Spaces:

elmarcito
/

nvidia

Sleeping

App Files Files Community

elmarcito commited on Jul 4

Commit

3cbc846

verified ·

1 Parent(s): 2be6704

Update app.py

Browse files

Files changed (1) hide show

app.py +89 -273

app.py CHANGED Viewed

@@ -1,9 +1,11 @@
 """
 🚀 NVIDIA AI Multi-Model Space
-- Tutti i modelli NVIDIA NIM
-- Memoria breve e lunga
-- API REST GET/POST + SSE streaming
 - Server MCP integrato
 """
 import os
@@ -12,15 +14,18 @@ import time
 import uuid
 import base64
 import asyncio
-import hashlib
 from typing import Optional, AsyncGenerator
 import gradio as gr
 import httpx
 from dotenv import load_dotenv
 from fastapi import FastAPI, Request, Query, HTTPException
-from fastapi.responses import StreamingResponse, JSONResponse, HTMLResponse
 from fastapi.middleware.cors import CORSMiddleware
 from sse_starlette.sse import EventSourceResponse
 from pydantic import BaseModel, Field
@@ -389,6 +394,11 @@ app.add_middleware(
     allow_headers=["*"],
 )
 # ── Pydantic Models ───────────────────────────────────────────
@@ -433,15 +443,42 @@ class MemoryRequest(BaseModel):
     summary: Optional[str] = None
-# ── API Endpoints ──────────────────────────────────────────────
-@app.get("/")
 async def root():
     return {
         "service": "🚀 NVIDIA AI Multi-Model Space",
         "version": "2.0.0",
         "models_available": len(NVIDIA_MODELS),
         "endpoints": {
             "docs": "/docs",
             "api_docs_html": "/api-docs",
             "models": "/v1/models",
@@ -466,6 +503,10 @@ async def health():
     }
 @app.get("/v1/models")
 async def list_models(category: Optional[str] = None):
     """Lista tutti i modelli disponibili, filtrabile per categoria"""
@@ -490,7 +531,9 @@ async def get_model_info(model_id: str):
     return {"model_id": model_id, **NVIDIA_MODELS[model_id]}
-# ── Chat POST (non-streaming) ─────────────────────────────────
 @app.post("/v1/chat")
 async def chat_post(req: ChatRequest):
@@ -503,7 +546,6 @@ async def chat_post(req: ChatRequest):
     if req.stream:
         return await _stream_response(req, session_id)
-    # Costruisci contesto con memoria
     if req.use_memory:
         memory.add_short(session_id, "user", req.message)
         messages = memory.get_full_context(session_id, req.system_prompt)
@@ -536,8 +578,6 @@ async def chat_post(req: ChatRequest):
         raise HTTPException(500, detail=str(e))
-# ── Chat GET ───────────────────────────────────────────────────
 @app.get("/v1/chat/get")
 async def chat_get(
     message: str = Query(..., description="Messaggio utente"),
@@ -569,8 +609,6 @@ async def chat_get(
     return await chat_post(req)
-# ── Chat SSE Stream ───────────────────────────────────────────
 @app.post("/v1/chat/stream")
 async def chat_stream_post(req: ChatRequest):
     """Endpoint dedicato per streaming SSE via POST"""
@@ -619,7 +657,6 @@ async def _stream_response(req: ChatRequest, session_id: str):
     async def event_generator():
         full_response = []
         try:
-            # Evento di inizio
             yield {
                 "event": "start",
                 "data": json.dumps({
@@ -642,12 +679,10 @@ async def _stream_response(req: ChatRequest, session_id: str):
                     "data": json.dumps({"token": chunk})
                 }
-            # Salva in memoria
             complete = "".join(full_response)
             if req.use_memory:
                 memory.add_short(session_id, "assistant", complete)
-            # Evento di fine
             yield {
                 "event": "done",
                 "data": json.dumps({
@@ -667,7 +702,9 @@ async def _stream_response(req: ChatRequest, session_id: str):
     return EventSourceResponse(event_generator())
-# ── Memory Management ─────────────────────────────────────────
 @app.post("/v1/memory")
 async def manage_memory(req: MemoryRequest):
@@ -830,7 +867,6 @@ async def mcp_endpoint(request: Request):
             "error": {"code": code, "message": message}
         })
-    # ── Initialize ──
     if method == "initialize":
         return _mcp_response({
             "protocolVersion": "2024-11-05",
