diff --git a/src/app.py b/src/app.py
index cf09fd5..65c22cc 100644
--- a/src/app.py
+++ b/src/app.py
@@ -1,48 +1,83 @@
 import gradio as gr
-
-from dotenv import load_dotenv
-from app.pipeline import Pipeline
 from agno.utils.log import log_info
+from dotenv import load_dotenv
+
+from app.chat_manager import ChatManager
 
 ########################################
-# MAIN APP & GRADIO INTERFACE
+# MAIN APP & GRADIO CHAT INTERFACE
 ########################################
 if __name__ == "__main__":
-    ######################################
-    # DA FARE PRIMA DI ESEGUIRE L'APP
-    # qui carichiamo le variabili d'ambiente dal file .env
-    # una volta fatto, possiamo usare le API keys senza problemi
-    # quindi non è necessario richiamare load_dotenv() altrove
+    # Carica variabili d’ambiente (.env)
     load_dotenv()
-    ######################################
 
-    pipeline = Pipeline()
+    # Inizializza ChatManager
+    chat = ChatManager()
 
+    ########################################
+    # Funzioni Gradio
+    ########################################
+    def respond(message, history):
+        response = chat.send_message(message)
+        history.append({"role": "user", "content": message})
+        history.append({"role": "assistant", "content": response})
+        return history, history, ""
+
+    def save_current_chat():
+        chat.save_chat("chat.json")
+        return "💾 Chat salvata in chat.json"
+
+    def load_previous_chat():
+        chat.load_chat("chat.json")
+        history = []
+        for m in chat.get_history():
+            history.append({"role": m["role"], "content": m["content"]})
+        return history, history
+
+    def reset_chat():
+        chat.reset_chat()
+        return [], []
+
+    ########################################
+    # Interfaccia Gradio
+    ########################################
     with gr.Blocks() as demo:
-        gr.Markdown("# 🤖 Agente di Analisi e Consulenza Crypto")
+        gr.Markdown("# 🤖 Agente di Analisi e Consulenza Crypto (Chat)")
 
+        # Dropdown provider e stile
         with gr.Row():
             provider = gr.Dropdown(
-                choices=pipeline.list_providers(),
+                choices=chat.list_providers(),
                 type="index",
                 label="Modello da usare"
             )
-            provider.change(fn=pipeline.choose_provider, inputs=provider, outputs=None)
+            provider.change(fn=chat.choose_provider, inputs=provider, outputs=None)
 
             style = gr.Dropdown(
-                choices=pipeline.list_styles(),
+                choices=chat.list_styles(),
                 type="index",
                 label="Stile di investimento"
             )
-            style.change(fn=pipeline.choose_style, inputs=style, outputs=None)
+            style.change(fn=chat.choose_style, inputs=style, outputs=None)
 
-        user_input = gr.Textbox(label="Richiesta utente")
-        output = gr.Textbox(label="Risultato analisi", lines=12)
+        chatbot = gr.Chatbot(label="Conversazione", height=500, type="messages")
+        msg = gr.Textbox(label="Scrivi la tua richiesta", placeholder="Es: Quali sono le crypto interessanti oggi?")
 
-        analyze_btn = gr.Button("🔎 Analizza")
-        analyze_btn.click(fn=pipeline.interact, inputs=[user_input], outputs=output)
+        with gr.Row():
+            clear_btn = gr.Button("🗑️ Reset Chat")
+            save_btn = gr.Button("💾 Salva Chat")
+            load_btn = gr.Button("📂 Carica Chat")
 
