Llama-cpp inference, database stored models

backend/src/agents/expert.py

@@ -19,7 +19,7 @@ class ExpertAgent(Agent):
     description = CLASS_DESCRIPTION
     schema_fields = [
         AgentField('system_prompt', 'string', default="You are an expert AI assistant."),
-        AgentField('model', 'string', default="qwen-big"),
+        AgentField('model', 'model_id', default=None),
     ]
 
     def __init__(self, app, data):
@@ -51,7 +51,15 @@ class ExpertAgent(Agent):
 
         model = self.config['model']
 
-        if model.startswith("ollama:"):
+        if type(model) is int:
+            db_model = await self.app.get_model(model)
+            
+            if db_model.value["engine"] == "llama-cpp":
+                response = await inference.local_chat_completion(self.app, messages, model=db_model.value['model'])
+            else:
+                raise Exception("Unknown model: " + repr(model))
+
+        elif model.startswith("ollama:"):
             response = await inference.ollama_chat_completion(self.app, messages, model=model[7:])
         elif model == "qwen-big":
             response = await inference.replicate_qwen_completion(self.app, messages, model=inference.Models.qwen3_235b)