perf(local-llm): Sprint 5 — request dedup + cache TTLs (P1, P2, P3)

Performance fixes:
- P1: Add fetchingRef guard to fetchAll() — prevents duplicate requests
  from rapid Refresh button clicks or overlapping interval ticks
- P2: Add 5-minute TTL to staticCache (chip, GPU, brew packages) —
  previously cached indefinitely per server process, now refreshes
  after brew upgrades without requiring a restart
- P3: Add 60-second TTL cache for Ollama models disk usage (du command)
  — previously traversed ~/.ollama/models on every 15s refresh cycle,
  now reuses cached value for 60s

__LOCAL_LLMs/dashboard/src/app/api/system/route.ts

@@ -6,12 +6,17 @@ import os from 'os';
 const execAsync = promisify(exec);
 const execFileAsync = promisify(execFile);
 
-// Cache slow commands (chip, gpu, brew don't change during a session)
+// Cache slow commands with TTL
 let staticCache: {
   chip: string;
   gpu: string;
   brewPackages: Array<{ name: string; version: string }>;
+  ts: number;
 } | null = null;
+const STATIC_TTL = 5 * 60 * 1000; // 5 minutes
+
+let ollamaDiskCache: { value: number; ts: number } | null = null;
+const OLLAMA_DISK_TTL = 60 * 1000; // 60 seconds
 
 async function getChipInfo(): Promise<string> {
   try {
@@ -80,16 +85,24 @@ async function getBrewPackages(): Promise<Array<{ name: string; version: string
 }
 
 async function getStaticInfo() {
-  if (staticCache) return staticCache;
+  if (staticCache && Date.now() - staticCache.ts < STATIC_TTL) return staticCache;
   const [chip, gpu, brewPackages] = await Promise.all([
     getChipInfo(),
     getGpuInfo(),
     getBrewPackages(),
   ]);
-  staticCache = { chip, gpu, brewPackages };
+  staticCache = { chip, gpu, brewPackages, ts: Date.now() };
   return staticCache;
 }
 
+async function getCachedOllamaDiskUsage(): Promise<number> {
+  if (ollamaDiskCache && Date.now() - ollamaDiskCache.ts < OLLAMA_DISK_TTL)
+    return ollamaDiskCache.value;
+  const value = await getOllamaModelsDiskUsage();
+  ollamaDiskCache = { value, ts: Date.now() };
+  return value;
+}
+
 // macOS vm_stat gives accurate memory breakdown (os.freemem() excludes reclaimable cache)
 async function getAccurateMemory(): Promise<{
   total: number;
@@ -140,7 +153,7 @@ export async function GET() {
   const [staticInfo, disk, ollamaDisk, memory] = await Promise.all([
     getStaticInfo(),
     getDiskSpace(),
-    getOllamaModelsDiskUsage(),
+    getCachedOllamaDiskUsage(),
     getAccurateMemory(),
   ]);
 

__LOCAL_LLMs/dashboard/src/app/page.tsx

@@ -68,6 +68,7 @@ export default function Dashboard() {
   const [modelfileData, setModelfileData] = useState<Record<string, string>>({});
   const responseRef = useRef<HTMLDivElement>(null);
   const abortRef = useRef<AbortController | null>(null);
+  const fetchingRef = useRef(false);
 
   const addToast = useCallback((message: string, type: Toast['type'] = 'info') => {
     const id = Date.now() + Math.random();
@@ -76,6 +77,8 @@ export default function Dashboard() {
   }, []);
 
   const fetchAll = useCallback(async () => {
+    if (fetchingRef.current) return;
+    fetchingRef.current = true;
     setLoading(true);
     const [oRes, wRes, sRes] = await Promise.allSettled([
       fetch('/api/ollama').then(r => r.json()),
@@ -87,6 +90,7 @@ export default function Dashboard() {
     if (sRes.status === 'fulfilled') setSystem(sRes.value);
     setLastRefresh(new Date());
     setLoading(false);
+    fetchingRef.current = false;
   }, []);
 
   useEffect(() => {