@@ -845,11 +881,9 @@ async def mcp_endpoint(request: Request):
             }
         })
-    # ── Tools List ──
     elif method == "tools/list":
         return _mcp_response({"tools": MCP_TOOLS})
-    # ── Tools Call ──
     elif method == "tools/call":
         tool_name = params.get("name", "")
         tool_args = params.get("arguments", {})
@@ -862,7 +896,6 @@ async def mcp_endpoint(request: Request):
         except Exception as e:
             return _mcp_error(-32000, str(e))
-    # ── Resources List ──
     elif method == "resources/list":
         return _mcp_response({
             "resources": [
@@ -881,7 +914,6 @@ async def mcp_endpoint(request: Request):
             ]
         })
-    # ── Resources Read ──
     elif method == "resources/read":
         uri = params.get("uri", "")
         if uri == "nvidia://models":
@@ -1010,19 +1042,7 @@ async def api_docs_page():
   .param-table { width: 100%; border-collapse: collapse; margin: 10px 0; }
   .param-table th, .param-table td { padding: 10px; border: 1px solid var(--border); text-align: left; }
   .param-table th { background: var(--code-bg); color: var(--accent); }
-  .badge { display: inline-block; padding: 2px 8px; border-radius: 4px; font-size: 0.8em; margin: 2px; }
-  .badge-llm { background: #1a5276; color: #85c1e9; }
-  .badge-code { background: #1e4620; color: #82e0aa; }
-  .badge-vision { background: #4a235a; color: #d7bde2; }
-  .badge-reasoning { background: #7d3c00; color: #f0b27a; }
-  .tab-container { margin: 20px 0; }
-  .tabs { display: flex; gap: 5px; margin-bottom: -1px; }
-  .tab { padding: 10px 20px; background: var(--code-bg); border: 1px solid var(--border); border-bottom: none; border-radius: 8px 8px 0 0; cursor: pointer; color: #888; }
-  .tab.active { background: var(--card); color: var(--accent); border-color: var(--accent); }
-  .tab-content { display: none; background: var(--card); border: 1px solid var(--border); border-radius: 0 8px 8px 8px; padding: 20px; }
-  .tab-content.active { display: block; }
   .note { background: #1a3a1a; border-left: 4px solid var(--accent); padding: 15px; margin: 15px 0; border-radius: 0 8px 8px 0; }
-  .warning { background: #3a2a1a; border-left: 4px solid #f0b27a; padding: 15px; margin: 15px 0; border-radius: 0 8px 8px 0; }
   a { color: var(--accent); }
 </style>
 </head>
@@ -1034,163 +1054,56 @@ async def api_docs_page():
 <div class="note">
   <strong>Base URL:</strong> <code>https://YOUR-SPACE.hf.space</code><br>
-  <strong>Autenticazione:</strong> API key NVIDIA pre-configurata (default) oppure fornire la propria.
 </div>
-<h2>📋 Endpoints Rapidi</h2>
 <div class="endpoint">
 <table class="param-table">
 <tr><th>Metodo</th><th>Endpoint</th><th>Descrizione</th></tr>
-<tr><td><span class="method get">GET</span></td><td><code>/v1/models</code></td><td>Lista tutti i modelli</td></tr>
 <tr><td><span class="method get">GET</span></td><td><code>/v1/chat/get</code></td><td>Chat via GET</td></tr>
 <tr><td><span class="method post">POST</span></td><td><code>/v1/chat</code></td><td>Chat via POST</td></tr>
-<tr><td><span class="method get">GET</span></td><td><code>/v1/chat/stream</code></td><td>Chat streaming SSE via GET</td></tr>
-<tr><td><span class="method post">POST</span></td><td><code>/v1/chat/stream</code></td><td>Chat streaming SSE via POST</td></tr>
 <tr><td><span class="method post">POST</span></td><td><code>/v1/memory</code></td><td>Gestione memoria</td></tr>
-<tr><td><span class="method get">GET</span></td><td><code>/v1/memory/{session_id}</code></td><td>Stats memoria sessione</td></tr>
 <tr><td><span class="method post">POST</span></td><td><code>/v1/mcp</code></td><td>Server MCP (JSON-RPC)</td></tr>
 </table>
 </div>
-<h2>💬 Chat API</h2>
-<h3>POST /v1/chat</h3>
 <div class="endpoint">
-<span class="method post">POST</span> <span class="url">/v1/chat</span>
-<p>Chat completo con supporto memoria e streaming opzionale.</p>
-<pre><code># cURL - Chat semplice
-curl -X POST https://YOUR-SPACE.hf.space/v1/chat \\
   -H "Content-Type: application/json" \\
   -d '{
     "message": "Spiegami la relativit&agrave;",
