fix(memory): isolate in-memory long-term memory per user

tests/memory/long_term_memory_backends/test_in_memory_ltm_backend.py

@@ -0,0 +1,50 @@
+# Copyright (c) 2025 Beijing Volcano Engine Technology Co., Ltd. and/or its affiliates.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+from llama_index.core.embeddings import MockEmbedding
+
+from veadk.memory.long_term_memory_backends import in_memory_backend
+from veadk.memory.long_term_memory_backends.in_memory_backend import (
+    InMemoryLTMBackend,
+)
+
+
+@pytest.fixture
+def backend(monkeypatch):
+    monkeypatch.setenv("MODEL_EMBEDDING_API_KEY", "test-key")
+    monkeypatch.setattr(
+        in_memory_backend,
+        "create_embedding_model",
+        lambda **_: MockEmbedding(embed_dim=8),
+    )
+    return InMemoryLTMBackend(index="test")
+
+
+def test_search_memory_only_returns_memories_of_the_given_user(backend):
+    backend.save_memory(user_id="alice", event_strings=["alice likes tea"])
+    backend.save_memory(user_id="bob", event_strings=["bob likes coffee"])
+
+    assert backend.search_memory(user_id="alice", query="likes", top_k=5) == [
+        "alice likes tea"
+    ]
+    assert backend.search_memory(user_id="bob", query="likes", top_k=5) == [
+        "bob likes coffee"
+    ]
+
+
+def test_search_memory_for_user_without_memories_is_empty(backend):
+    backend.save_memory(user_id="alice", event_strings=["alice likes tea"])
+
+    assert backend.search_memory(user_id="carol", query="likes", top_k=5) == []

veadk/memory/long_term_memory_backends/in_memory_backend.py

@@ -35,25 +35,37 @@ def model_post_init(self, __context: Any) -> None:
             api_key=self.embedding_config.api_key,
             api_base=self.embedding_config.api_base,
         )
-        self._vector_index = VectorStoreIndex([], embed_model=self._embed_model)
+        # One index per user, so memories never leak across users.
+        self._vector_indexes: dict[str, VectorStoreIndex] = {}
+
+    def _get_vector_index(self, user_id: str) -> VectorStoreIndex:
+        if user_id not in self._vector_indexes:
+            self._vector_indexes[user_id] = VectorStoreIndex(
+                [], embed_model=self._embed_model
+            )
+        return self._vector_indexes[user_id]
 
     def precheck_index_naming(self):
         # no checking
         pass
 
     @override
     def save_memory(self, user_id: str, event_strings: list[str], **kwargs) -> bool:
+        vector_index = self._get_vector_index(user_id)
         for event_string in event_strings:
             document = Document(text=event_string)
             nodes = self._split_documents([document])
-            self._vector_index.insert_nodes(nodes)
+            vector_index.insert_nodes(nodes)
         return True
 
     @override
     def search_memory(
         self, user_id: str, query: str, top_k: int, **kwargs
     ) -> list[str]:
-        _retriever = self._vector_index.as_retriever(similarity_top_k=top_k)
+        vector_index = self._vector_indexes.get(user_id)
+        if vector_index is None:
+            return []
+        _retriever = vector_index.as_retriever(similarity_top_k=top_k)
         retrieved_nodes = _retriever.retrieve(query)
         return [node.text for node in retrieved_nodes]