feat: add reasoning_effort parameter support for reasoning models (#4461)

mem0/configs/llms/azure.py

@@ -21,6 +21,7 @@ class AzureOpenAIConfig(BaseLlmConfig):
         top_k: int = 1,
         enable_vision: bool = False,
         vision_details: Optional[str] = "auto",
+        reasoning_effort: Optional[str] = None,
         http_client_proxies: Optional[dict] = None,
         # Azure OpenAI-specific parameters
         azure_kwargs: Optional[Dict[str, Any]] = None,
@@ -37,6 +38,7 @@ class AzureOpenAIConfig(BaseLlmConfig):
             top_k: Top-k sampling parameter, defaults to 1
             enable_vision: Enable vision capabilities, defaults to False
             vision_details: Vision detail level, defaults to "auto"
+            reasoning_effort: Effort level for reasoning models ("low", "medium", "high"), defaults to None
             http_client_proxies: HTTP client proxy settings, defaults to None
             azure_kwargs: Azure-specific configuration, defaults to None
         """
@@ -50,6 +52,7 @@ class AzureOpenAIConfig(BaseLlmConfig):
             top_k=top_k,
             enable_vision=enable_vision,
             vision_details=vision_details,
+            reasoning_effort=reasoning_effort,
             http_client_proxies=http_client_proxies,
         )
 

mem0/configs/llms/base.py

@@ -23,6 +23,7 @@ class BaseLlmConfig(ABC):
         top_k: int = 1,
         enable_vision: bool = False,
         vision_details: Optional[str] = "auto",
+        reasoning_effort: Optional[str] = None,
         http_client_proxies: Optional[Union[Dict, str]] = None,
     ):
         """
@@ -48,6 +49,9 @@ class BaseLlmConfig(ABC):
                 Only applicable to vision-enabled models. Defaults to False
             vision_details: Level of detail for vision processing.
                 Options: "low", "high", "auto". Defaults to "auto"
+            reasoning_effort: Effort level for reasoning models (e.g., o1, o3, gpt-5).
+                Options: "low", "medium", "high". Only applicable to reasoning models.
+                Defaults to None (uses the model's default reasoning effort)
             http_client_proxies: Proxy settings for HTTP client.
                 Can be a dict or string. Defaults to None
         """
@@ -59,4 +63,5 @@ class BaseLlmConfig(ABC):
         self.top_k = top_k
         self.enable_vision = enable_vision
         self.vision_details = vision_details
+        self.reasoning_effort = reasoning_effort
         self.http_client = httpx.Client(proxies=http_client_proxies) if http_client_proxies else None

mem0/configs/llms/openai.py

@@ -20,6 +20,7 @@ class OpenAIConfig(BaseLlmConfig):
         top_k: int = 1,
         enable_vision: bool = False,
         vision_details: Optional[str] = "auto",
+        reasoning_effort: Optional[str] = None,
         http_client_proxies: Optional[dict] = None,
         # OpenAI-specific parameters
         openai_base_url: Optional[str] = None,
@@ -44,6 +45,7 @@ class OpenAIConfig(BaseLlmConfig):
             top_k: Top-k sampling parameter, defaults to 1
             enable_vision: Enable vision capabilities, defaults to False
             vision_details: Vision detail level, defaults to "auto"
+            reasoning_effort: Effort level for reasoning models ("low", "medium", "high"), defaults to None
             http_client_proxies: HTTP client proxy settings, defaults to None
             openai_base_url: OpenAI API base URL, defaults to None
             models: List of models for OpenRouter, defaults to None
@@ -63,6 +65,7 @@ class OpenAIConfig(BaseLlmConfig):
             top_k=top_k,
             enable_vision=enable_vision,
             vision_details=vision_details,
+            reasoning_effort=reasoning_effort,
             http_client_proxies=http_client_proxies,
         )
 

mem0/llms/azure_openai.py

@@ -31,6 +31,7 @@ class AzureOpenAILLM(LLMBase):
                 top_k=config.top_k,
                 enable_vision=config.enable_vision,
                 vision_details=config.vision_details,
+                reasoning_effort=getattr(config, 'reasoning_effort', None),
                 http_client_proxies=config.http_client,
             )
 

mem0/llms/base.py

@@ -88,7 +88,12 @@ class LLMBase(ABC):
                 supported_params["tools"] = kwargs["tools"]
             if "tool_choice" in kwargs:
                 supported_params["tool_choice"] = kwargs["tool_choice"]
-                
+
+            # Add reasoning_effort if configured
+            reasoning_effort = getattr(self.config, 'reasoning_effort', None)
+            if reasoning_effort:
+                supported_params["reasoning_effort"] = reasoning_effort
+
             return supported_params
         else:
             # For regular models, include all common parameters

mem0/llms/openai.py

@@ -29,6 +29,7 @@ class OpenAILLM(LLMBase):
                 top_k=config.top_k,
                 enable_vision=config.enable_vision,
                 vision_details=config.vision_details,
+                reasoning_effort=getattr(config, 'reasoning_effort', None),
                 http_client_proxies=config.http_client,
             )
 

tests/llms/test_azure_openai.py

@@ -91,6 +91,57 @@ def test_generate_response_with_tools(mock_openai_client):
     assert response["tool_calls"][0]["arguments"] == {"data": "Today is a sunny day."}
 