-    server, port = ("0.0.0.0", 8000) # 0.0.0.0 per docker compatibility
+        # Invio messaggio
+        msg.submit(respond, inputs=[msg, chatbot], outputs=[chatbot, chatbot, msg])
+        # Reset
+        clear_btn.click(reset_chat, inputs=None, outputs=[chatbot, chatbot])
+        # Salvataggio
+        save_btn.click(save_current_chat, inputs=None, outputs=None)
+        # Caricamento
+        load_btn.click(load_previous_chat, inputs=None, outputs=[chatbot, chatbot])
+
+    server, port = ("0.0.0.0", 8000)
     server_log = "localhost" if server == "0.0.0.0" else server
-    log_info(f"Starting UPO AppAI on http://{server_log}:{port}")
+    log_info(f"Starting UPO AppAI Chat on http://{server_log}:{port}") # noqa
     demo.launch(server_name=server, server_port=port, quiet=True)
diff --git a/src/app/chat_manager.py b/src/app/chat_manager.py
new file mode 100644
index 0000000..7928c95
--- /dev/null
+++ b/src/app/chat_manager.py
@@ -0,0 +1,78 @@
+import os
+import json
+from typing import List, Dict
+from app.pipeline import Pipeline
+
+SAVE_DIR = os.path.join(os.path.dirname(__file__), "..", "saves")
+os.makedirs(SAVE_DIR, exist_ok=True)
+
+class ChatManager:
+    """
+    Gestisce la conversazione con la Pipeline:
+    - mantiene lo storico dei messaggi
+    - invoca la Pipeline per generare risposte
+    - salva e ricarica le chat
+    """
+
+    def __init__(self):
+        self.pipeline = Pipeline()
+        self.history: List[Dict[str, str]] = []  # [{"role": "user"/"assistant", "content": "..."}]
+
+    def send_message(self, message: str) -> str:
+        """
+        Aggiunge un messaggio utente, chiama la Pipeline e salva la risposta nello storico.
+        """
+        # Aggiungi messaggio utente allo storico
+        self.history.append({"role": "user", "content": message})
+
+        # Pipeline elabora la query
+        response = self.pipeline.interact(message)
+
+        # Aggiungi risposta assistente allo storico
+        self.history.append({"role": "assistant", "content": response})
+
+        return response
+
+    def save_chat(self, filename: str = "chat.json") -> None:
+        """
+        Salva la chat corrente in src/saves/<filename>.
+        """
+        path = os.path.join(SAVE_DIR, filename)
+        with open(path, "w", encoding="utf-8") as f:
+            json.dump(self.history, f, ensure_ascii=False, indent=2)
+
+    def load_chat(self, filename: str = "chat.json") -> None:
+        """
+        Carica una chat salvata da src/saves/<filename>.
+        """
+        path = os.path.join(SAVE_DIR, filename)
+        if not os.path.exists(path):
+            self.history = []
+            return
+        with open(path, "r", encoding="utf-8") as f:
+            self.history = json.load(f)
+
+    def reset_chat(self) -> None:
+        """
+        Resetta lo storico della chat.
+        """
+        self.history = []
+
+    def get_history(self) -> List[Dict[str, str]]:
+        """
+        Restituisce lo storico completo della chat.
+        """
+        return self.history
+
+    # Facciamo pass-through di provider e style, così Gradio può usarli
+    def choose_provider(self, index: int):
+        self.pipeline.choose_provider(index)
+
+    def choose_style(self, index: int):
+        self.pipeline.choose_style(index)
+
+    def list_providers(self) -> List[str]:
+        return self.pipeline.list_providers()
+
+    def list_styles(self) -> List[str]:
+        return self.pipeline.list_styles()
diff --git a/src/app/markets/__init__.py b/src/app/markets/__init__.py
index b782b8f..ef73f68 100644
--- a/src/app/markets/__init__.py
+++ b/src/app/markets/__init__.py
@@ -7,7 +7,7 @@ from .binance import BinanceWrapper
 from .cryptocompare import CryptoCompareWrapper
 from .yfinance import YFinanceWrapper
 
-__all__ = [ "MarketAPIs", "BinanceWrapper", "CoinBaseWrapper", "CryptoCompareWrapper", "YFinanceWrapper" ]
+__all__ = [ "MarketAPIsTool", "BinanceWrapper", "CoinBaseWrapper", "CryptoCompareWrapper", "YFinanceWrapper", "MARKET_INSTRUCTIONS" ]
 
 
 class MarketAPIsTool(BaseWrapper, Toolkit):
diff --git a/src/app/pipeline.py b/src/app/pipeline.py
index 10dddab..a7ae9d4 100644
--- a/src/app/pipeline.py
+++ b/src/app/pipeline.py
@@ -1,5 +1,5 @@
+from agno.run.agent import RunOutput
 from agno.team import Team
-from agno.utils.log import log_info
 
 from app.news import NewsAPIsTool, NEWS_INSTRUCTIONS
 from app.social import SocialAPIsTool, SOCIAL_INSTRUCTIONS
@@ -10,98 +10,139 @@ from app.predictor import PredictorStyle, PredictorInput, PredictorOutput, PREDI
 
