feat: add support for system messages to RunnableRails

nemoguardrails/integrations/langchain/runnable_rails.py

@@ -18,7 +18,7 @@
 from typing import Any, List, Optional
 
 from langchain_core.language_models import BaseLanguageModel
-from langchain_core.messages import AIMessage, HumanMessage
+from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
 from langchain_core.prompt_values import ChatPromptValue, StringPromptValue
 from langchain_core.runnables import Runnable
 from langchain_core.runnables.config import RunnableConfig
@@ -139,6 +139,8 @@ def _transform_input_to_rails_format(self, _input):
                         messages.append({"role": "assistant", "content": msg.content})
                     elif isinstance(msg, HumanMessage):
                         messages.append({"role": "user", "content": msg.content})
+                    elif isinstance(msg, SystemMessage):
+                        messages.append({"role": "system", "content": msg.content})
             elif isinstance(_input, StringPromptValue):
                 messages.append({"role": "user", "content": _input.text})
             elif isinstance(_input, dict):

tests/test_runnable_rails.py

@@ -16,7 +16,8 @@
 from typing import List, Optional
 
 import pytest
-from langchain_core.messages import AIMessage
+from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
+from langchain_core.prompt_values import ChatPromptValue
 from langchain_core.prompts import ChatPromptTemplate, PromptTemplate
 from langchain_core.runnables import (
     Runnable,
@@ -153,6 +154,63 @@ def test_dict_messages_in_dict_messages_out():
     assert result["output"] == {"role": "assistant", "content": "Paris."}
 
 
+def test_dict_system_message_in_dict_messages_out():
+    """Tests that SystemMessage is correctly handled."""
+    llm = FakeLLM(
+        responses=[
+            "Okay.",
+        ]
+    )
+    config = RailsConfig.from_content(config={"models": []})
+    model_with_rails = RunnableRails(config, llm=llm)
+
+    original_generate_async = model_with_rails.rails.generate_async
+    messages_passed = None
+
+    async def mock_generate_async(*args, **kwargs):
+        nonlocal messages_passed
+        messages_passed = kwargs.get("messages")
+        return await original_generate_async(*args, **kwargs)
+
+    model_with_rails.rails.generate_async = mock_generate_async
+
+    result = model_with_rails.invoke(
+        input={
+            "input": [
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": "Got it?"},
+            ]
+        }
+    )
+
+    assert isinstance(result, dict)
+    assert result["output"] == {"role": "assistant", "content": "Okay."}
+    assert messages_passed == [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": "Got it?"},
+    ]
+
+
+def test_list_system_message_in_list_messages_out():
+    """Tests that SystemMessage is correctly handled when input is ChatPromptValue."""
+    llm_response = "Intent: user asks question"
+    llm = FakeLLM(responses=[llm_response])
+
+    config = RailsConfig.from_content(config={"models": []})
+    model_with_rails = RunnableRails(config)
+
+    chain = model_with_rails | llm
+
+    input_messages = [
+        SystemMessage(content="You are a helpful assistant."),
+        HumanMessage(content="Got it?"),
+    ]
+    result = chain.invoke(input=ChatPromptValue(messages=input_messages))
+
+    assert isinstance(result, AIMessage)
+    assert result.content == llm_response
+
+
 def test_context_passing():
     llm = FakeLLM(
         responses=[