Protocol 22: Universal Model Gating & Network Layer Fusion

logos/agent_dispatcher.py

@@ -348,8 +348,8 @@ class LogosSwarm:
     Protocol 21: The Async Interference Bus.
     Agents pulse in parallel; the solution is the geometric intersection.
     """
-    def __init__(self, base_url="http://localhost:1234/v1"):
-        self.connector = get_connector('local', base_url=base_url)
+    def __init__(self, base_url="http://localhost:1234/v1", model="google/gemma-3-4b"):
+        self.connector = get_connector('local', base_url=base_url, model=model)
         self.state = {
             "last_node": 1,
             "tensor_history": [],
@@ -490,7 +490,7 @@ class NeuralRouter:
     Implements the Routing Logic to dispatch tasks to the Nano Swarm.
     """
     
-    def __init__(self, base_url: str = "http://192.168.0.105:1234/v1", router_model: str = "google/gemma-3-4b"):
+    def __init__(self, base_url: str = "http://localhost:1234/v1", router_model: str = "google/gemma-3-4b"):
         # The router itself uses a fast model (Gemma 4B)
         self.connector = get_connector('local', base_url=base_url, model=router_model)
         self.history = []

logos/connectors.py

@@ -99,7 +99,6 @@ class OCRConnector:
     
     def __init__(self, languages: List[str] = None, gpu: bool = False):
         # We rely on the local LLM connector, 'gpu' arg is ignored as it's handled by LM Studio
-        from .connectors import get_connector
         # Hardcoded to Gemma as requested by user ("gemma is your vision model")
         self.client = get_connector('local', model="google/gemma-3-4b")
     
@@ -286,7 +285,7 @@ class LocalLLMConnector:
     Optimization: Direct localhost access (no Docker bridge lag).
     """
     
-    def __init__(self, base_url: str = None, model: str = "local-model"):
+    def __init__(self, base_url: str = None, model: str = "google/gemma-3-4b"):
         # Prioritize Environment -> Argument -> Default
         env_url = os.environ.get("LOGOS_LLM_ENDPOINT")
         self.base_url = base_url or env_url or "http://localhost:1234/v1"
@@ -322,7 +321,7 @@ class LocalLLMConnector:
                         logprobs = data['choices'][0].get('logprobs')
                         return content, logprobs
                     else:
-                        return f"[Error] Local LLM returned status {response.status}"
+                        return f"[Error] Local LLM returned status {response.status}", None
         except Exception as e:
             return f"[Async Local LLM Error] {e}", None
 

logos/server.py

@@ -62,7 +62,7 @@ logger = logging.getLogger("LOGOS_Router")
 # Protocol 5: Tiered Token Consumption
 # Protocol 10: Swarm Optimization (Unified Model Mode)
 FORCE_SINGLE_MODEL = True  # Set to True to prevent VRAM trashing by model swapping
-UNIFIED_MODEL_ID = "local-model" # Points to whatever is currently loaded in LM Studio
+UNIFIED_MODEL_ID = "google/gemma-3-4b" # Points to the active model in LM Studio
 
 SHELL_CONFIG = {
     "INNER_SHELL": {