feat: implémentation de l'orchestrateur asynchrone multi-agents (prompteur/travailleurs/concaténeur) et API de synchronisation des modèles

backend/ai.py

@@ -1,33 +1,29 @@
 from openai import OpenAI
 from fastapi import HTTPException
 from .models import SystemSettings
-from .logger import system_logger
 
 def get_ai_response(messages: list, settings: SystemSettings) -> str:
     """Route la conversation vers le premier fournisseur IA disponible, priorité à Gemini Direct."""
     if not settings:
-        system_logger.error("Configuration système introuvable lors de l'appel IA.")
         raise HTTPException(status_code=500, detail="Configuration système introuvable.")
 
     formatted_messages = [{"role": msg.role, "content": msg.content} for msg in messages]
 
     try:
-        # 1. Priorité absolue : Test Gemini direct (Rapide, sans intermédiaire)
+        # 1. Priorité absolue : Test Gemini direct
         if settings.gemini_api_key:
             client = OpenAI(
                 base_url="https://generativelanguage.googleapis.com/v1beta/openai/", 
                 api_key=settings.gemini_api_key
             )
-            # Utilisation de l'identifiant standard API pour la version Flash
-            model = "gemini-1.5-flash" 
-            system_logger.info(f"Appel IA via Google Gemini direct (Modèle: {model})")
+            # Correction : Utilisation exacte du modèle 3.5 Flash-Lite
+            model = "gemini-3.5-flash-lite" 
             response = client.chat.completions.create(model=model, messages=formatted_messages)
 
         # 2. Test OpenRouter (Fallback)
         elif settings.openrouter_api_key:
             client = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=settings.openrouter_api_key)
-            model = "google/gemini-1.5-pro" # Identifiant mis à jour et valide
-            system_logger.info(f"Appel IA via OpenRouter (Modèle: {model})")
+            model = "google/gemini-3.5-flash-lite"
             response = client.chat.completions.create(
                 model=model,
                 messages=formatted_messages,
@@ -38,14 +34,12 @@ def get_ai_response(messages: list, settings: SystemSettings) -> str:
         elif settings.deepseek_api_key:
             client = OpenAI(base_url="https://api.deepseek.com/v1", api_key=settings.deepseek_api_key)
             model = "deepseek-chat"
-            system_logger.info(f"Appel IA via DeepSeek (Modèle: {model})")
             response = client.chat.completions.create(model=model, messages=formatted_messages)
             
         # 4. Test Groq
         elif settings.groq_api_key:
             client = OpenAI(base_url="https://api.groq.com/openai/v1", api_key=settings.groq_api_key)
             model = "llama3-8b-8192"
-            system_logger.info(f"Appel IA via Groq (Modèle: {model})")
             response = client.chat.completions.create(model=model, messages=formatted_messages)
             
         else:
@@ -54,5 +48,5 @@ def get_ai_response(messages: list, settings: SystemSettings) -> str:
         return response.choices[0].message.content
 
     except Exception as e:
-        system_logger.error(f"Erreur API IA : {str(e)}")
+        print(f"Erreur API IA : {str(e)}")
         raise HTTPException(status_code=502, detail=f"Détail du fournisseur : {str(e)}")

backend/main.py

@@ -10,7 +10,8 @@ from .database import engine, Base, get_db
 from .auth import get_password_hash, generate_totp_secret, get_totp_uri, verify_password, verify_totp, create_access_token, verify_token
 from .schemas import AdminCreate, LoginRequest, ProjectCreate, ProjectResponse, ProjectRename, MessageCreate, MessageResponse
 from .models import User, Project, Message, SystemSettings
-from .ai import get_ai_response
+from .orchestrator import run_orchestrator, sync_providers_models
+from .models import User, Project, Message, SystemSettings, AIModel
 
 # Création des tables dans la base de données
 Base.metadata.create_all(bind=engine)
@@ -151,7 +152,8 @@ def get_messages(project_id: int, db: Session = Depends(get_db), current_user: U
     return db.query(Message).filter(Message.project_id == project_id).order_by(Message.created_at.asc()).all()
 
 @app.post("/api/projects/{project_id}/messages", response_model=List[MessageResponse])
-def create_message(project_id: int, message: MessageCreate, db: Session = Depends(get_db), current_user: User = Depends(get_current_user)):
+async def create_message(project_id: int, message: MessageCreate, db: Session = Depends(get_db), current_user: User = Depends(get_current_user)):
+    """Ajoute un message, lance l'Orchestrateur asynchrone et retourne l'historique."""
     project = db.query(Project).filter(Project.id == project_id, Project.user_id == current_user.id).first()
     if not project: raise HTTPException(status_code=404, detail="Projet introuvable")
     