     "model": "meta/llama-3.1-405b-instruct",
     "session_id": "user-123",
-    "temperature": 0.7,
     "use_memory": true
   }'</code></pre>
-<pre><code># Python
-import requests
-resp = requests.post("https://YOUR-SPACE.hf.space/v1/chat", json={
-    "message": "Ciao! Come funziona la memoria?",
-    "model": "meta/llama-3.1-405b-instruct",
-    "session_id": "my-session",
-    "use_memory": True
-})
-print(resp.json()["response"])</code></pre>
-<pre><code># JavaScript (fetch)
-const resp = await fetch('https://YOUR-SPACE.hf.space/v1/chat', {
-  method: 'POST',
-  headers: {'Content-Type': 'application/json'},
-  body: JSON.stringify({
-    message: 'Hello!',
-    model: 'meta/llama-3.1-405b-instruct',
-    session_id: 'js-session'
-  })
-});
-const data = await resp.json();
-console.log(data.response);</code></pre>
 </div>
-<h3>GET /v1/chat/get</h3>
 <div class="endpoint">
-<span class="method get">GET</span> <span class="url">/v1/chat/get?message=...&amp;model=...</span>
-<p>Chat via query parameters — comodo per test rapidi.</p>
-<pre><code># Browser o cURL
-curl "https://YOUR-SPACE.hf.space/v1/chat/get?message=Ciao&model=meta/llama-3.1-8b-instruct&session_id=test"</code></pre>
-</div>
-<h2>📡 Streaming SSE</h2>
-<div class="endpoint">
-<span class="method get">GET</span> <span class="url">/v1/chat/stream?message=...</span><br>
-<span class="method post">POST</span> <span class="url">/v1/chat/stream</span>
-<p>Risposta in tempo reale via Server-Sent Events.</p>
-<p><strong>Eventi SSE:</strong></p>
-<ul>
-  <li><code>event: start</code> — Inizio generazione (contiene session_id, model)</li>
-  <li><code>event: token</code> — Singolo token generato</li>
-  <li><code>event: done</code> — Fine (contiene risposta completa + stats memoria)</li>
-  <li><code>event: error</code> — Errore</li>
-</ul>
-<pre><code># JavaScript - EventSource (GET)
 const es = new EventSource(
-  'https://YOUR-SPACE.hf.space/v1/chat/stream?message=Raccontami+una+storia&session_id=sse-demo'
 );
-es.addEventListener('token', (e) => {
-  const {token} = JSON.parse(e.data);
-  process.stdout.write(token);  // o document.body.innerHTML += token;
-});
-es.addEventListener('done', (e) => {
-  const {full_response, memory_stats} = JSON.parse(e.data);
-  console.log('\\n--- DONE ---');
-  console.log('Stats:', memory_stats);
-  es.close();
 });
-es.addEventListener('error', (e) => {
-  console.error('Error:', e.data);
-  es.close();
-});</code></pre>
-<pre><code># Python - SSE streaming con httpx
-import httpx
-import json
-with httpx.stream('GET', 'https://YOUR-SPACE.hf.space/v1/chat/stream',
-                   params={'message': 'Ciao!', 'session_id': 'py-sse'}) as r:
-    for line in r.iter_lines():
-        if line.startswith('data: '):
-            data = json.loads(line[6:])
-            if 'token' in data:
-                print(data['token'], end='', flush=True)</code></pre>
-<pre><code># cURL - SSE streaming
-curl -N "https://YOUR-SPACE.hf.space/v1/chat/stream?message=Ciao&session_id=curl-test"</code></pre>
-</div>
-<h2>🧠 Memoria API</h2>
-<div class="endpoint">
-<p>Il sistema di memoria gestisce automaticamente:</p>
-<ul>
-  <li><strong>Memoria Breve:</strong> ultimi 20 messaggi della sessione attiva (in RAM)</li>
-  <li><strong>Memoria Lunga:</strong> messaggi precedenti salvati in SQLite con limite token</li>
-  <li><strong>Riassunti:</strong> possibilit&agrave; di salvare riassunti di sessione</li>
-</ul>
-<pre><code># Statistiche memoria
-curl https://YOUR-SPACE.hf.space/v1/memory/my-session
-# Cronologia completa
-curl https://YOUR-SPACE.hf.space/v1/memory/my-session/history
-# Pulisci memoria breve
-curl -X POST https://YOUR-SPACE.hf.space/v1/memory \\
-  -H "Content-Type: application/json" \\
-  -d '{"session_id": "my-session", "action": "clear_short"}'
-# Elimina tutta la sessione
-curl -X DELETE https://YOUR-SPACE.hf.space/v1/sessions/my-session
-# Salva riassunto
-curl -X POST https://YOUR-SPACE.hf.space/v1/memory \\
-  -H "Content-Type: application/json" \\
-  -d '{"session_id": "my-session", "action": "save_summary", "summary": "L utente sta imparando Python"}'</code></pre>
 </div>
