feat(ai-search): optimize dev server, add qdrant boot sync, fix orb overflow

src/lib/qdrant.ts

@@ -1,4 +1,5 @@
 import { QdrantClient } from '@qdrant/js-client-rest';
+import redis from './redis';
 
 const isDockerContainer =
   process.env.IS_DOCKER === 'true' || process.env.HOSTNAME?.includes('klz-app');
@@ -10,15 +11,26 @@ const qdrantApiKey = process.env.QDRANT_API_KEY || '';
 export const qdrant = new QdrantClient({
   url: qdrantUrl,
   apiKey: qdrantApiKey || undefined,
+  // Disable qdrant client's own version check to avoid the warning spam
+  checkCompatibility: false,
 });
 
 export const COLLECTION_NAME = 'klz_products';
 export const VECTOR_SIZE = 1536; // OpenAI text-embedding-3-small
 
+// Cache TTLs
+const EMBEDDING_CACHE_TTL = 60 * 60 * 24; // 24h — embeddings are deterministic
+const SEARCH_CACHE_TTL = 60 * 30; // 30 min — product data could change
+
+// Track collection existence in-memory (don't re-check every request)
+let collectionVerified = false;
+
 /**
- * Ensure the collection exists in Qdrant.
+ * Ensure the collection exists in Qdrant (only checks once per process lifetime).
  */
 export async function ensureCollection() {
+  if (collectionVerified) return;
+
   try {
     const collections = await qdrant.getCollections();
     const exists = collections.collections.some((c) => c.name === COLLECTION_NAME);
@@ -31,15 +43,47 @@ export async function ensureCollection() {
       });
       console.log(`Successfully created Qdrant collection: ${COLLECTION_NAME}`);
     }
+    collectionVerified = true;
   } catch (error) {
     console.error('Error ensuring Qdrant collection:', error);
   }
 }
 
 /**
- * Generate an embedding for a given text using OpenRouter (OpenAI embedding proxy)
+ * Simple hash for cache keys
+ */
+function hashKey(text: string): string {
+  let hash = 0;
+  for (let i = 0; i < text.length; i++) {
+    const chr = text.charCodeAt(i);
+    hash = (hash << 5) - hash + chr;
+    hash |= 0;
+  }
+  return hash.toString(36);
+}
+
+/**
+ * Generate an embedding for a given text using OpenRouter (OpenAI embedding proxy).
+ * Results are cached in Redis for 24h since embeddings are deterministic.
+ *
+ * NOTE: We keep OpenRouter for embeddings because the Qdrant collection uses 1536-dim
+ * vectors (OpenAI text-embedding-3-small). Switching to Mistral embed (1024-dim) would
+ * require re-indexing the entire product catalog.
+ * User-facing chat uses Mistral AI directly for DSGVO compliance.
  */
 export async function generateEmbedding(text: string): Promise<number[]> {
+  const cacheKey = `emb:${hashKey(text.toLowerCase().trim())}`;
+
+  // Try Redis cache first
+  try {
+    const cached = await redis.get(cacheKey);
+    if (cached) {
+      return JSON.parse(cached);
+    }
+  } catch {
+    // Redis down — proceed without cache
+  }
+
   const openRouterKey = process.env.OPENROUTER_API_KEY;
   if (!openRouterKey) {
     throw new Error('OPENROUTER_API_KEY is not set');
@@ -67,7 +111,16 @@ export async function generateEmbedding(text: string): Promise<number[]> {
   }
 
   const data = await response.json();
-  return data.data[0].embedding;
+  const embedding = data.data[0].embedding;
+
+  // Cache the embedding in Redis
+  try {
+    await redis.set(cacheKey, JSON.stringify(embedding), 'EX', EMBEDDING_CACHE_TTL);
+  } catch {
+    // Redis down — proceed without caching
+  }
+
+  return embedding;
 }
 
 /**
@@ -113,9 +166,23 @@ export async function deleteProductVector(id: string | number) {
 }
 
 /**
- * Search products in Qdrant
+ * Search products in Qdrant.
+ * Results are cached in Redis for 30 minutes keyed by query text.
  */
 export async function searchProducts(query: string, limit = 5) {
+  const cacheKey = `search:${hashKey(query.toLowerCase().trim())}:${limit}`;
+
+  // Try Redis cache first
+  try {
+    const cached = await redis.get(cacheKey);
+    if (cached) {
+      console.log(`[Qdrant] Cache HIT for query: "${query.substring(0, 50)}"`);
+      return JSON.parse(cached);
+    }
+  } catch {
+    // Redis down — proceed without cache
+  }
+
   try {
     await ensureCollection();
     const vector = await generateEmbedding(query);
@@ -126,6 +193,13 @@ export async function searchProducts(query: string, limit = 5) {
       with_payload: true,
     });
 
+    // Cache results in Redis
+    try {
+      await redis.set(cacheKey, JSON.stringify(results), 'EX', SEARCH_CACHE_TTL);
+    } catch {
+      // Redis down — proceed without caching
+    }
+
     return results;
   } catch (error) {
     console.error('Error searching in Qdrant:', error);