@@ -162,10 +164,27 @@ def create_message(project_id: int, message: MessageCreate, db: Session = Depend
     history = db.query(Message).filter(Message.project_id == project_id).order_by(Message.created_at.asc()).all()
     settings = db.query(SystemSettings).first()
 
-    ai_response_text = get_ai_response(history, settings)
+    # Configuration temporaire (sera remplacée par les choix du Frontend à la Phase 2)
+    # Ex pour tester le pipeline complet : {"workers": ["gemini-3.5-flash-lite", "google/gemini-1.5-pro"]}
+    orchestrator_config = {"workers": ["gemini-3.5-flash-lite"]} 
+    
+    ai_response_text = await run_orchestrator(history, settings, orchestrator_config)
 
     ai_message = Message(role="assistant", content=ai_response_text, project_id=project_id)
     db.add(ai_message)
     db.commit()
 
-    return db.query(Message).filter(Message.project_id == project_id).order_by(Message.created_at.asc()).all()
+    return db.query(Message).filter(Message.project_id == project_id).order_by(Message.created_at.asc()).all()
+
+@app.post("/api/models/sync")
+async def trigger_model_sync(db: Session = Depends(get_db), current_user: User = Depends(get_current_user)):
+    """Déclenche manuellement l'extraction et la mise à jour des modèles depuis les API."""
+    if not current_user.is_admin:
+        raise HTTPException(status_code=403, detail="Accès réservé aux administrateurs.")
+    settings = db.query(SystemSettings).first()
+    return await sync_providers_models(db, settings)
+
+@app.get("/api/models")
+def get_models(db: Session = Depends(get_db), current_user: User = Depends(get_current_user)):
+    """Renvoie la liste complète des modèles stockés en base de données."""
+    return db.query(AIModel).order_by(AIModel.name.asc()).all()

backend/models.py

@@ -1,60 +1,44 @@
-from sqlalchemy import Column, Integer, String, Boolean, DateTime, ForeignKey, Text
+from sqlalchemy import Column, Integer, String, Boolean, ForeignKey, DateTime, Float
 from sqlalchemy.orm import relationship
 from datetime import datetime, timezone
 from .database import Base
-from sqlalchemy import Float, DateTime
-from datetime import datetime, timezone
 
 class User(Base):
     __tablename__ = "users"
-
     id = Column(Integer, primary_key=True, index=True)
-    email = Column(String, unique=True, index=True, nullable=False)
-    username = Column(String, unique=True, index=True, nullable=False)
-    hashed_password = Column(String, nullable=False)
-    totp_secret = Column(String, nullable=False)
+    email = Column(String, unique=True, index=True)
+    username = Column(String, unique=True, index=True)
+    hashed_password = Column(String)
+    totp_secret = Column(String)
     is_admin = Column(Boolean, default=False)
-    
-    # Relation : Un utilisateur peut avoir plusieurs projets
-    projects = relationship("Project", back_populates="owner", cascade="all, delete-orphan")
-
+    projects = relationship("Project", back_populates="owner")
 
 class Project(Base):
     __tablename__ = "projects"
-
     id = Column(Integer, primary_key=True, index=True)
-    title = Column(String, nullable=False)
+    title = Column(String, index=True)
     created_at = Column(DateTime, default=lambda: datetime.now(timezone.utc))
     is_pinned = Column(Boolean, default=False)
-    user_id = Column(Integer, ForeignKey("users.id"), nullable=False)
-    
-    # Relations
+    user_id = Column(Integer, ForeignKey("users.id"))
     owner = relationship("User", back_populates="projects")
     messages = relationship("Message", back_populates="project", cascade="all, delete-orphan")
 
