fix: apply RAG_EMBEDDING_QUERY_PREFIX to memory search queries (#24921)

The query_memory endpoint embeds the search query without the configured
RAG_EMBEDDING_QUERY_PREFIX, while every RAG retrieval path in
retrieval/utils.py correctly passes it. Instruction-tuned embedding
models (e.g. Qwen3-Embedding) produce poor results without the prefix,
causing memory search to return semantically unrelated results.

backend/open_webui/routers/memories.py

@@ -9,6 +9,7 @@ from open_webui.constants import ERROR_MESSAGES
 from open_webui.internal.db import get_async_session
 from open_webui.models.memories import Memories, MemoryModel
 from open_webui.retrieval.vector.async_client import ASYNC_VECTOR_DB_CLIENT
+from open_webui.config import RAG_EMBEDDING_QUERY_PREFIX
 from open_webui.utils.access_control import has_permission
 from open_webui.utils.auth import get_verified_user
 from pydantic import BaseModel
@@ -137,7 +138,7 @@ async def query_memory(
     if not memories:
         raise HTTPException(status_code=404, detail='No memories found for user')
 
-    vector = await request.app.state.EMBEDDING_FUNCTION(form_data.content, user=user)
+    vector = await request.app.state.EMBEDDING_FUNCTION(form_data.content, RAG_EMBEDDING_QUERY_PREFIX, user=user)
 
     results = await ASYNC_VECTOR_DB_CLIENT.search(
         collection_name=f'user-memory-{user.id}',