-<h2>🔌 Server MCP</h2>
 <div class="endpoint">
-<span class="method post">POST</span> <span class="url">/v1/mcp</span>
-<p>Endpoint MCP (Model Context Protocol) per integrazioni con Claude Desktop, Cursor, etc.</p>
-<pre><code># Configurazione MCP per Claude Desktop / Cursor
-// In mcp_settings.json:
 {
   "mcpServers": {
     "nvidia-ai": {
@@ -1199,78 +1112,6 @@ curl -X POST https://YOUR-SPACE.hf.space/v1/memory \\
     }
   }
 }</code></pre>
-<pre><code># Test MCP - Initialize
-curl -X POST https://YOUR-SPACE.hf.space/v1/mcp \\
-  -H "Content-Type: application/json" \\
-  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}'
-# Test MCP - List Tools
-curl -X POST https://YOUR-SPACE.hf.space/v1/mcp \\
-  -H "Content-Type: application/json" \\
-  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}'
-# Test MCP - Call Tool
-curl -X POST https://YOUR-SPACE.hf.space/v1/mcp \\
-  -H "Content-Type: application/json" \\
-  -d '{
-    "jsonrpc":"2.0","id":3,
-    "method":"tools/call",
-    "params":{
-      "name":"nvidia_chat",
-      "arguments":{
-        "message":"What is quantum computing?",
-        "model":"meta/llama-3.1-70b-instruct",
-        "session_id":"mcp-test"
-      }
-    }
-  }'</code></pre>
-<h3>Tools MCP Disponibili:</h3>
-<table class="param-table">
-<tr><th>Tool</th><th>Descrizione</th></tr>
-<tr><td><code>nvidia_chat</code></td><td>Chat con qualsiasi modello NVIDIA + memoria</td></tr>
-<tr><td><code>nvidia_chat_stream</code></td><td>Info per streaming SSE</td></tr>
-<tr><td><code>list_models</code></td><td>Lista modelli (filtrabile per categoria)</td></tr>
-<tr><td><code>memory_stats</code></td><td>Statistiche memoria sessione</td></tr>
-<tr><td><code>memory_clear</code></td><td>Pulisci memoria (short/all)</td></tr>
-<tr><td><code>get_conversation_history</code></td><td>Cronologia completa conversazione</td></tr>
-</table>
-</div>
-<h2>🤖 Modelli Disponibili</h2>
-<div class="endpoint">
-<p>
-<span class="badge badge-llm">LLM</span>
-<span class="badge badge-code">Code</span>
-<span class="badge badge-vision">Vision</span>
-<span class="badge badge-reasoning">Reasoning</span>
-e altri...
-</p>
-<pre><code># Lista tutti i modelli
-curl https://YOUR-SPACE.hf.space/v1/models
-# Filtra per categoria
-curl "https://YOUR-SPACE.hf.space/v1/models?category=Code"
-# Info singolo modello
-curl https://YOUR-SPACE.hf.space/v1/models/meta/llama-3.1-405b-instruct</code></pre>
-</div>
-<h2>📐 Schema Parametri Chat</h2>
-<div class="endpoint">
-<table class="param-table">
-<tr><th>Parametro</th><th>Tipo</th><th>Default</th><th>Descrizione</th></tr>
-<tr><td><code>message</code></td><td>string</td><td><em>richiesto</em></td><td>Messaggio dell'utente</td></tr>
-<tr><td><code>model</code></td><td>string</td><td>meta/llama-3.1-405b-instruct</td><td>ID modello NVIDIA</td></tr>
-<tr><td><code>session_id</code></td><td>string</td><td>auto-generated</td><td>ID sessione per memoria</td></tr>
-<tr><td><code>system_prompt</code></td><td>string</td><td>"Sei un assistente AI..."</td><td>System prompt</td></tr>
-<tr><td><code>temperature</code></td><td>float</td><td>0.7</td><td>Creativit&agrave; (0.0 - 2.0)</td></tr>
-<tr><td><code>max_tokens</code></td><td>int</td><td>4096</td><td>Max token risposta</td></tr>
-<tr><td><code>top_p</code></td><td>float</td><td>0.9</td><td>Nucleus sampling</td></tr>
-<tr><td><code>use_memory</code></td><td>bool</td><td>true</td><td>Usa sistema memoria</td></tr>
-<tr><td><code>stream</code></td><td>bool</td><td>false</td><td>Abilita SSE streaming</td></tr>
-</table>
 </div>
 </div>
@@ -1280,7 +1121,7 @@ curl https://YOUR-SPACE.hf.space/v1/models/meta/llama-3.1-405b-instruct</code></
 # ══════════════════════════════════════════════════════════════
-# GRADIO UI
 # ══════════════════════════════════════════════════════════════
 def get_model_choices():
@@ -1315,10 +1156,11 @@ async def gradio_chat(
     else:
         messages = [{"role": "system", "content": system_prompt}]
         for h in history:
-            if h["role"] == "user":
-                messages.append({"role": "user", "content": h["content"]})
-            elif h["role"] == "assistant":