 
+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")
+    llm = AzureOpenAILLM(config)
+    messages = [
+        {"role": "system", "content": "You are a helpful ai."},
+        {"role": "user", "content": "Hello"},
+    ]
+
+    mock_response = Mock()
+    mock_response.choices = [Mock(message=Mock(content="Response from o3-mini"))]
+    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
+    assert call_kwargs[1]["reasoning_effort"] == "low"
+    assert "temperature" not in call_kwargs[1]
+    assert response == "Response from o3-mini"
+
+
+def test_azure_reasoning_effort_not_passed_when_none(mock_openai_client):
+    """Test that reasoning_effort is not passed when not configured on Azure."""
+    config = AzureOpenAIConfig(model="o3-mini")
+    llm = AzureOpenAILLM(config)
+    messages = [
+        {"role": "system", "content": "You are a helpful ai."},
+        {"role": "user", "content": "Hello"},
+    ]
+
+    mock_response = Mock()
+    mock_response.choices = [Mock(message=Mock(content="Response"))]
+    mock_openai_client.chat.completions.create.return_value = mock_response
+
+    llm.generate_response(messages)
+
+    call_kwargs = mock_openai_client.chat.completions.create.call_args
+    assert "reasoning_effort" not in call_kwargs[1]
+
+
+def test_azure_config_accepts_reasoning_effort():
+    """Test that AzureOpenAIConfig accepts reasoning_effort without TypeError (issue #3651)."""
+    config = AzureOpenAIConfig(
+        model="o3-mini",
+        reasoning_effort="low",
+        azure_kwargs={"api_key": "test"},
+    )
+    assert config.reasoning_effort == "low"
+    assert config.model == "o3-mini"
+
+
 @pytest.mark.parametrize(
     "default_headers",
     [None, {"Firstkey": "FirstVal", "SecondKey": "SecondVal"}],

tests/llms/test_openai.py

@@ -170,6 +170,67 @@ def test_callback_exception_handling(mock_openai_client):
     assert llm.config.response_callback is faulty_callback
 
 
+def test_reasoning_model_with_reasoning_effort(mock_openai_client):
+    """Test that reasoning_effort is passed to the API for reasoning models."""
+    config = OpenAIConfig(model="o3-mini", reasoning_effort="low")
+    llm = OpenAILLM(config)
+    messages = [{"role": "user", "content": "Hello"}]
+
+    mock_response = Mock()
+    mock_response.choices = [Mock(message=Mock(content="Response from o3-mini"))]
+    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
+    assert call_kwargs[1]["reasoning_effort"] == "low"
+    assert "temperature" not in call_kwargs[1]  # reasoning models don't get temperature
+    assert response == "Response from o3-mini"
+
+
+def test_reasoning_model_without_reasoning_effort(mock_openai_client):
+    """Test that reasoning_effort is not passed when not configured."""
+    config = OpenAIConfig(model="o3-mini")
+    llm = OpenAILLM(config)
+    messages = [{"role": "user", "content": "Hello"}]
+
+    mock_response = Mock()
+    mock_response.choices = [Mock(message=Mock(content="Response"))]
+    mock_openai_client.chat.completions.create.return_value = mock_response
+
+    llm.generate_response(messages)
+
+    call_kwargs = mock_openai_client.chat.completions.create.call_args
+    assert "reasoning_effort" not in call_kwargs[1]
+
+
+def test_non_reasoning_model_ignores_reasoning_effort(mock_openai_client):
+    """Test that reasoning_effort is not passed for non-reasoning models."""
+    config = OpenAIConfig(model="gpt-4.1-nano-2025-04-14", reasoning_effort="high")
+    llm = OpenAILLM(config)
+    messages = [{"role": "user", "content": "Hello"}]
+
+    mock_response = Mock()
+    mock_response.choices = [Mock(message=Mock(content="Response"))]
+    mock_openai_client.chat.completions.create.return_value = mock_response
+
+    llm.generate_response(messages)
+
+    call_kwargs = mock_openai_client.chat.completions.create.call_args
+    # Non-reasoning models use common params path, reasoning_effort not added there
+    assert "reasoning_effort" not in call_kwargs[1]
+
+
+def test_reasoning_effort_config_values():
+    """Test that reasoning_effort can be set to all valid values."""
+    for effort in ["low", "medium", "high"]:
+        config = OpenAIConfig(model="o3", reasoning_effort=effort)
+        assert config.reasoning_effort == effort
+
+    config = OpenAIConfig(model="o3")
+    assert config.reasoning_effort is None
+
+
 def test_callback_with_tools(mock_openai_client):
     mock_callback = Mock()
     config = OpenAIConfig(model="gpt-4.1-nano-2025-04-14", response_callback=mock_callback)