnalyticsService.logRequest({
         user_id: 7001,
         backend_id: backendId,
@@ -175,8 +180,8 @@ describe('Auth & Proxy API', () => {
         total_tokens: 240,
         response_time_ms: 380,
         status_code: 200,
-        local_date: '2026-03-10',
-        created_at: '2026-03-10T03:00:00.000Z',
+        local_date: primaryDate,
+        created_at: getUtcTimestamp(),
       });
 
       AnalyticsService.logRequest({
@@ -191,8 +196,8 @@ describe('Auth & Proxy API', () => {
         response_time_ms: 510,
         status_code: 500,
         error_message: 'synthetic-error',
-        local_date: '2026-03-11',
-        created_at: '2026-03-11T03:00:00.000Z',
+        local_date: secondaryDate,
+        created_at: getUtcTimestamp(secondarySourceDate),
       });
 
       const [dailyTotals, backendQuality, modelTrends, histogram, boxPlot] = await Promise.all([
@@ -221,7 +226,7 @@ describe('Auth & Proxy API', () => {
 
       expect(boxPlot.status).toBe(200);
       expect(Array.isArray(boxPlot.body)).toBe(true);
-      expect(boxPlot.body.some((row: any) => row.date === '2026-03-10' && row.median === 120)).toBe(true);
+      expect(boxPlot.body.some((row: any) => row.date === primaryDate && row.median === 120)).toBe(true);
     });
 
     it('should expose dashboard summary data for the ops cockpit', async () => {

server/tests/integration/routing.test.ts

@@ -280,6 +280,8 @@ describe('OpenAI Compatible Backend Integration', () => {
   });
 
   afterEach(async () => {
+    delete process.env.MODEL_LIST_INCLUDE_ROUTING_METADATA;
+
     if (mockServer) {
       await new Promise<void>(resolve => mockServer.close(resolve));
       mockServer = undefined;
@@ -374,6 +376,57 @@ describe('OpenAI Compatible Backend Integration', () => {
       expect(receivedAuthorization).toBe('Bearer upstream-secret-key');
       expect(receivedAuthorization).not.toBe(`Bearer ${userApiKey}`);
     });
+
+    it('should preserve multimodal image messages and chat template kwargs when proxying', async () => {
+      let receivedBody: any;
+      const { server, port } = createMockBackend({
+        onRequest: (req) => {
+          if (req.path === '/v1/chat/completions') {
+            receivedBody = req.body;
+          }
+        },
+        modelsResponse: [{ id: 'vision-test-model', object: 'model' }],
+      });
+      mockServer = server;
+      mockPort = port;
+
+      const userResponse = await admin.post('/admin/users').send({ name: 'Vision Payload User 6-6' });
+      const userApiKey = userResponse.body.api_key;
+      const userId = userResponse.body.id;
+
+      const backendResponse = await admin.post('/admin/backends').send({
+        name: 'Vision Payload Backend 6-6',
+        base_url: `http://localhost:${mockPort}`,
+      });
+      const backendId = backendResponse.body.id;
+
+      await admin
+        .post('/admin/permissions')
+        .send({ user_id: userId, backend_id: backendId });
+
+      const imageDataUrl = `data:image/jpeg;base64,${'a'.repeat(128 * 1024)}`;
+      const payload = {
+        model: 'vision-test-model',
+        messages: [{
+          role: 'user',
+          content: [
+            { type: 'text', text: '이 이미지는 무엇인가요?' },
+            { type: 'image_url', image_url: { url: imageDataUrl } },
+          ],
+        }],
+        chat_template_kwargs: {
+          enable_thinking: true,
+        },
+      };
+
+      const response = await request(app)
+        .post('/v1/chat/completions')
+        .set('Authorization', `Bearer ${userApiKey}`)
+        .send(payload);
+
+      expect(response.status).toBe(200);
+      expect(receivedBody).toEqual(payload);
+    });
   });
 