-                messages.append({"role": "assistant", "content": h["content"]})
         messages.append({"role": "user", "content": message})
     full_response = ""
@@ -1341,9 +1183,8 @@ async def gradio_chat(
 def get_memory_info(session_id: str):
     if not session_id:
-        return "⚠️ Inserisci un Session ID"
-    stats = memory.get_stats(session_id)
-    return json.dumps(stats, indent=2, ensure_ascii=False)
 def clear_session_memory(session_id: str):
@@ -1353,8 +1194,6 @@ def clear_session_memory(session_id: str):
     return f"🗑️ Memoria sessione '{session_id}' eliminata"
-# ── Build Gradio Interface ────────────────────────────────────
 with gr.Blocks(
     title="🚀 NVIDIA AI Multi-Model Space",
     theme=gr.themes.Soft(
@@ -1372,7 +1211,7 @@ with gr.Blocks(
     # 🚀 NVIDIA AI Multi-Model Space
     **Chat con tutti i modelli NVIDIA NIM** • Memoria Breve/Lunga • API REST • SSE Streaming • MCP Server
-    📖 [Documentazione API](/api-docs) &nbsp;|&nbsp; 🔧 [OpenAPI Swagger](/docs) &nbsp;|&nbsp; 🔌 [MCP Endpoint](/v1/mcp)
     """)
     with gr.Tab("💬 Chat"):
@@ -1382,7 +1221,6 @@ with gr.Blocks(
                     height=600,
                     type="messages",
                     show_copy_button=True,
-                    avatar_images=(None, "https://upload.wikimedia.org/wikipedia/sco/thumb/2/21/Nvidia_logo.svg/200px-Nvidia_logo.svg.png"),
                     placeholder="Seleziona un modello e inizia a chattare..."
                 )
                 msg = gr.Textbox(
@@ -1425,7 +1263,6 @@ with gr.Blocks(
                 mem_output = gr.JSON(label="📊 Memoria", visible=True)
-        # Chat handler
         msg.submit(
             gradio_chat,
             inputs=[msg, chatbot, model, system_prompt, temperature, max_tokens, session_id, use_memory],
@@ -1458,8 +1295,7 @@ with gr.Blocks(
     with gr.Tab("🧠 Memoria"):
         gr.Markdown("""
         ## 🧠 Gestione Memoria
-        Il sistema gestisce automaticamente:
-        - **Memoria Breve**: ultimi 20 messaggi (configurabile)
         - **Memoria Lunga**: messaggi più vecchi salvati in SQLite
         - **Riassunti**: riassunti di sessione persistenti
         """)
@@ -1470,7 +1306,7 @@ with gr.Blocks(
                 stats_btn = gr.Button("📊 Mostra Statistiche")
                 clear_mem_btn = gr.Button("🗑️ Elimina Sessione", variant="stop")
-        mem_display = gr.Textbox(label="Risultato", lines=10, interactive=False)
         stats_btn.click(get_memory_info, inputs=[mem_session], outputs=[mem_display])
         clear_mem_btn.click(clear_session_memory, inputs=[mem_session], outputs=[mem_display])
@@ -1479,15 +1315,13 @@ with gr.Blocks(
         gr.Markdown("""
         ## 📖 Documentazione API
-        👉 **[Apri documentazione completa](/api-docs)** (pagina HTML interattiva)
-        👉 **[OpenAPI / Swagger UI](/docs)** (auto-generata da FastAPI)
-        ---
         ### Quick Start
-        #### Chat via GET (test veloce)
         ```
         GET /v1/chat/get?message=Ciao&model=meta/llama-3.1-8b-instruct&session_id=test
         ```
@@ -1504,24 +1338,12 @@ with gr.Blocks(
         curl -N "/v1/chat/stream?message=Raccontami+una+storia"
         ```
-        #### MCP (Model Context Protocol)
         ```bash
         curl -X POST /v1/mcp \\
           -H "Content-Type: application/json" \\
           -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'
         ```
-        ### Configurazione MCP per Claude Desktop / Cursor
-        ```json
-        {
-          "mcpServers": {
-            "nvidia-ai": {
-              "url": "https://YOUR-SPACE.hf.space/v1/mcp",
-              "transport": "http"
-            }
-          }
-        }
-        ```
         """)
@@ -1532,12 +1354,6 @@ with gr.Blocks(
 app = gr.mount_gradio_app(app, demo, path="/ui")
-# Redirect root HTML to UI for browser visits
-@app.get("/app", response_class=HTMLResponse)
-async def redirect_to_ui():
-    return '<html><head><meta http-equiv="refresh" content="0;url=/ui"></head></html>'
 if __name__ == "__main__":
     import uvicorn
     uvicorn.run(app, host="0.0.0.0", port=7860)