 class Pipeline:
     """
-    Pipeline coordinata: esegue tutti gli agenti del Team, aggrega i risultati e invoca il Predictor.
+    Coordina gli agenti di servizio (Market, News, Social) e il Predictor finale.
+    Il Team è orchestrato da qwen3:latest (Ollama), mentre il Predictor è dinamico
+    e scelto dall'utente tramite i dropdown dell'interfaccia grafica.
     """
-
     def __init__(self):
         # Inizializza gli agenti
-        market_agent = AppModels.OLLAMA_QWEN_1B.get_agent(
+        self.market_agent = AppModels.OLLAMA_QWEN.get_agent(
             instructions=MARKET_INSTRUCTIONS,
             name="MarketAgent",
             tools=[MarketAPIsTool()]
         )
-        news_agent = AppModels.OLLAMA_QWEN_1B.get_agent(
+        self.news_agent = AppModels.OLLAMA_QWEN.get_agent(
             instructions=NEWS_INSTRUCTIONS,
             name="NewsAgent",
             tools=[NewsAPIsTool()]
         )
-        social_agent = AppModels.OLLAMA_QWEN_1B.get_agent(
+        self.social_agent = AppModels.OLLAMA_QWEN.get_agent(
             instructions=SOCIAL_INSTRUCTIONS,
             name="SocialAgent",
             tools=[SocialAPIsTool()]
         )
 
-        # Crea il Team
-        prompt = """
-        You are the coordinator of a team of analysts specialized in cryptocurrency market analysis.
-        Your role is to gather insights from various sources, including market data, news articles, and social media trends.
-        Based on the information provided by your team members, you will synthesize a comprehensive sentiment analysis for each cryptocurrency discussed.
-        Your analysis should consider the following aspects:
-        1. Market Trends: Evaluate the current market trends and price movements.
-        2. News Impact: Assess the impact of recent news articles on market sentiment.
-        3. Social Media Buzz: Analyze social media discussions and trends related to the cryptocurrencies.
-        Your final output should be a well-rounded sentiment analysis that can guide investment decisions.
-        """ # TODO migliorare il prompt
-        self.team = Team(
-            model = AppModels.OLLAMA_QWEN_1B.get_model(prompt),
-            name="CryptoAnalysisTeam",
-            members=[market_agent, news_agent, social_agent]
+        # === Modello di orchestrazione del Team ===
+        team_model = AppModels.OLLAMA_QWEN.get_model(
+            # TODO: migliorare le istruzioni del team
+            "Agisci come coordinatore: smista le richieste tra MarketAgent, NewsAgent e SocialAgent."
         )
 
-        # Modelli disponibili e Predictor
+        # === Team ===
+        self.team = Team(
+            name="CryptoAnalysisTeam",
+            members=[self.market_agent, self.news_agent, self.social_agent],
+            model=team_model
+        )
+
+        # === Predictor ===
         self.available_models = AppModels.availables()
-        self.predictor_model = self.available_models[0]
-        self.predictor = self.predictor_model.get_agent(PREDICTOR_INSTRUCTIONS, output=PredictorOutput) # type: ignore[arg-type]
+        self.all_styles = list(PredictorStyle)
 
-        # Stili
-        self.styles = list(PredictorStyle)
-        self.style = self.styles[0]
+        # Scelte di default
+        self.chosen_model = self.available_models[0] if self.available_models else None
+        self.style = self.all_styles[0] if self.all_styles else None
 
+        self._init_predictor() # Inizializza il predictor con il modello di default
+
+    # ======================
+    # Dropdown handlers
+    # ======================
     def choose_provider(self, index: int):
-        self.predictor_model = self.available_models[index]
-        self.predictor = self.predictor_model.get_agent(PREDICTOR_INSTRUCTIONS, output=PredictorOutput) # type: ignore[arg-type]
+        """
+        Sceglie il modello LLM da usare per il Predictor.
+        """
+        self.chosen_model = self.available_models[index]
+        self._init_predictor()
 
     def choose_style(self, index: int):
-        self.style = self.styles[index]
-
-    def interact(self, query: str) -> str:
         """
-        Esegue il Team (Market + News + Social), aggrega i risultati e invoca il Predictor.
+        Sceglie lo stile (conservativo/aggressivo) da usare per il Predictor.
         """
-        # Step 1: raccogli output del Team
-        team_results = self.team.run(query)
-        if isinstance(team_results, dict):  # alcuni Team possono restituire dict
-            pieces = [str(v) for v in team_results.values()]
-        elif isinstance(team_results, list):
-            pieces = [str(r) for r in team_results]
-        else:
-            pieces = [str(team_results)]
-        aggregated_text = "\n\n".join(pieces)
+        self.style = self.all_styles[index]
 