   describe('Scenario 7: Models endpoint routing', () => {
@@ -568,5 +621,271 @@ describe('OpenAI Compatible Backend Integration', () => {
       expect(receivedModel).toBe('fallback-model');
       expect(response.body.model).toBe('fallback-model');
     });
+
+    it('should follow force rewrite chains before upstream forwarding', async () => {
+      let receivedModel: string | undefined;
+      const { server, port } = createMockBackend({
+        onRequest: (req) => {
+          if (req.path === '/v1/chat/completions') {
+            receivedModel = req.body.model;
+          }
+        },
+        chatResponse: {
+          id: 'force-chain-success',
+          model: 'chain-final-c',
+          choices: [{ index: 0, message: { role: 'assistant', content: 'chain' }, finish_reason: 'stop' }],
+          usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 },
+        },
+        modelsResponse: [{ id: 'chain-final-c', object: 'model' }],
+      });
+      mockServer = server;
+      mockPort = port;
+
+      const userResponse = await admin.post('/admin/users').send({ name: 'Force Chain User 8-10' });
+      const userApiKey = userResponse.body.api_key;
+      const userId = userResponse.body.id;
+      const backendResponse = await admin.post('/admin/backends').send({
+        name: 'Force Chain Backend 8-10',
+        base_url: `http://localhost:${port}`,
+      });
+
+      await admin.post('/admin/permissions').send({ user_id: userId, backend_id: backendResponse.body.id });
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'chain-start-a', target_model: 'chain-mid-b', force: true })).status).toBe(201);
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'chain-mid-b', target_model: 'chain-final-c', force: true })).status).toBe(201);
+
+      const response = await request(app)
+        .post('/v1/chat/completions')
+        .set('Authorization', `Bearer ${userApiKey}`)
+        .send({ model: 'chain-start-a', messages: [] });
+
+      expect(response.status).toBe(200);
+      expect(receivedModel).toBe('chain-final-c');
+    });
+
+    it('should continue a mixed chain only when the current fallback model is unavailable', async () => {
+      let unavailableReceivedModel: string | undefined;
+      const unavailableBackend = createMockBackend({
+        onRequest: (req) => {
+          if (req.path === '/v1/chat/completions') {
+            unavailableReceivedModel = req.body.model;
+          }
+        },
+        chatResponse: {
+          id: 'mixed-chain-unavailable',
+          model: 'mixed-final-e',
+          choices: [{ index: 0, message: { role: 'assistant', content: 'fallback-chain' }, finish_reason: 'stop' }],
+          usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 },
+        },
+        modelsResponse: [{ id: 'mixed-final-e', object: 'model' }],
+      });
+      mockServer = unavailableBackend.server;
+      mockPort = unavailableBackend.port;
+
+      const unavailableUser = await admin.post('/admin/users').send({ name: 'Mixed Chain Missing User 8-11' });
+      const unavailableBackendResponse = await admin.post('/admin/backends').send({
+        name: 'Mixed Chain Missing Backend 8-11',
+        base_url: `http://localhost:${unavailableBackend.port}`,
+      });
+      await admin.post('/admin/permissions').send({ user_id: unavailableUser.body.id, backend_id: unavailableBackendResponse.body.id });
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'mixed-missing-a', target_model: 'mixed-missing-b', force: true })).status).toBe(201);
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'mixed-missing-b', target_model: 'mixed-missing-c', force: true })).status).toBe(201);
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'mixed-missing-c', target_model: 'mixed-missing-d', force: false })).status).toBe(201);
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'mixed-missing-d', target_model: 'mixed-final-e', force: true })).status).toBe(201);
+
+      const unavailableResponse = await request(app)
+        .post('/v1/chat/completions')
+        .set('Authorization', `Bearer ${unavailableUser.body.api_key}`)
+        .send({ model: 'mixed-missing-a', messages: [] });
+
+      expect(unavailableResponse.status).toBe(200);
+      expect(unavailableReceivedModel).toBe('mixed-final-e');
+
+      let availableReceivedModel: string | undefined;
+      const availableBackend = createMockBackend({
+        onRequest: (req) => {
+          if (req.path === '/v1/chat/completions') {
+            availableReceivedModel = req.body.model;
+          }
+        },
+        chatResponse: {
+          id: 'mixed-chain-available',
+          model: 'mixed-available-c',
+          choices: [{ index: 0, message: { role: 'assistant', content: 'available-chain' }, finish_reason: 'stop' }],
+          usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 },
+        },
+        modelsResponse: [{ id: 'mixed-available-c', object: 'model' }, { id: 'mixed-available-e', object: 'model' }],
+      });
+
+      try {
+        const availableUser = await admin.post('/admin/users').send({ name: 'Mixed Chain Available User 8-12' });
+        const availableBackendResponse = await admin.post('/admin/backends').send({
+          name: 'Mixed Chain Available Backend 8-12',
+          base_url: `http://localhost:${availableBackend.port}`,
+        });
+        await admin.post('/admin/permissions').send({ user_id: availableUser.body.id, backend_id: availableBackendResponse.body.id });
+        expect((await admin.post('/admin/model-rewrites').send({ source_model: 'mixed-available-a', target_model: 'mixed-available-b', force: true })).status).toBe(201);
+        expect((await admin.post('/admin/model-rewrites').send({ source_model: 'mixed-available-b', target_model: 'mixed-available-c', force: true })).status).toBe(201);
+        expect((await admin.post('/admin/model-rewrites').send({ source_model: 'mixed-available-c', target_model: 'mixed-available-d', force: false })).status).toBe(201);
+        expect((await admin.post('/admin/model-rewrites').send({ source_model: 'mixed-available-d', target_model: 'mixed-available-e', force: true })).status).toBe(201);
+
+        const availableResponse = await request(app)
+          .post('/v1/chat/completions')
+          .set('Authorization', `Bearer ${availableUser.body.api_key}`)
+          .send({ model: 'mixed-available-a', messages: [] });
+
+        expect(availableResponse.status).toBe(200);
+        expect(availableReceivedModel).toBe('mixed-available-c');
+      } finally {
+        await new Promise<void>((resolve) => availableBackend.server.close(() => resolve()));
+      }
+    });
+
+    it('should expose only requestable native models and rewrite aliases from /v1/models', async () => {
+      const allowedBackend = createMockBackend({
+        modelsResponse: [
+          { id: 'models-visible-final', object: 'model' },
+          { id: 'models-native-forced-away', object: 'model' },
+        ],
+      });
+      const deniedBackend = createMockBackend({
+        modelsResponse: [{ id: 'models-denied-final', object: 'model' }],
