Auto-updater, limites CPU 60% et facturation corrigés

modules/model_updater.py

@@ -0,0 +1,55 @@
+import os
+import json
+import requests
+import threading
+import time
+from modules.logger import log_event
+
+CATALOGUE_FILE = "/DATA/AppData/MULTI-IA-AETHAS38/catalogue_dynamique.json"
+
+def _do_update():
+    ollama_host = os.getenv("OLLAMA_HOST", "http://192.168.1.20:11434")
+    try:
+        # MAJ locale Abliterated
+        resp_local = requests.post(
+            f"{ollama_host}/api/pull", 
+            json={"name": "richardyoung/llama-3.2-3b-instruct-abliterated"},
+            timeout=300
+        )
+        if resp_local.status_code == 200:
+            log_event("INFO", "UPDATER", "Modèle local Abliterated vérifié/mis à jour.")
+        
+        # MAJ du catalogue OpenRouter (Gratuits ET Payants)
+        resp_or = requests.get("https://openrouter.ai/api/v1/models", timeout=30)
+        if resp_or.status_code == 200:
+            data = resp_or.json().get("data", [])
+            modeles_gratuits = []
+            modeles_payants = []
+            
+            for m in data:
+                pricing = m.get("pricing", {})
+                # Gratuits purs
+                if pricing.get("prompt") == "0" and pricing.get("completion") == "0":
+                    modeles_gratuits.append(m["id"])
+                # Ciblage des payants ultra-performants (Claude & OpenAI)
+                elif "claude-3.5-sonnet" in m["id"].lower() or "gpt-4o" in m["id"].lower():
+                    modeles_payants.append(m["id"])
+                    
+            with open(CATALOGUE_FILE, "w", encoding="utf-8") as f:
+                json.dump({"free_models": modeles_gratuits, "paid_models": modeles_payants}, f)
+            log_event("INFO", "UPDATER", f"Catalogue dynamique mis à jour ({len(modeles_gratuits)} gratuits, {len(modeles_payants)} payants).")
+            
+    except Exception as e:
+        log_event("ERROR", "UPDATER", f"Erreur lors de la MAJ auto : {e}")
+
+def trigger_update_now():
+    """Déclenchement forcé du scan en cas d'erreur de modèle."""
+    threading.Thread(target=_do_update, daemon=True).start()
+
+def update_models_background():
+    _do_update()
+    while True:
+        time.sleep(7200) # Attente 2 heures
+        _do_update()
+
+threading.Thread(target=update_models_background, daemon=True).start()