create separate models for managed agents

LLM/__init__.py

@@ -0,0 +1,12 @@
+from .llm_models import orchestrator_model, supabase_model, websearch_model
+from .llm_provider import LLMProvider
+from .models import LLMProviderType, LLMModelType
+
+__all__ = [
+    "orchestrator_model",
+    "supabase_model",
+    "websearch_model",
+    "LLMProvider",
+    "LLMProviderType",
+    "LLMModelType",
+]

LLM/llm_models.py

@@ -0,0 +1,9 @@
+
+from .models import LLMProviderType, LLMModelType
+from .llm_provider import LLMProvider
+
+
+orchestrator_model = LLMProvider(LLMProviderType.OPENAI, LLMModelType.openai.gpt_5_1).get_model()
+
+supabase_model = LLMProvider(LLMProviderType.OPENAI, LLMModelType.openai.gpt_4o).get_model()
+websearch_model = LLMProvider(LLMProviderType.OPENAI, LLMModelType.openai.gpt_4o).get_model()

llm_provider.py → LLM/llm_provider.py

@@ -2,52 +2,7 @@ import os
 from typing import Any
 from enum import Enum
 from smolagents import InferenceClientModel, OpenAIServerModel  # type: ignore
-
-class OpenAIModel(str, Enum):
-    gpt_5_1 = "gpt-5.1"
-    gpt_4o = "gpt-4o"
-    gpt_4o_mini = "gpt-4o-mini"
-    gpt_4_1 = "gpt-4.1"
-    gpt_4_1_mini = "gpt-4.1-mini"
-    gpt_4_turbo = "gpt-4-turbo"
-    gpt_3_5_turbo = "gpt-3.5-turbo-0125"
-
-
-class AnthropicModel(str, Enum):
-    claude_3_5_sonnet = "claude-3-5-sonnet-20240620"
-    claude_3_5_haiku = "claude-3-5-haiku-20241022"
-    claude_3_opus = "claude-3-opus-20240229"
-
-
-class GeminiModel(str, Enum):
-    gemini_2_5_flash = "gemini-2.5-flash"
-    gemini_2_5_pro = "gemini-2.5-pro"
-
-
-class OSSModel(str, Enum):
-    llama3_70b_instruct = "meta-llama/Meta-Llama-3-70B-Instruct"
-    llama3_8b_instruct = "meta-llama/Meta-Llama-3-8B-Instruct"
-    mixtral_8x7b = "mistralai/Mixtral-8x7B-Instruct-v0.1"
-    mistral_small_instruct = "mistralai/Mistral-Small-Instruct-2409"
-    qwen2_72b_instruct = "Qwen/Qwen2-72B-Instruct"
-    deepseek_v2 = "deepseek-ai/DeepSeek-V2"
-
-
-class LLMModelType:
-    """Namespaced access to curated model enums, e.g. LLMModelType.openai.gpt_5_1."""
-
-    openai = OpenAIModel
-    claude = AnthropicModel
-    gemini = GeminiModel
-    open_source = OSSModel
-
-class LLMProviderType(str, Enum):
-    OPENAI = "openai"
-    GEMINI = "gemini"
-    CLAUDE = "claude"
-    HF = "hf"
-    OPENROUTER = "openrouter"
-    
+from .models import LLMProviderType
 
 ENV_KEY_MAP = {
     LLMProviderType.OPENAI: "OPENAI_API_KEY",
@@ -106,7 +61,6 @@ class LLMProvider:
             api_base = BASE_URL_MAP.get(self.provider)
             if not api_base:
                 raise ValueError(f"Base URL not configured for provider: {self.provider}")
-            print(api_base, self.api_key, self.model_id)
             return OpenAIServerModel(
                 model_id=self.model_id,
                 api_key=self.api_key,