+      });
+      mockServer = allowedBackend.server;
+      mockPort = allowedBackend.port;
+
+      try {
+        const userResponse = await admin.post('/admin/users').send({ name: 'Requestable Models User 8-13' });
+        const allowedBackendResponse = await admin.post('/admin/backends').send({
+          name: 'Requestable Models Allowed Backend 8-13',
+          base_url: `http://localhost:${allowedBackend.port}`,
+        });
+        await admin.post('/admin/backends').send({
+          name: 'Requestable Models Denied Backend 8-13',
+          base_url: `http://localhost:${deniedBackend.port}`,
+        });
+        await admin.post('/admin/permissions').send({ user_id: userResponse.body.id, backend_id: allowedBackendResponse.body.id });
+        expect((await admin.post('/admin/model-rewrites').send({ source_model: 'models-visible-alias', target_model: 'models-visible-final', force: true })).status).toBe(201);
+        expect((await admin.post('/admin/model-rewrites').send({ source_model: 'models-missing-alias', target_model: 'models-missing-final', force: true })).status).toBe(201);
+        expect((await admin.post('/admin/model-rewrites').send({ source_model: 'models-denied-alias', target_model: 'models-denied-final', force: true })).status).toBe(201);
+        expect((await admin.post('/admin/model-rewrites').send({ source_model: 'models-native-forced-away', target_model: 'models-missing-final', force: true })).status).toBe(201);
+
+        const response = await request(app)
+          .get('/v1/models')
+          .set('Authorization', `Bearer ${userResponse.body.api_key}`);
+
+        expect(response.status).toBe(200);
+        const ids = response.body.data.map((item: any) => item.id);
+        expect(ids).toContain('models-visible-final');
+        expect(ids).toContain('models-visible-alias');
+        expect(ids).not.toContain('models-missing-alias');
+        expect(ids).not.toContain('models-denied-alias');
+        expect(ids).not.toContain('models-native-forced-away');
+        expect(response.body.data.every((item: any) => item.kyush_router === undefined)).toBe(true);
+      } finally {
+        await new Promise<void>((resolve) => deniedBackend.server.close(() => resolve()));
+      }
+    });
+
+    it('should include kyush_router metadata for /v1/models when enabled', async () => {
+      process.env.MODEL_LIST_INCLUDE_ROUTING_METADATA = 'true';
+
+      const { server, port } = createMockBackend({
+        modelsResponse: [
+          { id: 'metadata-native', object: 'model' },
+          { id: 'metadata-final', object: 'model' },
+          { id: 'metadata-skip-current', object: 'model' },
+        ],
+      });
+      mockServer = server;
+      mockPort = port;
+
+      const userResponse = await admin.post('/admin/users').send({ name: 'Model Metadata User 8-14' });
+      const backendResponse = await admin.post('/admin/backends').send({
+        name: 'Model Metadata Backend 8-14',
+        base_url: `http://localhost:${port}`,
+      });
+      await admin.post('/admin/permissions').send({ user_id: userResponse.body.id, backend_id: backendResponse.body.id });
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'metadata-alias-a', target_model: 'metadata-alias-b', force: true })).status).toBe(201);
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'metadata-alias-b', target_model: 'metadata-missing-c', force: true })).status).toBe(201);
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'metadata-missing-c', target_model: 'metadata-final', force: false })).status).toBe(201);
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'metadata-skip-current', target_model: 'metadata-skip-target', force: false })).status).toBe(201);
+
+      const response = await request(app)
+        .get('/v1/models')
+        .set('Authorization', `Bearer ${userResponse.body.api_key}`);
+
+      expect(response.status).toBe(200);
+      expect(response.body.data.length).toBeGreaterThan(0);
+      expect(response.body.data.every((item: any) => item.kyush_router && !('backend_ids' in item.kyush_router))).toBe(true);
+
+      const native = response.body.data.find((item: any) => item.id === 'metadata-native');
+      expect(native.kyush_router).toEqual({
+        requested_model: 'metadata-native',
+        routed_model: 'metadata-native',
+        was_rewritten: false,
+        rule_type: 'none',
+        rewrite_path: [],
+      });
+
+      const alias = response.body.data.find((item: any) => item.id === 'metadata-alias-a');
+      expect(alias.kyush_router).toEqual({
+        requested_model: 'metadata-alias-a',
+        routed_model: 'metadata-final',
+        was_rewritten: true,
+        rule_type: 'chain',
+        rewrite_path: [
+          { source_model: 'metadata-alias-a', target_model: 'metadata-alias-b', mode: 'force' },
+          { source_model: 'metadata-alias-b', target_model: 'metadata-missing-c', mode: 'force' },
+          { source_model: 'metadata-missing-c', target_model: 'metadata-final', mode: 'fallback' },
+        ],
+      });
+
+      const skippedFallback = response.body.data.find((item: any) => item.id === 'metadata-skip-current');
+      expect(skippedFallback.kyush_router).toEqual({
+        requested_model: 'metadata-skip-current',
+        routed_model: 'metadata-skip-current',
+        was_rewritten: false,
+        rule_type: 'none',
+        rewrite_path: [],
+      });
+    });
+
+    it('should reject active rewrite cycles while allowing inactive cycles until activation', async () => {
+      const selfLoop = await admin.post('/admin/model-rewrites').send({
+        source_model: 'cycle-self-a',
+        target_model: 'cycle-self-a',
+        force: true,
+      });
+      expect(selfLoop.status).toBe(409);
+      expect(selfLoop.body.error).toBe('Model rewrite cycle detected');
+
+      expect((await admin.post('/admin/model-rewrites').send({ source_model: 'cycle-active-a', target_model: 'cycle-active-b', force: true })).status).toBe(201);
+      const activeCycle = await admin.post('/admin/model-rewrites').send({
+        source_model: 'cycle-active-b',
+        target_model: 'cycle-active-a',
+        force: true,
+      });
+      expect(activeCycle.status).toBe(409);
+
+      const inactiveA = await admin.post('/admin/model-rewrites').send({
+        source_model: 'cycle-inactive-a',
+        target_model: 'cycle-inactive-b',
+        is_active: false,
+        force: true,
+      });
+      const inactiveB = await admin.post('/admin/model-rewrites').send({
+        source_model: 'cycle-inactive-b',
+        target_model: 'cycle-inactive-a',
+        is_active: false,
+        force: true,
+      });
+      expect(inactiveA.status).toBe(201);
+      expect(inactiveB.status).toBe(201);
+
+      const activation = await admin.put(`/admin/model-rewrites/${inactiveA.body.id}`).send({ is_active: true });
+      expect(activation.status).toBe(200);
+      const secondActivation = await admin.put(`/admin/model-rewrites/${inactiveB.body.id}`).send({ is_active: true });
+      expect(secondActivation.status).toBe(409);
+    });
   });
 });

