diff --git a/python/flink_agents/e2e_tests/e2e_tests_resource_cross_language/embedding_model_cross_language_test.py b/python/flink_agents/e2e_tests/e2e_tests_resource_cross_language/embedding_model_cross_language_test.py
index a0afc56f2..1e5de5268 100644
--- a/python/flink_agents/e2e_tests/e2e_tests_resource_cross_language/embedding_model_cross_language_test.py
+++ b/python/flink_agents/e2e_tests/e2e_tests_resource_cross_language/embedding_model_cross_language_test.py
@@ -41,7 +41,6 @@
 current_dir = Path(__file__).parent
 
 OLLAMA_MODEL = os.environ.get("OLLAMA_EMBEDDING_MODEL", "nomic-embed-text:latest")
-os.environ["OLLAMA_EMBEDDING_MODEL"] = OLLAMA_MODEL
 
 client = pull_model(OLLAMA_MODEL)
 
@@ -51,7 +50,10 @@
 @pytest.mark.skipif(
     client is None, reason="Ollama client is not available or test model is missing."
 )
-def test_java_embedding_model_integration(tmp_path: Path) -> None:
+def test_java_embedding_model_integration(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    monkeypatch.setenv("OLLAMA_EMBEDDING_MODEL", OLLAMA_MODEL)
     env = StreamExecutionEnvironment.get_execution_environment()
     env.set_runtime_mode(RuntimeExecutionMode.STREAMING)
     env.set_parallelism(1)
diff --git a/python/flink_agents/e2e_tests/e2e_tests_resource_cross_language/vector_store_cross_language_test.py b/python/flink_agents/e2e_tests/e2e_tests_resource_cross_language/vector_store_cross_language_test.py
index ac49eecdf..da987e011 100644
--- a/python/flink_agents/e2e_tests/e2e_tests_resource_cross_language/vector_store_cross_language_test.py
+++ b/python/flink_agents/e2e_tests/e2e_tests_resource_cross_language/vector_store_cross_language_test.py
@@ -43,7 +43,6 @@
 current_dir = Path(__file__).parent
 
 OLLAMA_MODEL = os.environ.get("OLLAMA_EMBEDDING_MODEL", "nomic-embed-text:latest")
-os.environ["OLLAMA_EMBEDDING_MODEL"] = OLLAMA_MODEL
 
 ES_HOST = os.environ.get("ES_HOST")
 MILVUS_URI = os.environ.get("MILVUS_URI")
@@ -61,6 +60,7 @@ def _run_vector_store_integration(
     backend: str,
 ) -> None:
     print(f"[TEST][{backend}] Vector store e2e embedding={embedding_type}")
+    monkeypatch.setenv("OLLAMA_EMBEDDING_MODEL", OLLAMA_MODEL)
     monkeypatch.setenv("EMBEDDING_TYPE", embedding_type)
     monkeypatch.setenv("VECTOR_STORE_BACKEND", backend)
     suffix = uuid.uuid4().hex
diff --git a/python/flink_agents/integrations/embedding_models/local/tests/start_ollama_server.sh b/python/flink_agents/integrations/embedding_models/local/tests/start_ollama_server.sh
index 674ef6e70..92da52d79 100755
--- a/python/flink_agents/integrations/embedding_models/local/tests/start_ollama_server.sh
+++ b/python/flink_agents/integrations/embedding_models/local/tests/start_ollama_server.sh
@@ -36,5 +36,4 @@ if [[ $os == "Linux" ]]; then
   fi
 
   ollama pull all-minilm:22m
-  ollama run all-minilm:22m
 fi
\ No newline at end of file
diff --git a/python/flink_agents/integrations/embedding_models/local/tests/test_ollama_embedding_model.py b/python/flink_agents/integrations/embedding_models/local/tests/test_ollama_embedding_model.py
index ab983cfa9..b770eea3f 100644
--- a/python/flink_agents/integrations/embedding_models/local/tests/test_ollama_embedding_model.py
+++ b/python/flink_agents/integrations/embedding_models/local/tests/test_ollama_embedding_model.py
@@ -78,6 +78,7 @@ def get_resource(name: str, type: ResourceType) -> Resource:
         truncate=True,
         resource_context=mock_ctx,
     )
+    setup.open()
 
     # Test embedding through setup
     embedding = setup.embed("This is a test sentence for embedding.")
diff --git a/python/flink_agents/integrations/embedding_models/tests/test_openai_embedding_model.py b/python/flink_agents/integrations/embedding_models/tests/test_openai_embedding_model.py
index 753820358..e76cc3faa 100644
--- a/python/flink_agents/integrations/embedding_models/tests/test_openai_embedding_model.py
+++ b/python/flink_agents/integrations/embedding_models/tests/test_openai_embedding_model.py
@@ -48,6 +48,7 @@ def get_resource(name: str, type: ResourceType) -> Resource:
     embedding_model = OpenAIEmbeddingModelSetup(
         name="openai", model=test_model, connection="openai", resource_context=mock_ctx
     )
+    embedding_model.open()
 
     response = embedding_model.embed("Hello, Flink Agent!")
     assert response is not None