 """
 🚀 NVIDIA AI Multi-Model Space
+- Tutti i modelli NVIDIA NIM (35+)
+- Memoria breve (RAM) e lunga (SQLite)
+- API REST GET/POST + SSE Streaming
 - Server MCP integrato
+- UI Web moderna alla root
+- Interfaccia Gradio su /ui
 """
 import os
 import uuid
 import base64
 import asyncio
+from pathlib import Path
 from typing import Optional, AsyncGenerator
 import gradio as gr
 import httpx
 from dotenv import load_dotenv
 from fastapi import FastAPI, Request, Query, HTTPException
+from fastapi.responses import (
+    StreamingResponse, JSONResponse, HTMLResponse, FileResponse
+)
 from fastapi.middleware.cors import CORSMiddleware
+from fastapi.staticfiles import StaticFiles
 from sse_starlette.sse import EventSourceResponse
 from pydantic import BaseModel, Field
     allow_headers=["*"],
 )
+# Mount statici se la cartella esiste
+STATIC_DIR = Path(__file__).parent / "static"
+if STATIC_DIR.exists():
+    app.mount("/static", StaticFiles(directory=str(STATIC_DIR)), name="static")
 # ── Pydantic Models ───────────────────────────────────────────
     summary: Optional[str] = None
+# ══════════════════════════════════════════════════════════════
+# ROOT UI + INFO
+# ══════════════════════════════════════════════════════════════
+@app.get("/", response_class=HTMLResponse)
 async def root():
+    """Serve la UI web principale (static/index.html)"""
+    ui_file = STATIC_DIR / "index.html"
+    if ui_file.exists():
+        return FileResponse(str(ui_file))
+    # Fallback se manca il file
+    return HTMLResponse("""
+    <!DOCTYPE html><html><head><title>NVIDIA AI</title></head>
+    <body style="font-family:sans-serif;background:#0a0e14;color:#e4e6eb;padding:40px">
+      <h1>🚀 NVIDIA AI Multi-Model</h1>
+      <p>UI non trovata (static/index.html mancante).</p>
+      <ul>
+        <li><a href="/ui" style="color:#76b900">Interfaccia Gradio</a></li>
+        <li><a href="/docs" style="color:#76b900">Swagger UI</a></li>
+        <li><a href="/api-docs" style="color:#76b900">API Docs HTML</a></li>
+        <li><a href="/info" style="color:#76b900">Info JSON</a></li>
+      </ul>
+    </body></html>
+    """)
+@app.get("/info")
+async def info():
+    """Info del servizio (JSON)"""
     return {
         "service": "🚀 NVIDIA AI Multi-Model Space",
         "version": "2.0.0",
         "models_available": len(NVIDIA_MODELS),
         "endpoints": {
+            "ui": "/",
+            "gradio": "/ui",
             "docs": "/docs",
             "api_docs_html": "/api-docs",
             "models": "/v1/models",
     }
+# ══════════════════════════════════════════════════════════════
+# MODELS
+# ══════════════════════════════════════════════════════════════
 @app.get("/v1/models")
 async def list_models(category: Optional[str] = None):
     """Lista tutti i modelli disponibili, filtrabile per categoria"""
     return {"model_id": model_id, **NVIDIA_MODELS[model_id]}
+# ══════════════════════════════════════════════════════════════
+# CHAT ENDPOINTS
+# ══════════════════════════════════════════════════════════════
 @app.post("/v1/chat")
 async def chat_post(req: ChatRequest):
     if req.stream:
         return await _stream_response(req, session_id)
     if req.use_memory:
         memory.add_short(session_id, "user", req.message)
         messages = memory.get_full_context(session_id, req.system_prompt)
         raise HTTPException(500, detail=str(e))
 @app.get("/v1/chat/get")
 async def chat_get(
     message: str = Query(..., description="Messaggio utente"),
     return await chat_post(req)
 @app.post("/v1/chat/stream")
 async def chat_stream_post(req: ChatRequest):
     """Endpoint dedicato per streaming SSE via POST"""
     async def event_generator():
         full_response = []
         try:
             yield {
                 "event": "start",
                 "data": json.dumps({
                     "data": json.dumps({"token": chunk})
                 }
             complete = "".join(full_response)
             if req.use_memory:
                 memory.add_short(session_id, "assistant", complete)
             yield {
                 "event": "done",
                 "data": json.dumps({
     return EventSourceResponse(event_generator())
+# ══════════════════════════════════════════════════════════════
+# MEMORY MANAGEMENT
+# ══════════════════════════════════════════════════════════════
 @app.post("/v1/memory")
 async def manage_memory(req: MemoryRequest):
             "error": {"code": code, "message": message}
         })
     if method == "initialize":
         return _mcp_response({
             "protocolVersion": "2024-11-05",
             }
         })
     elif method == "tools/list":
         return _mcp_response({"tools": MCP_TOOLS})
     elif method == "tools/call":
         tool_name = params.get("name", "")
         tool_args = params.get("arguments", {})
         except Exception as e:
             return _mcp_error(-32000, str(e))
     elif method == "resources/list":
         return _mcp_response({
             "resources": [
             ]
         })
     elif method == "resources/read":
         uri = params.get("uri", "")
         if uri == "nvidia://models":
   .param-table { width: 100%; border-collapse: collapse; margin: 10px 0; }
   .param-table th, .param-table td { padding: 10px; border: 1px solid var(--border); text-align: left; }
   .param-table th { background: var(--code-bg); color: var(--accent); }
   .note { background: #1a3a1a; border-left: 4px solid var(--accent); padding: 15px; margin: 15px 0; border-radius: 0 8px 8px 0; }
   a { color: var(--accent); }
 </style>
 </head>
 <div class="note">
   <strong>Base URL:</strong> <code>https://YOUR-SPACE.hf.space</code><br>
+  <strong>Autenticazione:</strong> API key NVIDIA pre-configurata (default)
 </div>
+<h2>📋 Endpoints</h2>
 <div class="endpoint">
 <table class="param-table">
 <tr><th>Metodo</th><th>Endpoint</th><th>Descrizione</th></tr>
+<tr><td><span class="method get">GET</span></td><td><code>/</code></td><td>UI Web principale</td></tr>
+<tr><td><span class="method get">GET</span></td><td><code>/ui</code></td><td>Interfaccia Gradio</td></tr>
+<tr><td><span class="method get">GET</span></td><td><code>/docs</code></td><td>Swagger UI</td></tr>
+<tr><td><span class="method get">GET</span></td><td><code>/v1/models</code></td><td>Lista modelli</td></tr>
 <tr><td><span class="method get">GET</span></td><td><code>/v1/chat/get</code></td><td>Chat via GET</td></tr>
 <tr><td><span class="method post">POST</span></td><td><code>/v1/chat</code></td><td>Chat via POST</td></tr>
+<tr><td><span class="method get">GET</span></td><td><code>/v1/chat/stream</code></td><td>SSE streaming via GET</td></tr>
+<tr><td><span class="method post">POST</span></td><td><code>/v1/chat/stream</code></td><td>SSE streaming via POST</td></tr>
 <tr><td><span class="method post">POST</span></td><td><code>/v1/memory</code></td><td>Gestione memoria</td></tr>
 <tr><td><span class="method post">POST</span></td><td><code>/v1/mcp</code></td><td>Server MCP (JSON-RPC)</td></tr>
 </table>
 </div>
+<h2>💬 Chat</h2>
 <div class="endpoint">
+<pre><code>curl -X POST https://YOUR-SPACE.hf.space/v1/chat \\
   -H "Content-Type: application/json" \\
   -d '{
     "message": "Spiegami la relativit&agrave;",
     "model": "meta/llama-3.1-405b-instruct",
     "session_id": "user-123",
     "use_memory": true
   }'</code></pre>
 </div>
+<h2>📡 SSE Streaming</h2>
 <div class="endpoint">
+<pre><code># JavaScript EventSource
 const es = new EventSource(