server/tests/integration/scripts.test.ts

@@ -3,6 +3,7 @@ import request from 'supertest';
 import { createTestApp } from '../utils/testApp';
 import { initDb } from '../../src/config/database';
 import { createAdminClient } from '../utils/adminClient';
+import { createMockBackend } from '../utils/mockBackend';
 
 describe('Script API Endpoints', () => {
   let app: ReturnType<typeof createTestApp>;
@@ -10,6 +11,8 @@ describe('Script API Endpoints', () => {
   let userId: number;
   let backendId: number;
   let scriptId: number;
+  let backendServer: ReturnType<typeof createMockBackend>['server'];
+  let receivedChatBody: any;
 
   beforeAll(() => {
     initDb();
@@ -22,12 +25,22 @@ describe('Script API Endpoints', () => {
 
   // Setup: Create user and backend for testing
   beforeAll(async () => {
+    const mockBackend = createMockBackend({
+      onRequest: (req) => {
+        if (req.path === '/v1/chat/completions') {
+          receivedChatBody = req.body;
+        }
+      },
+      modelsResponse: [{ id: 'test-model', object: 'model' }],
+    });
+    backendServer = mockBackend.server;
+
     const userResponse = await admin.post('/admin/users').send({ name: 'Script Test User' });
     userId = userResponse.body.id;
 
     const backendResponse = await admin.post('/admin/backends').send({
       name: 'Script Test Backend',
-      base_url: 'http://localhost:8006/v1'
+      base_url: `http://localhost:${mockBackend.port}/v1`
     });
     backendId = backendResponse.body.id;
   });
@@ -36,6 +49,7 @@ describe('Script API Endpoints', () => {
     // Cleanup: Delete created resources
     await admin.delete(`/admin/users/${userId}`);
     await admin.delete(`/admin/backends/${backendId}`);
+    await new Promise<void>((resolve) => backendServer.close(() => resolve()));
   });
 
   describe('GET /admin/scripts', () => {
@@ -369,9 +383,11 @@ export const onResponse = (context) => {
         name: 'Integration Test Script',
         script_code: `
 export const onRequest = (context) => {
-  const body = JSON.parse(context.request.body);
+  const body = typeof context.request.body === 'string'
+    ? JSON.parse(context.request.body)
+    : context.request.body;
   body.messages.push({ role: 'system', content: 'Modified by middleware' });
-  context.request.body = JSON.stringify(body);
+  context.request.body = body;
   return context;
 };
 
