Merge caitlin/CRM_ENVR: AI chat ref+health, tips panel, cyberpunk theme, app-shell/bg improvements

ai-server/index.mjs

@@ -202,6 +202,7 @@ Provide concise, actionable sales advice. When asked about a specific job catego
         { role: "user", content: userMessage },
       ],
       stream: false,
+      keep_alive: "30m",
       options: { temperature: 0.7, num_predict: 1024 },
     }),
   })
@@ -576,11 +577,40 @@ async function processRequest(req, res, body, startTime) {
   }
 })
 
+// ── Warm up model on startup ────────────────────────────────────
+// Pre-loads the model into Ollama's memory so the first user
+// request doesn't pay the cold-start penalty.
+async function warmModel() {
+  console.log(`Warming up model ${MODEL}...`)
+  try {
+    const res = await fetch(`${OLLAMA_URL}/api/generate`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      signal: AbortSignal.timeout(300000),
+      body: JSON.stringify({
+        model: MODEL,
+        prompt: "Hello",
+        keep_alive: "30m",
+        options: { num_predict: 1 },
+      }),
+    })
+    if (res.ok) {
+      console.log(`Model ${MODEL} loaded and kept alive for 30 minutes`)
+    } else {
+      console.warn(`Model warm-up returned ${res.status}`)
+    }
+  } catch (err) {
+    console.warn(`Model warm-up failed (will load on first request): ${err.message}`)
+  }
+}
+
 // ── Start ───────────────────────────────────────────────────────
 server.listen(PORT, HOST, () => {
   console.log(`CRM AI server listening on http://${HOST}:${PORT}`)
   console.log(`  Model: ${MODEL}`)
   console.log(`  Ollama: ${OLLAMA_URL}`)
+  // Kick off warm-up in background (don't block server start)
+  warmModel()
 })
 
 initPg()