feat(oss): port v3 pipeline with hybrid search, entity extraction, and additive scoring (#4805)

Co-authored-by: Soumil Rathi <soumilrathi@gmail.com>
Co-authored-by: Saket Aryan <saketaryan2002@gmail.com>
Co-authored-by: chaithanyak42 <chaithanya.kumar42a@gmail.com>
Co-authored-by: kartik-mem0 <kartik.labhshetwar@mem0.ai>

tests/llms/test_langchain.py

@@ -2,6 +2,8 @@ from unittest.mock import Mock
 
 import pytest
 
+pytest.importorskip("langchain", reason="langchain not installed")
+
 from mem0.configs.llms.base import BaseLlmConfig
 from mem0.llms.langchain import LangchainLLM
 

tests/llms/test_vllm.py

@@ -144,6 +144,7 @@ def create_mocked_memory():
 
         mock_embedder = MagicMock()
         mock_embedder.embed.return_value = [0.1, 0.2, 0.3]
+        mock_embedder.embed_batch.return_value = [[0.1, 0.2, 0.3]]
         mock_embedder_factory.return_value = mock_embedder
 
         mock_vector_store = MagicMock()
@@ -151,9 +152,12 @@ def create_mocked_memory():
         mock_vector_store.add.return_value = None
         mock_vector_factory.return_value = mock_vector_store
 
-        mock_sqlite.return_value = MagicMock()
+        mock_db = MagicMock()
+        mock_db.get_last_messages.return_value = []
+        mock_sqlite.return_value = mock_db
 
         memory = Memory()
+        memory.custom_instructions = None
         memory.api_version = "v1.0"
         return memory, mock_llm, mock_vector_store
 
@@ -170,6 +174,7 @@ def create_mocked_async_memory():
 
         mock_embedder = MagicMock()
         mock_embedder.embed.return_value = [0.1, 0.2, 0.3]
+        mock_embedder.embed_batch.return_value = [[0.1, 0.2, 0.3]]
         mock_embedder_factory.return_value = mock_embedder
 
         mock_vector_store = MagicMock()
@@ -177,9 +182,12 @@ def create_mocked_async_memory():
         mock_vector_store.add.return_value = None
         mock_vector_factory.return_value = mock_vector_store
 
-        mock_sqlite.return_value = MagicMock()
+        mock_db = MagicMock()
+        mock_db.get_last_messages.return_value = []
+        mock_sqlite.return_value = mock_db
 
         memory = AsyncMemory()
+        memory.custom_instructions = None
         memory.api_version = "v1.0"
         return memory, mock_llm, mock_vector_store
 
@@ -187,22 +195,21 @@ def create_mocked_async_memory():
 def test_thinking_tags_sync():
     """Test thinking tags handling in Memory._add_to_vector_store (sync)."""
     memory, mock_llm, mock_vector_store = create_mocked_memory()
-    
-    # Mock LLM responses for both phases
+
+    # v3 pipeline: single LLM call returning ADD-only memories
     mock_llm.generate_response.side_effect = [
-        '        <think>Sync fact extraction</think>  \n{"facts": ["User loves sci-fi"]}',
-        '        <think>Sync memory actions</think>  \n{"memory": [{"text": "Loves sci-fi", "event": "ADD"}]}'
+        '<think>Sync extraction</think>\n{"memory": [{"text": "Loves sci-fi", "attributed_to": "user"}]}'
     ]
-    
+
     mock_vector_store.search.return_value = []
-    
+
     result = memory._add_to_vector_store(
         messages=[{"role": "user", "content": "I love sci-fi movies"}],
-        metadata={}, 
-        filters={}, 
+        metadata={},
+        filters={},
         infer=True
     )
-    
+
     assert len(result) == 1
     assert result[0]["memory"] == "Loves sci-fi"
     assert result[0]["event"] == "ADD"
@@ -213,13 +220,12 @@ def test_thinking_tags_sync():
 async def test_async_thinking_tags_async():
     """Test thinking tags handling in AsyncMemory._add_to_vector_store."""
     memory, mock_llm, mock_vector_store = create_mocked_async_memory()
-    
-    # Directly mock llm.generate_response instead of via asyncio.to_thread
+
+    # v3 pipeline: single LLM call returning ADD-only memories
     mock_llm.generate_response.side_effect = [
-        '        <think>Async fact extraction</think>  \n{"facts": ["User loves sci-fi"]}',
-        '        <think>Async memory actions</think>  \n{"memory": [{"text": "Loves sci-fi", "event": "ADD"}]}'
+        '<think>Async extraction</think>\n{"memory": [{"text": "Loves sci-fi", "attributed_to": "user"}]}'
     ]
-    
+
     # Mock asyncio.to_thread to call the function directly (bypass threading)
     async def mock_to_thread(func, *args, **kwargs):
         if func == mock_llm.generate_response:
@@ -230,15 +236,15 @@ async def test_async_thinking_tags_async():
             return []
         else:
             return func(*args, **kwargs)
-    
+
     with patch('mem0.memory.main.asyncio.to_thread', side_effect=mock_to_thread):
         result = await memory._add_to_vector_store(
             messages=[{"role": "user", "content": "I love sci-fi movies"}],
-            metadata={}, 
-            effective_filters={}, 
+            metadata={},
+            effective_filters={},
             infer=True
         )
-    
+
     assert len(result) == 1
     assert result[0]["memory"] == "Loves sci-fi"
     assert result[0]["event"] == "ADD"