@@ -402,6 +418,8 @@ export const onResponse = (context) => {
     it('should execute script when making request to backend', async () => {
       // This test verifies that scripts are executed during actual API calls
       // The script should modify the request before forwarding to backend
+      receivedChatBody = undefined;
+
       const response = await request(app)
         .post('/v1/chat/completions')
         .set('Authorization', `Bearer ${userApiKey}`)
@@ -410,9 +428,8 @@ export const onResponse = (context) => {
           messages: [{ role: 'user', content: 'Hello' }]
         });
 
-      // Request will fail (502) because backend is not actually running,
-      // but we can verify the script was executed by checking logs
-      expect(response.status).toBe(502); // Backend unreachable
+      expect(response.status).toBe(200);
+      expect(receivedChatBody?.messages).toContainEqual({ role: 'system', content: 'Modified by middleware' });
       
       // Check that request was logged with script execution
       const analyticsResponse = await admin.get('/admin/analytics/requests?limit=10');

server/tests/integration/streaming.test.ts

@@ -57,6 +57,8 @@ describe('Streaming Response Proxying', () => {
   });
 
   afterEach(async () => {
+    delete process.env.DETAIL_STREAM_LOG_MODE;
+
     if (mockServer) {
       await new Promise<void>(resolve => mockServer.close(resolve));
       mockServer = undefined;
@@ -73,7 +75,7 @@ describe('Streaming Response Proxying', () => {
     }
   });
 
-  async function setupUserAndBackend(mockPort: number) {
+  async function setupUserAndBackend(mockPort: number, options: { detailLogging?: boolean; copyReasoning?: boolean } = {}) {
     // Deactivate all existing backends to ensure only our mock backend is selected
     const allBackendsResponse = await admin.get('/admin/backends');
     for (const backend of allBackendsResponse.body) {
@@ -82,7 +84,11 @@ describe('Streaming Response Proxying', () => {
       }
     }
 
-    const userResponse = await admin.post('/admin/users').send({ name: `Stream Test User ${Date.now()}` });
+    const userResponse = await admin.post('/admin/users').send({
+      name: `Stream Test User ${Date.now()}`,
+      detail_logging: options.detailLogging,
+      copy_reasoning_to_reasoning_content: options.copyReasoning,
+    });
     const userApiKey = userResponse.body.api_key;
     const userId = userResponse.body.id;
 
@@ -97,6 +103,15 @@ describe('Streaming Response Proxying', () => {
     return { userApiKey, userId, backendId };
   }
 
