From 3ffea554bc67d26efc6d96094005e58ac74d3d1f Mon Sep 17 00:00:00 2001
From: Rakhee Singh <rakhisingh2811@gmail.com>
Date: Wed, 8 Apr 2026 19:22:02 +0530
Subject: [PATCH] fix(azure_openai): forward response_format to Azure OpenAI
 API (#4689)

---
 mem0/llms/azure_openai.py       |  2 ++
 tests/llms/test_azure_openai.py | 44 +++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+)

diff --git a/mem0/llms/azure_openai.py b/mem0/llms/azure_openai.py
index 3abd7de49..7a088a17c 100644
--- a/mem0/llms/azure_openai.py
+++ b/mem0/llms/azure_openai.py
@@ -134,6 +134,8 @@ class AzureOpenAILLM(LLMBase):
             "messages": messages,
         })
 
+        if response_format:
+            params["response_format"] = response_format
         if tools:
             params["tools"] = tools
             params["tool_choice"] = tool_choice
diff --git a/tests/llms/test_azure_openai.py b/tests/llms/test_azure_openai.py
index c10dc1635..ee6c01d7c 100644
--- a/tests/llms/test_azure_openai.py
+++ b/tests/llms/test_azure_openai.py
@@ -91,6 +91,50 @@ def test_generate_response_with_tools(mock_openai_client):
     assert response["tool_calls"][0]["arguments"] == {"data": "Today is a sunny day."}
 
 
+def test_generate_response_with_response_format(mock_openai_client):
+    config = AzureOpenAIConfig(model=MODEL, temperature=TEMPERATURE, max_tokens=MAX_TOKENS, top_p=TOP_P)
+    llm = AzureOpenAILLM(config)
+    messages = [
+        {"role": "system", "content": "You are a memory extraction assistant."},
+        {"role": "user", "content": "I like hiking on weekends."},
+    ]
+
+    mock_response = Mock()
+    mock_response.choices = [Mock(message=Mock(content='{"facts": ["User likes hiking on weekends"]}'))]
+    mock_openai_client.chat.completions.create.return_value = mock_response
+
+    response = llm.generate_response(messages, response_format={"type": "json_object"})
+
+    mock_openai_client.chat.completions.create.assert_called_once_with(
+        model=MODEL,
+        messages=messages,
+        temperature=TEMPERATURE,
+        max_tokens=MAX_TOKENS,
+        top_p=TOP_P,
+        response_format={"type": "json_object"},
+    )
+    assert response == '{"facts": ["User likes hiking on weekends"]}'
+
+
+def test_generate_response_without_response_format(mock_openai_client):
+    config = AzureOpenAIConfig(model=MODEL, temperature=TEMPERATURE, max_tokens=MAX_TOKENS, top_p=TOP_P)
+    llm = AzureOpenAILLM(config)
+    messages = [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": "Tell me a joke."},
+    ]
+
+    mock_response = Mock()
+    mock_response.choices = [Mock(message=Mock(content="Why did the chicken cross the road?"))]
+    mock_openai_client.chat.completions.create.return_value = mock_response
+
+    response = llm.generate_response(messages)
+
+    call_kwargs = mock_openai_client.chat.completions.create.call_args[1]
+    assert "response_format" not in call_kwargs
+    assert response == "Why did the chicken cross the road?"
+
+
 def test_reasoning_model_with_reasoning_effort(mock_openai_client):
     """Test that reasoning_effort is passed to the API for Azure reasoning models."""
     config = AzureOpenAIConfig(model="o3-mini", reasoning_effort="low")