-        # Step 2: prepara input per Predictor
-        predictor_input = PredictorInput(
-            data=[],  # TODO: mappare meglio i dati di mercato in ProductInfo
-            style=self.style,
-            sentiment=aggregated_text
+    # ======================
+    # Helpers
+    # ======================
+    def _init_predictor(self):
+        """
+        Inizializza (o reinizializza) il Predictor in base al modello scelto.
+        """
+        if not self.chosen_model:
+            return
+        self.predictor = self.chosen_model.get_agent(
+            PREDICTOR_INSTRUCTIONS,
+            output=PredictorOutput, # type: ignore
+        )
+
+    def list_providers(self) -> list[str]:
+        """
+        Restituisce la lista dei nomi dei modelli disponibili.
+        """
+        return [model.name for model in self.available_models]
+
+    def list_styles(self) -> list[str]:
+        """
+        Restituisce la lista degli stili di previsione disponibili.
+        """
+        return [style.value for style in self.all_styles]
+
+    # ======================
+    # Core interaction
+    # ======================
+    def interact(self, query: str) -> str:
+        """
+        1. Raccoglie output dai membri del Team
+        2. Aggrega output strutturati
+        3. Invoca Predictor
+        4. Restituisce la strategia finale
+        """
+        if not self.predictor or not self.style:
+            return "⚠️ Devi prima selezionare un modello e una strategia validi dagli appositi menu."
+
+        # Step 1: raccolta output dai membri del Team
+        team_outputs = self.team.run(query)
+
+        # Step 2: aggregazione output strutturati
+        all_products = []
+        sentiments = []
+
+        for agent_output in team_outputs.member_responses:
+            if isinstance(agent_output, RunOutput):
+                if "products" in agent_output.metadata:
+                    all_products.extend(agent_output.metadata["products"])
+                if "sentiment_news" in agent_output.metadata:
+                    sentiments.append(agent_output.metadata["sentiment_news"])
+                if "sentiment_social" in agent_output.metadata:
+                    sentiments.append(agent_output.metadata["sentiment_social"])
+
+        aggregated_sentiment = "\n".join(sentiments)
+
+        # Step 3: invocazione Predictor
+        predictor_input = PredictorInput(
+            data=all_products,
+            style=self.style,
+            sentiment=aggregated_sentiment
         )
 
-        # Step 3: chiama Predictor
         result = self.predictor.run(predictor_input)
         prediction: PredictorOutput = result.content
 
-        # Step 4: formatta output finale
+        # Step 4: restituzione strategia finale
         portfolio_lines = "\n".join(
             [f"{item.asset} ({item.percentage}%): {item.motivation}" for item in prediction.portfolio]
         )
-        output = (
+        return (
             f"📊 Strategia ({self.style.value}): {prediction.strategy}\n\n"
             f"💼 Portafoglio consigliato:\n{portfolio_lines}"
         )
-
-        return output
-
-    def list_providers(self) -> list[str]:
-        return [m.name for m in self.available_models]
-
-    def list_styles(self) -> list[str]:
-        return [s.value for s in self.styles]
diff --git a/tests/agents/test_predictor.py b/tests/agents/test_predictor.py
index 9f28717..5867938 100644
--- a/tests/agents/test_predictor.py
+++ b/tests/agents/test_predictor.py
@@ -4,7 +4,7 @@ from app.markets.base import ProductInfo
 from app.models import AppModels
 
 def unified_checks(model: AppModels, input):
-    llm = model.get_agent(PREDICTOR_INSTRUCTIONS, output=PredictorOutput)
+    llm = model.get_agent(PREDICTOR_INSTRUCTIONS, output=PredictorOutput) # type: ignore[arg-type]
     result = llm.run(input)
     content = result.content