+  async function createUserForBackend(backendId: number, options: { copyReasoning?: boolean } = {}) {
+    const userResponse = await admin.post('/admin/users').send({
+      name: `Stream Compat User ${Date.now()} ${Math.random()}`,
+      copy_reasoning_to_reasoning_content: options.copyReasoning,
+    });
+    await admin.post('/admin/permissions').send({ user_id: userResponse.body.id, backend_id: backendId });
+    return { userApiKey: userResponse.body.api_key as string, userId: userResponse.body.id as number };
+  }
+
   it('should return Content-Type text/event-stream for stream requests', async () => {
     const { server, port } = createMockBackend({
       streamChunks: sampleStreamChunks,
@@ -237,4 +252,244 @@ describe('Streaming Response Proxying', () => {
     expect(receivedBody).toBeDefined();
     expect(receivedBody.stream).toBe(true);
   });
+
+  it('should store compact stream logs by default when detail logging is enabled', async () => {
+    process.env.DETAIL_STREAM_LOG_MODE = 'compact';
+    const compactStreamChunks = [
+      JSON.stringify({
+        id: 'chatcmpl-compact-1',
+        object: 'chat.completion.chunk',
+        created: 1776916142,
+        model: 'mock-model',
+        choices: [{ index: 0, delta: { role: 'assistant', reasoning: 'Think' }, finish_reason: null }],
+      }),
+      JSON.stringify({
+        id: 'chatcmpl-compact-1',
+        object: 'chat.completion.chunk',
+        created: 1776916142,
+        model: 'mock-model',
+        choices: [{ index: 0, delta: { reasoning: ' first. ' }, finish_reason: null }],
+      }),
+      JSON.stringify({
+        id: 'chatcmpl-compact-1',
+        object: 'chat.completion.chunk',
+        created: 1776916142,
+        model: 'mock-model',
+        choices: [{
+          index: 0,
+          delta: {
+            tool_calls: [{
+              id: 'tool-1',
+              type: 'function',
+              index: 0,
+              function: { name: 'search_web', arguments: '' },
+            }],
+          },
+          finish_reason: null,
+        }],
+      }),
+      JSON.stringify({
+        id: 'chatcmpl-compact-1',
+        object: 'chat.completion.chunk',
+        created: 1776916142,
+        model: 'mock-model',
+        choices: [{ index: 0, delta: { tool_calls: [{ index: 0, function: { arguments: '{"query":"' } }] }, finish_reason: null }],
+      }),
+      JSON.stringify({
+        id: 'chatcmpl-compact-1',
+        object: 'chat.completion.chunk',
+        created: 1776916142,
+        model: 'mock-model',
+        choices: [{ index: 0, delta: { tool_calls: [{ index: 0, function: { arguments: 'apple"}' } }] }, finish_reason: null }],
+      }),
+      JSON.stringify({
+        id: 'chatcmpl-compact-1',
+        object: 'chat.completion.chunk',
+        created: 1776916142,
+        model: 'mock-model',
+        choices: [{ index: 0, delta: { content: 'Hello' }, finish_reason: null }],
+      }),
+      JSON.stringify({
+        id: 'chatcmpl-compact-1',
+        object: 'chat.completion.chunk',
+        created: 1776916142,
+        model: 'mock-model',
+        choices: [{ index: 0, delta: { content: ' world' }, finish_reason: 'stop', stop_reason: 106 }],
+      }),
+      JSON.stringify({
+        id: 'chatcmpl-compact-1',
+        object: 'chat.completion.chunk',
+        created: 1776916142,
+        model: 'mock-model',
+        choices: [],
+        usage: { prompt_tokens: 7, completion_tokens: 5, total_tokens: 12 },
+      }),
+    ];
+    const { server, port } = createMockBackend({
+      streamChunks: compactStreamChunks,
+      modelsResponse: [{ id: 'mock-model', object: 'model' }],
+    });
+    mockServer = server;
+
+    const { userApiKey, userId } = await setupUserAndBackend(port, { detailLogging: true });
+
+    const response = await request(app)
+      .post('/v1/chat/completions')
+      .set('Authorization', `Bearer ${userApiKey}`)
+      .send({
+        model: 'mock-model',
+        messages: [{ role: 'user', content: 'Hello' }],
+        stream: true,
+      });
+
+    expect(response.status).toBe(200);
+    expect(response.headers['content-type']).toMatch(/text\/event-stream/);
+
+    const logsResponse = await admin.get(`/admin/analytics/requests?limit=1&userId=${userId}&detailLogged=1`);
+    expect(logsResponse.body.rows).toHaveLength(1);
+
+    const responseBody = logsResponse.body.rows[0].response_body;
+    expect(responseBody).not.toContain('data: ');
+
+    const parsed = JSON.parse(responseBody);
+    expect(parsed.format).toBe('kyush.chat_stream.compact.v1');
+    expect(parsed.id).toBe('chatcmpl-compact-1');
+    expect(parsed.model).toBe('mock-model');
+    expect(parsed.choices[0].reasoning).toBe('Think first. ');
+    expect(parsed.choices[0].content).toBe('Hello world');
+    expect(parsed.choices[0].tool_calls[0].function.arguments).toBe('{"query":"apple"}');
+    expect(parsed.choices[0].finish_reason).toBe('stop');
+    expect(parsed.choices[0].stop_reason).toBe('106');
+    expect(parsed.usage.total_tokens).toBe(12);
+    expect(parsed.stream.done).toBe(true);
+  });
+
+  it('should keep raw SSE logs when DETAIL_STREAM_LOG_MODE=raw', async () => {
+    process.env.DETAIL_STREAM_LOG_MODE = 'raw';
+    const { server, port } = createMockBackend({
+      streamChunks: sampleStreamChunks,
+      modelsResponse: [{ id: 'mock-model', object: 'model' }],
