fix: use knowledge embedding score rerank

src/memos/api/handlers/formatters_handler.py

@@ -34,7 +34,7 @@ def to_iter(running: Any) -> list[Any]:
 
 
 def format_memory_item(
-    memory_data: Any, include_embedding: bool = False, save_sources: bool = False
+    memory_data: Any, include_embedding: bool = False, save_sources: bool = True
 ) -> dict[str, Any]:
     """
     Format a single memory item for API response.
@@ -186,6 +186,13 @@ def rerank_knowledge_mem(
     # rerank set unuse
     reranked_knowledge_mem = knowledge_mem
 
+    # Sort by relativity in descending order
+    reranked_knowledge_mem = sorted(
+        reranked_knowledge_mem,
+        key=lambda item: item.get("metadata", {}).get("relativity", 0.0),
+        reverse=True,
+    )
+
     # TODO revoke sources replace memory value
     for item in reranked_knowledge_mem:
         item["memory"] = item["metadata"]["sources"][0]["content"]

src/memos/api/handlers/search_handler.py

@@ -5,9 +5,12 @@
 using dependency injection for better modularity and testability.
 """
 
+import time
+
 from typing import Any
 
 from memos.api.handlers.base_handler import BaseHandler, HandlerDependencies
+from memos.api.handlers.formatters_handler import rerank_knowledge_mem
 from memos.api.product_models import APISearchRequest, SearchResponse
 from memos.log import get_logger
 from memos.memories.textual.tree_text_memory.retrieve.retrieve_utils import (
@@ -69,6 +72,18 @@ def handle_search_memories(self, search_req: APISearchRequest) -> SearchResponse
             # Restore original top_k for downstream logic or response metadata
             search_req.top_k = original_top_k
 
+        start_time = time.time()
+        text_mem = results["text_mem"]
+        results["text_mem"] = rerank_knowledge_mem(
+            self.reranker,
+            query=search_req.query,
+            text_mem=text_mem,
+            top_k=original_top_k,
+            file_mem_proportion=0.5,
+        )
+        rerank_time = time.time() - start_time
+
+        self.logger.info(f"[Knowledge_replace_memory_time] Rerank time: {rerank_time} seconds")
         self.logger.info(
             f"[SearchHandler] Final search results: count={len(results)} results={results}"
         )