Add system message to KnowledgeEngine and LLMClient

src/ontogpt/clients/llm_client.py

@@ -33,6 +33,9 @@ class LLMClient:
 
     temperature: float = 1.0
 
+    system_message: str = ""
+    """System message to be provided to the LLM."""
+
     def __post_init__(self):
         # Get appropriate API key for the model source
         # and other details if needed
@@ -65,14 +68,19 @@ def complete(self, prompt, show_prompt: bool = False, **kwargs) -> str:
 
         response = None
 
+        these_messages = [{"content": prompt, "role": "user"}]
+
+        if self.system_message:
+            these_messages.insert(0, {"content": self.system_message, "role": "system"})
+
         try:
             # TODO: expose user prompt to CLI
             response = completion(
                 api_key=self.api_key,
                 api_base=self.api_base,
                 api_version=self.api_version,
                 model=self.model,
-                messages=[{"content": prompt, "role": "user"}],
+                messages=these_messages,
                 temperature=self.temperature,
                 caching=True,
                 custom_llm_provider=self.custom_llm_provider,

src/ontogpt/engines/knowledge_engine.py

@@ -144,6 +144,9 @@ class KnowledgeEngine(ABC):
     temperature: float = 1.0
     """Temperature for LLM completions - this is passed to the LLMClient."""
 
+    system_message: str = ""
+    """System message to be provided to the LLM."""
+
     def __post_init__(self):
         if self.template_details:
             (
@@ -167,6 +170,7 @@ def __post_init__(self):
                 api_version=self.api_version,
                 api_base=self.api_base,
                 custom_llm_provider=self.model_provider,
+                system_message=self.system_message,
             )
 
         # We retrieve encoding

src/ontogpt/prompts/qa/__init__.py

@@ -1,4 +1,3 @@
 from pathlib import Path
 
 QA_PROMPT_DIR_PATH = Path(__file__).parent
-GENERIC_QA_PROMPT = QA_PROMPT_DIR_PATH / "generic.jinja2"