+    });
+    mockServer = server;
+
+    const { userApiKey, userId } = await setupUserAndBackend(port, { detailLogging: true });
+
+    await request(app)
+      .post('/v1/chat/completions')
+      .set('Authorization', `Bearer ${userApiKey}`)
+      .send({
+        model: 'mock-model',
+        messages: [{ role: 'user', content: 'Hello' }],
+        stream: true,
+      });
+
+    const logsResponse = await admin.get(`/admin/analytics/requests?limit=1&userId=${userId}&detailLogged=1`);
+    expect(logsResponse.body.rows).toHaveLength(1);
+    expect(logsResponse.body.rows[0].response_body).toContain('data: ');
+    expect(logsResponse.body.rows[0].response_body).toContain('data: [DONE]');
+  });
+
+  it('should copy streaming reasoning to reasoning_content only for users with compatibility enabled', async () => {
+    const reasoningStreamChunks = [
+      JSON.stringify({
+        id: 'chatcmpl-reasoning-1',
+        object: 'chat.completion.chunk',
+        model: 'mock-model',
+        choices: [{ index: 0, delta: { role: 'assistant', reasoning: 'Think once.' }, finish_reason: null }],
+      }),
+      JSON.stringify({
+        id: 'chatcmpl-reasoning-1',
+        object: 'chat.completion.chunk',
+        model: 'mock-model',
+        choices: [{ index: 0, delta: { reasoning: ' Keep original.', reasoning_content: 'Existing wins.' }, finish_reason: null }],
+      }),
+      JSON.stringify({
+        id: 'chatcmpl-reasoning-1',
+        object: 'chat.completion.chunk',
+        model: 'mock-model',
+        choices: [{ index: 0, delta: { content: 'Hello' }, finish_reason: 'stop' }],
+      }),
+    ];
+    const { server, port } = createMockBackend({
+      streamChunks: reasoningStreamChunks,
+      modelsResponse: [{ id: 'mock-model', object: 'model' }],
+    });
+    mockServer = server;
+
+    const { backendId, userApiKey: defaultUserApiKey } = await setupUserAndBackend(port);
+    const { userApiKey: compatUserApiKey } = await createUserForBackend(backendId, { copyReasoning: true });
+
+    const defaultResponse = await request(app)
+      .post('/v1/chat/completions')
+      .set('Authorization', `Bearer ${defaultUserApiKey}`)
+      .send({
+        model: 'mock-model',
+        messages: [{ role: 'user', content: 'Hello' }],
+        stream: true,
+      });
+
+    const compatResponse = await request(app)
+      .post('/v1/chat/completions')
+      .set('Authorization', `Bearer ${compatUserApiKey}`)
+      .send({
+        model: 'mock-model',
+        messages: [{ role: 'user', content: 'Hello' }],
+        stream: true,
+      });
+
+    const defaultFirstChunk = JSON.parse(defaultResponse.text.split('\n').find((line: string) => line.startsWith('data: {'))!.replace('data: ', ''));
+    expect(defaultFirstChunk.choices[0].delta.reasoning).toBe('Think once.');
+    expect(defaultFirstChunk.choices[0].delta.reasoning_content).toBeUndefined();
+
+    const compatDataLines = compatResponse.text.split('\n').filter((line: string) => line.startsWith('data: {'));
+    const compatFirstChunk = JSON.parse(compatDataLines[0].replace('data: ', ''));
+    expect(compatFirstChunk.choices[0].delta.reasoning).toBe('Think once.');
+    expect(compatFirstChunk.choices[0].delta.reasoning_content).toBe('Think once.');
+
+    const compatSecondChunk = JSON.parse(compatDataLines[1].replace('data: ', ''));
+    expect(compatSecondChunk.choices[0].delta.reasoning).toBe(' Keep original.');
+    expect(compatSecondChunk.choices[0].delta.reasoning_content).toBe('Existing wins.');
+    expect(compatResponse.text).toContain('data: [DONE]');
+  });
+
+  it('should copy non-stream reasoning to reasoning_content only for users with compatibility enabled', async () => {
+    const { server, port } = createMockBackend({
+      chatResponse: {
+        id: 'non-stream-reasoning-1',
+        model: 'mock-model',
+        choices: [{
+          index: 0,
+          message: { role: 'assistant', content: 'Hello', reasoning: 'Think non-stream.' },
+          finish_reason: 'stop',
+        }],
+        usage: { prompt_tokens: 10, completion_tokens: 5, total_tokens: 15 },
+      },
+      modelsResponse: [{ id: 'mock-model', object: 'model' }],
+    });
+    mockServer = server;
+
+    const { backendId, userApiKey: defaultUserApiKey } = await setupUserAndBackend(port);
+    const { userApiKey: compatUserApiKey } = await createUserForBackend(backendId, { copyReasoning: true });
+
+    const defaultResponse = await request(app)
+      .post('/v1/chat/completions')
+      .set('Authorization', `Bearer ${defaultUserApiKey}`)
+      .send({
+        model: 'mock-model',
+        messages: [{ role: 'user', content: 'Hello' }],
+      });
+
+    const compatResponse = await request(app)
+      .post('/v1/chat/completions')
+      .set('Authorization', `Bearer ${compatUserApiKey}`)
+      .send({
+        model: 'mock-model',
+        messages: [{ role: 'user', content: 'Hello' }],
+      });
+
+    expect(defaultResponse.body.choices[0].message.reasoning).toBe('Think non-stream.');
+    expect(defaultResponse.body.choices[0].message.reasoning_content).toBeUndefined();
+    expect(compatResponse.body.choices[0].message.reasoning).toBe('Think non-stream.');
+    expect(compatResponse.body.choices[0].message.reasoning_content).toBe('Think non-stream.');
+  });
 });