-
 class Message(Base):
     __tablename__ = "messages"
-
     id = Column(Integer, primary_key=True, index=True)
-    role = Column(String, nullable=False)  # Ex: 'user', 'assistant', 'worker_1', 'system'
-    content = Column(Text, nullable=False)
+    role = Column(String)
+    content = Column(String)
     created_at = Column(DateTime, default=lambda: datetime.now(timezone.utc))
-    project_id = Column(Integer, ForeignKey("projects.id"), nullable=False)
-    
-    # Relation
+    project_id = Column(Integer, ForeignKey("projects.id"))
     project = relationship("Project", back_populates="messages")
 
 class SystemSettings(Base):
     __tablename__ = "system_settings"
-
     id = Column(Integer, primary_key=True, index=True)
-    # Serveur Mail (Obligatoire)
     smtp_host = Column(String, nullable=False)
     smtp_port = Column(Integer, nullable=False)
     smtp_user = Column(String, nullable=False)
     smtp_password = Column(String, nullable=False)
-    # Clés API (Optionnelles)
     openrouter_api_key = Column(String, nullable=True)
     openrouter_management_key = Column(String, nullable=True)
     groq_api_key = Column(String, nullable=True)
@@ -65,25 +49,23 @@ class SystemSettings(Base):
     cloudflare_api_token = Column(String, nullable=True)
     huggingface_api_key = Column(String, nullable=True)
 
-    class AIModel(Base):
+class AIModel(Base):
     """Stocke la liste des modèles extraits depuis les fournisseurs."""
     __tablename__ = "ai_models"
-
     id = Column(Integer, primary_key=True, index=True)
-    provider = Column(String, index=True) # ex: openrouter, groq
-    model_id = Column(String, unique=True, index=True) # ex: google/gemini-1.5-pro
+    provider = Column(String, index=True)
+    model_id = Column(String, unique=True, index=True)
     name = Column(String)
     context_length = Column(Integer)
-    pricing_prompt = Column(Float) # Coût pour 1M tokens
+    pricing_prompt = Column(Float)
     pricing_completion = Column(Float)
     last_updated = Column(DateTime, default=lambda: datetime.now(timezone.utc))
 
 class FinancialLog(Base):
     """Suivi financier par fournisseur."""
     __tablename__ = "financial_logs"
-
     id = Column(Integer, primary_key=True, index=True)
     provider = Column(String, index=True)
-    balance = Column(Float) # Crédit restant
-    total_usage = Column(Float) # Consommation totale
+    balance = Column(Float)
+    total_usage = Column(Float)
     checked_at = Column(DateTime, default=lambda: datetime.now(timezone.utc))