LLM/models.py

@@ -0,0 +1,47 @@
+from enum import Enum
+
+class OpenAIModel(str, Enum):
+    gpt_5_1 = "gpt-5.1"
+    gpt_4o = "gpt-4o"
+    gpt_4o_mini = "gpt-4o-mini"
+    gpt_4_1 = "gpt-4.1"
+    gpt_4_1_mini = "gpt-4.1-mini"
+    gpt_4_turbo = "gpt-4-turbo"
+    gpt_3_5_turbo = "gpt-3.5-turbo-0125"
+
+
+class AnthropicModel(str, Enum):
+    claude_3_5_sonnet = "claude-3-5-sonnet-20240620"
+    claude_3_5_haiku = "claude-3-5-haiku-20241022"
+    claude_3_opus = "claude-3-opus-20240229"
+
+
+class GeminiModel(str, Enum):
+    gemini_2_5_flash = "gemini-2.5-flash"
+    gemini_2_5_pro = "gemini-2.5-pro"
+
+
+class OSSModel(str, Enum):
+    llama3_70b_instruct = "meta-llama/Meta-Llama-3-70B-Instruct"
+    llama3_8b_instruct = "meta-llama/Meta-Llama-3-8B-Instruct"
+    mixtral_8x7b = "mistralai/Mixtral-8x7B-Instruct-v0.1"
+    mistral_small_instruct = "mistralai/Mistral-Small-Instruct-2409"
+    qwen2_72b_instruct = "Qwen/Qwen2-72B-Instruct"
+    deepseek_v2 = "deepseek-ai/DeepSeek-V2"
+
+
+class LLMModelType:
+    """Namespaced access to curated model enums, e.g. LLMModelType.openai.gpt_5_1."""
+
+    openai = OpenAIModel
+    claude = AnthropicModel
+    gemini = GeminiModel
+    open_source = OSSModel
+
+class LLMProviderType(str, Enum):
+    OPENAI = "openai"
+    GEMINI = "gemini"
+    CLAUDE = "claude"
+    HF = "hf"
+    OPENROUTER = "openrouter"
+    

agents/orchestrator_agent/managed_agents/supabase_agent.py

@@ -1,6 +1,9 @@
-from smolagents import CodeAgent, MCPClient
 import os
 
+from smolagents import CodeAgent, MCPClient
+from LLM.llm_models import supabase_model
+
+
 mcp_client = MCPClient(
         {
             "url": "https://mcp.supabase.com/mcp?project_ref=kzrdiydhjgsaxomkuijq",
@@ -10,15 +13,15 @@ mcp_client = MCPClient(
     , structured_output=True)
 
 
-_supabase_agent = None
 
 
+_supabase_agent = None
 
-def get_supabase_agent(model):
+def get_supabase_agent():
     global _supabase_agent
     if _supabase_agent is None:
         _supabase_agent = CodeAgent(
-            model=model, 
+            model=supabase_model, 
             tools=mcp_client.get_tools(), 
             name="supabase_agent", 
             description="An agent that queries the database for project details, team members, and employee skills."

agents/orchestrator_agent/managed_agents/web_search_agent.py

@@ -1,16 +1,17 @@
-from typing import List, Optional
 
 from smolagents import CodeAgent, WebSearchTool
 
+from LLM.llm_models import websearch_model
+
 
 
 _web_search_agent = None
 
-def get_web_search_agent(model):
+def get_web_search_agent():
     global _web_search_agent
     if _web_search_agent is None:
         _web_search_agent = CodeAgent(
-            model=model, 
+            model=websearch_model, 
             tools=[WebSearchTool()], 
             name="web_search_agent", 
             description="An agent that searches for relevant online courses, certifications, and learning paths."

agents/orchestrator_agent/orchestrator_agent.py

@@ -1,16 +1,12 @@
 from typing import List, Optional
 from smolagents import CodeAgent, tool
 
+from LLM.llm_models import orchestrator_model
 from models import AnalysisResult, EmployeeTrainingPlan
-from llm_provider import LLMModelType, LLMProvider, LLMProviderType
-
 
 from .get_orchestrator_prompt import get_orchestrator_prompt
 from .managed_agents import get_web_search_agent, get_supabase_agent
 
-llm_provider = LLMProviderType.OPENAI
-llm_model = LLMModelType.openai.gpt_5_1
-
 @tool
 def final_answer_callback(project_id: str, project_name: str, team: List[EmployeeTrainingPlan]) -> AnalysisResult:
     """
@@ -47,15 +43,14 @@ class OrchestratorAgent:
         # Initialize the agent as a singleton
         self._initialized = True
 
-        model = LLMProvider(llm_provider, llm_model).get_model()
+       
 
         self._agent = CodeAgent(
-            model=model, 
+            model=orchestrator_model, 
             tools=[final_answer_callback], 
-            managed_agents=[get_web_search_agent(model), get_supabase_agent(model)], 
+            managed_agents=[get_web_search_agent(), get_supabase_agent()], 
             name="orchestrator_agent", 
             description="An agent that orchestrates the training-plan analysis process.",
-            stream_outputs=True
             )
 
     def analyze_and_plan(self, user_prompt: str):