+  '/v1/chat/stream?message=Ciao&session_id=demo'
 );
+es.addEventListener('token', (e) =&gt; {
+  console.log(JSON.parse(e.data).token);
 });
+es.addEventListener('done', (e) =&gt; es.close());</code></pre>
+<pre><code># cURL
+curl -N "https://YOUR-SPACE.hf.space/v1/chat/stream?message=Ciao"</code></pre>
 </div>
+<h2>🔌 MCP Server</h2>
 <div class="endpoint">
+<pre><code># Config Claude Desktop / Cursor
 {
   "mcpServers": {
     "nvidia-ai": {
     }
   }
 }</code></pre>
 </div>
 </div>
 # ══════════════════════════════════════════════════════════════
+# GRADIO UI (secondaria, su /ui)
 # ══════════════════════════════════════════════════════════════
 def get_model_choices():
     else:
         messages = [{"role": "system", "content": system_prompt}]
         for h in history:
+            if isinstance(h, dict):
+                if h.get("role") == "user":
+                    messages.append({"role": "user", "content": h["content"]})
+                elif h.get("role") == "assistant":
+                    messages.append({"role": "assistant", "content": h["content"]})
         messages.append({"role": "user", "content": message})
     full_response = ""
 def get_memory_info(session_id: str):
     if not session_id:
+        return {"error": "Inserisci un Session ID"}
+    return memory.get_stats(session_id)
 def clear_session_memory(session_id: str):
     return f"🗑️ Memoria sessione '{session_id}' eliminata"
 with gr.Blocks(
     title="🚀 NVIDIA AI Multi-Model Space",
     theme=gr.themes.Soft(
     # 🚀 NVIDIA AI Multi-Model Space
     **Chat con tutti i modelli NVIDIA NIM** • Memoria Breve/Lunga • API REST • SSE Streaming • MCP Server
+    🏠 [UI Web Principale](/) &nbsp;|&nbsp; 📖 [API Docs HTML](/api-docs) &nbsp;|&nbsp; 🔧 [Swagger UI](/docs) &nbsp;|&nbsp; 🔌 [MCP Endpoint](/v1/mcp)
     """)
     with gr.Tab("💬 Chat"):
                     height=600,
                     type="messages",
                     show_copy_button=True,
                     placeholder="Seleziona un modello e inizia a chattare..."
                 )
                 msg = gr.Textbox(
                 mem_output = gr.JSON(label="📊 Memoria", visible=True)
         msg.submit(
             gradio_chat,
             inputs=[msg, chatbot, model, system_prompt, temperature, max_tokens, session_id, use_memory],
     with gr.Tab("🧠 Memoria"):
         gr.Markdown("""
         ## 🧠 Gestione Memoria
+        - **Memoria Breve**: ultimi 20 messaggi (in RAM)
         - **Memoria Lunga**: messaggi più vecchi salvati in SQLite
         - **Riassunti**: riassunti di sessione persistenti
         """)
                 stats_btn = gr.Button("📊 Mostra Statistiche")
                 clear_mem_btn = gr.Button("🗑️ Elimina Sessione", variant="stop")
+        mem_display = gr.JSON(label="Risultato")
         stats_btn.click(get_memory_info, inputs=[mem_session], outputs=[mem_display])
         clear_mem_btn.click(clear_session_memory, inputs=[mem_session], outputs=[mem_display])
         gr.Markdown("""
         ## 📖 Documentazione API
+        👉 **[Apri documentazione completa](/api-docs)**
+        👉 **[OpenAPI / Swagger UI](/docs)**
         ### Quick Start
+        #### Chat via GET
         ```
         GET /v1/chat/get?message=Ciao&model=meta/llama-3.1-8b-instruct&session_id=test
         ```
         curl -N "/v1/chat/stream?message=Raccontami+una+storia"
         ```
+        #### MCP
         ```bash
         curl -X POST /v1/mcp \\
           -H "Content-Type: application/json" \\
           -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'
         ```
         """)
 app = gr.mount_gradio_app(app, demo, path="/ui")
 if __name__ == "__main__":
     import uvicorn
     uvicorn.run(app, host="0.0.0.0", port=7860)