backend/orchestrator.py

@@ -0,0 +1,127 @@
+import asyncio
+import httpx
+from openai import AsyncOpenAI
+from fastapi import HTTPException
+from sqlalchemy.orm import Session
+from .models import SystemSettings, AIModel
+from .logger import system_logger
+from datetime import datetime, timezone
+
+# --- PARTIE 1 : EXTRACTION DES MODÈLES ---
+async def sync_providers_models(db: Session, settings: SystemSettings):
+    """Extrait et met à jour les modèles depuis les fournisseurs configurés."""
+    added_or_updated = 0
+    
+    # httpx.AsyncClient permet des requêtes non-bloquantes (ultra rapide)
+    async with httpx.AsyncClient() as client:
+        # 1. OpenRouter (Exemple principal pour l'extraction massive)
+        if settings.openrouter_api_key:
+            try:
+                response = await client.get("https://openrouter.ai/api/v1/models")
+                if response.status_code == 200:
+                    for item in response.json().get("data", []):
+                        model_id = item["id"]
+                        existing = db.query(AIModel).filter(AIModel.model_id == model_id).first()
+                        
+                        pricing = item.get("pricing", {})
+                        # Conversion en coût pour 1 Million de tokens
+                        p_prompt = float(pricing.get("prompt", 0)) * 1000000 if pricing.get("prompt") else 0.0
+                        p_comp = float(pricing.get("completion", 0)) * 1000000 if pricing.get("completion") else 0.0
+
+                        if existing:
+                            existing.pricing_prompt = p_prompt
+                            existing.pricing_completion = p_comp
+                            existing.last_updated = datetime.now(timezone.utc)
+                        else:
+                            new_model = AIModel(
+                                provider="openrouter",
+                                model_id=model_id,
+                                name=item["name"],
+                                context_length=item.get("context_length", 0),
+                                pricing_prompt=p_prompt,
+                                pricing_completion=p_comp
+                            )
+                            db.add(new_model)
+                        added_or_updated += 1
+            except Exception as e:
+                system_logger.error(f"Erreur Sync OpenRouter: {e}")
+    
+    db.commit()
+    return {"status": "success", "models_processed": added_or_updated}
+
+# --- PARTIE 2 : MOTEUR MULTI-AGENTS ---
+def get_client_for_model(model_id: str, settings: SystemSettings):
+    """Retourne le client AsyncOpenAI approprié selon le modèle sélectionné."""
+    if "gemini" in model_id.lower() and settings.gemini_api_key:
+        return AsyncOpenAI(base_url="https://generativelanguage.googleapis.com/v1beta/openai/", api_key=settings.gemini_api_key)
+    elif settings.openrouter_api_key:
+        return AsyncOpenAI(base_url="https://openrouter.ai/api/v1", api_key=settings.openrouter_api_key)
+    raise ValueError(f"Aucun fournisseur configuré pour {model_id}")
+
+async def ask_agent(client, model_id, messages, is_openrouter=False):
+    """Appel asynchrone à un modèle IA."""
+    kwargs = {"model": model_id, "messages": messages}
+    if is_openrouter:
+        kwargs["extra_headers"] = {"HTTP-Referer": "https://aethas38.duckdns.org", "X-Title": "AETHAS38 Orchestrator"}
+    
+    response = await client.chat.completions.create(**kwargs)
+    return response.choices[0].message.content
+
+async def run_orchestrator(history: list, settings: SystemSettings, config: dict) -> str:
+    """
+    Gère la logique : 1 Prompteur -> N Travailleurs -> 1 Concaténeur.
+    Le dictionnaire 'config' proviendra de l'interface graphique.
+    """
+    workers = config.get("workers", [])
+    if not workers:
+        workers = ["gemini-3.5-flash-lite"] # Fallback de sécurité
+    
+    user_prompt = history[-1].content
+    formatted_history = [{"role": msg.role, "content": msg.content} for msg in history[:-1]]
+
+    # SCÉNARIO 1 : Un seul travailleur (Pas besoin de prompteur/concaténeur)
+    if len(workers) == 1:
+        worker_model = workers[0]
+        client = get_client_for_model(worker_model, settings)
+        messages = formatted_history + [{"role": "user", "content": user_prompt}]
+        return await ask_agent(client, worker_model, messages, "openrouter" in worker_model.lower())
+
+    # SCÉNARIO 2 : Multi-Travailleurs (Le pipeline complet)
+    try:
+        # Étape 1 : Le Prompteur améliore la requête
+        prompter_model = config.get("prompter", "gemini-3.5-flash-lite")
+        p_client = get_client_for_model(prompter_model, settings)
+        p_messages = [{"role": "system", "content": "Tu es un expert en Prompt Engineering. Optimise la requête de l'utilisateur pour qu'elle soit claire, directive et parfaite pour des IAs de génération. Retourne UNIQUEMENT le prompt optimisé."}]
+        p_messages.append({"role": "user", "content": user_prompt})
+        
+        system_logger.info("Démarrage du Prompteur...")
+        optimized_prompt = await ask_agent(p_client, prompter_model, p_messages)
+
+        # Étape 2 : Les Travailleurs en parallèle (Magie de l'Asynchrone)
+        system_logger.info(f"Lancement de {len(workers)} travailleurs en parallèle...")
+        w_tasks = []
+        for w_model in workers:
+            w_client = get_client_for_model(w_model, settings)
+            w_messages = formatted_history + [{"role": "user", "content": optimized_prompt}]
+            # On stocke les tâches sans les attendre immédiatement
+            w_tasks.append(ask_agent(w_client, w_model, w_messages, "openrouter" in w_model.lower()))
+        
+        # 'gather' exécute toutes les requêtes en même temps !
+        workers_responses = await asyncio.gather(*w_tasks, return_exceptions=True)
+
+        # Étape 3 : Le Concaténeur synthétise
+        concat_model = config.get("concatenator", "gemini-3.5-flash-lite")
+        c_client = get_client_for_model(concat_model, settings)
+        
+        synthesis_prompt = f"Voici la requête initiale : {user_prompt}\n\nVoici les réponses de {len(workers)} experts IA différents :\n"
+        for i, resp in enumerate(workers_responses):
+            synthesis_prompt += f"--- EXPERT {i+1} ---\n{resp if not isinstance(resp, Exception) else 'Erreur de génération'}\n\n"
+        synthesis_prompt += "Fais une synthèse finale parfaite, complète et structurée de ces réponses, en gardant le meilleur de chacune."
+
+        system_logger.info("Démarrage du Concaténeur...")
+        c_messages = [{"role": "user", "content": synthesis_prompt}]
+        return await ask_agent(c_client, concat_model, c_messages)
+
+    except Exception as e:
+        system_logger.error(f"Erreur Pipeline Multi-Agents: {e}")
+        raise HTTPException(status_code=502, detail=f"Échec de l'orchestration : {str(e)}")