server/tests/unit/request-body.test.ts

@@ -0,0 +1,37 @@
+import express from 'express';
+import request from 'supertest';
+import { describe, expect, it, vi } from 'vitest';
+import { createJsonBodyParser, requestBodyErrorHandler } from '../../src/utils/requestBody';
+import { logger } from '../../src/utils/logger';
+
+describe('request body parser errors', () => {
+  it('should log payload size details when JSON body exceeds the configured limit', async () => {
+    const warnSpy = vi.spyOn(logger, 'warn').mockImplementation(() => {});
+    const app = express();
+
+    app.use(createJsonBodyParser('1kb'));
+    app.use(requestBodyErrorHandler);
+    app.post('/v1/chat/completions', (_req, res) => res.json({ ok: true }));
+
+    const response = await request(app)
+      .post('/v1/chat/completions')
+      .send({ model: 'vision-test-model', data: 'x'.repeat(2048) });
+
+    expect(response.status).toBe(413);
+    expect(response.body.error).toBe('Request body too large');
+    expect(response.body.payload_size_bytes).toBeGreaterThan(1024);
+    expect(response.body.limit_bytes).toBe(1024);
+
+    expect(warnSpy).toHaveBeenCalledTimes(1);
+    expect(warnSpy.mock.calls[0][0]).toContain('Request body too large');
+    expect(warnSpy.mock.calls[0][0]).toContain('payload=');
+    expect(warnSpy.mock.calls[0][0]).toContain('limit=1.00 KB (1024 B)');
+    expect(warnSpy.mock.calls[0][1]).toMatchObject({
+      payload_size_bytes: expect.any(Number),
+      parser_limit_bytes: 1024,
+      parser_error_type: 'entity.too.large',
+    });
+
+    warnSpy.mockRestore();
+  });
+});

server/tests/utils/mockBackend.ts

@@ -1,4 +1,5 @@
 import express from 'express';
+import { createJsonBodyParser, requestBodyErrorHandler } from '../../src/utils/requestBody';
 
 export interface MockBackendOptions {
   port?: number;
@@ -8,7 +9,7 @@ export interface MockBackendOptions {
     model: string;
     choices: Array<{
       index: number;
-      message: { role: string; content: string };
+      message: { role: string; content: string; reasoning?: string; reasoning_content?: string };
       finish_reason: string;
     }>;
     usage: { prompt_tokens: number; completion_tokens: number; total_tokens: number };
@@ -33,7 +34,8 @@ export function createMockBackend(options: MockBackendOptions = {}) {
   } = options;
 
   const app = express();
-  app.use(express.json());
+  app.use(createJsonBodyParser());
+  app.use(requestBodyErrorHandler);
 
   app.post('/v1/chat/completions', (req, res) => {
     onRequest?.(req);

server/tests/utils/testApp.ts

@@ -10,6 +10,7 @@ import { initRequestLogsDb } from '../../src/config/request-logs-db';
 import { getUtcTimestamp } from '../../src/utils/time';
 import { requireAdminAccess, requireSessionCsrf } from '../../src/utils/adminAuth';
 import { ModelCatalogService } from '../../src/services/ModelCatalogService';
+import { createJsonBodyParser, requestBodyErrorHandler } from '../../src/utils/requestBody';
 
 export function createTestApp() {
   // Initialize both databases
@@ -22,7 +23,8 @@ export function createTestApp() {
   const app = express();
   
   app.use(cors());
-  app.use(express.json());
+  app.use(createJsonBodyParser());
+  app.use(requestBodyErrorHandler);
   
   app.use('/admin/auth', adminAuthRoutes);
   app.use('/admin/analytics', requireAdminAccess, requireSessionCsrf, analyticsRoutes);

shared/types.ts

@@ -5,6 +5,7 @@ export interface User {
   email?: string;
   is_active: boolean;
   detail_logging: boolean;
+  copy_reasoning_to_reasoning_content: boolean;
   created_at: string;
   updated_at: string;
 }
@@ -102,6 +103,7 @@ export interface CreateUserData {
   email?: string;
   api_key?: string;
   detail_logging?: boolean;
+  copy_reasoning_to_reasoning_content?: boolean;
 }
 
 export interface CreateBackendData {
@@ -122,6 +124,7 @@ export interface UpdateUserData {
   api_key?: string;
   is_active?: boolean;
   detail_logging?: boolean;
+  copy_reasoning_to_reasoning_content?: boolean;
 }
 
 export interface UpdateBackendData {