Add batch embeddings to avoid MCP timeout on store_conversation

store_conversation now collects all texts (summary + key_memories)
and calls Voyage AI once via get_embeddings_batch() instead of N
sequential calls. Prevents MCP timeout for sessions with many memories.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Knowledgebase/server.py

@@ -74,21 +74,37 @@ def get_conn() -> psycopg.Connection:
 
 _voyage_client = None
 
-def get_embedding(text: str) -> Optional[list[float]]:
+def _get_voyage_client():
-    """Return 1024-dim embedding via Voyage AI, or None if unavailable."""
     global _voyage_client
-    if not VOYAGE_API_KEY:
-        return None
-    try:
     if _voyage_client is None:
         import voyageai
         _voyage_client = voyageai.Client(api_key=VOYAGE_API_KEY)
-        result = _voyage_client.embed([text[:8000]], model=EMBED_MODEL)
+    return _voyage_client
+
+def get_embedding(text: str) -> Optional[list[float]]:
+    """Jeden embedding přes Voyage AI, nebo None."""
+    if not VOYAGE_API_KEY:
+        return None
+    try:
+        result = _get_voyage_client().embed([text[:8000]], model=EMBED_MODEL)
         return result.embeddings[0]
     except Exception as e:
         log(f"Embedding error: {e}")
         return None
 
+def get_embeddings_batch(texts: list[str]) -> list[Optional[list[float]]]:
+    """Batch embedding — jedno API volání pro N textů. Mnohem rychlejší než N jednotlivých."""
+    if not VOYAGE_API_KEY or not texts:
+        return [None] * len(texts)
+    try:
+        truncated = [t[:8000] for t in texts]
+        result = _get_voyage_client().embed(truncated, model=EMBED_MODEL)
+        return result.embeddings
+    except Exception as e:
+        log(f"Batch embedding error: {e}")
+        return [None] * len(texts)
+        return None
+
 # ─── Helpers ─────────────────────────────────────────────────────────────────
 
 def _row_to_dict(row: dict) -> dict:
@@ -266,9 +282,22 @@ def store_conversation(
                     (sid, msg.get("role", "unknown"), msg.get("content", ""), i),
                 )
 
+            # ── Batch embedding: summary + všechny key_memories najednou ──
+            valid_kms = [km for km in (key_memories or []) if km.get("content")]
+
+            batch_texts = []
+            if summary:
+                batch_texts.append(f"{title or ''} {summary}")
+            for km in valid_kms:
+                batch_texts.append(f"{km.get('title','') or ''} {km.get('content','')}")
+
+            embeddings = get_embeddings_batch(batch_texts)  # 1 API volání
+
+            emb_iter = iter(embeddings)
+
             # summary jako paměťový záznam
             if summary:
-                emb = get_embedding(f"{title or ''} {summary}")
+                emb = next(emb_iter)
                 _insert_memory_in_tx(conn, {
                     "mem_type": "summary",
                     "title": title or f"Session {sid}",
@@ -282,12 +311,8 @@ def store_conversation(
 
             # key memories
             stored_km = 0
-            for km in (key_memories or []):
+            for km in valid_kms:
-                if not km.get("content"):
+                emb = next(emb_iter)
-                    continue
-                emb = get_embedding(
-                    f"{km.get('title','') or ''} {km.get('content','')}"
-                )
                 _insert_memory_in_tx(conn, {
                     "mem_type":   km.get("mem_type", "fact"),
                     "title":      km.get("title"),