v2.0 — Streaming, Session-Cookies, Gedächtnis-Fix, Multi-Device

Features:
- Streaming Responses: Token-für-Token via Server-Sent Events
- Session-Cookies: Cookie-basierte Session, läuft 1 Jahr
- Multi-Device: Anderes Gerät = gleiche Session = gleicher Chat
- /api/history endpoint: Lädt vollen Chat-Verlauf beim Reconnect
- Gedächtnis-Fix: Chat-Verlauf beim Neuladen korrekt aus DB
- Session-basiertes Gedächtnis (session_id in jeder Nachricht)
- UI: Animated typing indicator (3 dots)
- UI: Session-ID Anzeige im Header
- Auto-History-Load beim Seitenaufruf (falls nicht server-seitig gerendert)
- Prompt: Anti-Repetition Rule hinzugefügt
- Prompt: Erinnerungen-Referenz-Regel hinzugefügt
- DB Migration: Alte Nachrichten werden zu Default-Session migriert
- DB Schema: session_id Spalte in messages und summaries

main.py

@@ -1,17 +1,16 @@
 #!/usr/bin/env python3
 """
-NeonChat — Lokaler RP Chatbot mit Ollama
+NeonChat v2 — Lokaler RP Chatbot mit Ollama
-Character-basiert, mit Kurz- und Langzeitgedächtnis
+Features: Streaming, Session-Cookies, Multi-Device, Langzeitgedächtnis
 """
 
-import json, os, time, asyncio, sqlite3, re
+import json, os, time, asyncio, sqlite3, re, secrets
 from pathlib import Path
 from datetime import datetime
-from contextlib import asynccontextmanager
 
 import httpx
-from fastapi import FastAPI, Request, UploadFile, File, HTTPException
+from fastapi import FastAPI, Request, UploadFile, File, HTTPException, Response, Cookie
-from fastapi.responses import HTMLResponse, JSONResponse, RedirectResponse
+from fastapi.responses import HTMLResponse, JSONResponse, RedirectResponse, StreamingResponse
 from fastapi.staticfiles import StaticFiles
 from fastapi.templating import Jinja2Templates
 from pydantic import BaseModel
@@ -22,7 +21,9 @@ CHARACTERS_DIR = BASE_DIR / "characters"
 DATA_DIR = BASE_DIR / "data"
 DB_PATH = DATA_DIR / "memory.db"
 OLLAMA_URL = os.environ.get("OLLAMA_URL", "http://localhost:11434")
-DEFAULT_MODEL = os.environ.get("OLLAMA_MODEL", "llama3.1:8b")
+DEFAULT_MODEL = os.environ.get("OLLAMA_MODEL", "gemma4:12b")
+SESSION_COOKIE_NAME = "neonchat_session"
+SESSION_DURATION = 60 * 60 * 24 * 365  # 1 Jahr
 
 # === Datenbank ===
 def init_db():
@@ -31,6 +32,7 @@ def init_db():
     conn.executescript("""
         CREATE TABLE IF NOT EXISTS messages (
             id INTEGER PRIMARY KEY AUTOINCREMENT,
+            session_id TEXT NOT NULL,
             character TEXT NOT NULL,
             role TEXT NOT NULL,
             content TEXT NOT NULL,
@@ -39,21 +41,65 @@ def init_db():
         );
         CREATE TABLE IF NOT EXISTS summaries (
             id INTEGER PRIMARY KEY AUTOINCREMENT,
+            session_id TEXT NOT NULL,
             character TEXT NOT NULL,
             summary TEXT NOT NULL,
             timestamp REAL NOT NULL
         );
-        CREATE INDEX IF NOT EXISTS idx_messages_character ON messages(character, timestamp);
+        CREATE TABLE IF NOT EXISTS sessions (
-        CREATE INDEX IF NOT EXISTS idx_summaries_character ON summaries(character, timestamp);
+            id TEXT PRIMARY KEY,
+            created REAL NOT NULL,
+            name TEXT DEFAULT ''
+        );
+        CREATE INDEX IF NOT EXISTS idx_messages_session ON messages(session_id, character, timestamp);
+        CREATE INDEX IF NOT EXISTS idx_summaries_session ON summaries(session_id, character, timestamp);
     """)
     conn.commit()
     conn.close()
 
 init_db()
 
+# === Session Management ===
+def get_or_create_session(session_id: str = None) -> str:
+    conn = sqlite3.connect(str(DB_PATH))
+    if session_id:
+        cur = conn.execute("SELECT id FROM sessions WHERE id = ?", (session_id,))
+        if cur.fetchone():
+            conn.close()
+            return session_id
+    # Create new session
+    new_id = secrets.token_hex(16)
+    conn.execute("INSERT INTO sessions (id, created) VALUES (?, ?)", (new_id, time.time()))
+    conn.commit()
+    conn.close()
+    return new_id
+
+def migrate_old_messages():
+    """Migriert alte Nachrichten ohne session_id zu einer default Session."""
+    conn = sqlite3.connect(str(DB_PATH))
+    cur = conn.execute("SELECT COUNT(*) FROM messages WHERE session_id = '' OR session_id IS NULL")
+    count = cur.fetchone()[0]
+    if count > 0:
+        # Create default session
+        default_session = "default_" + secrets.token_hex(8)
+        conn.execute("INSERT OR IGNORE INTO sessions (id, created, name) VALUES (?, ?, 'default')", (default_session, time.time()))
+        conn.execute("UPDATE messages SET session_id = ? WHERE session_id = '' OR session_id IS NULL", (default_session,))
+        conn.execute("UPDATE summaries SET session_id = ? WHERE session_id = '' OR session_id IS NULL", (default_session,))
+        conn.commit()
+    conn.close()
+
+migrate_old_messages()
+
 # === Character Loading ===
 def load_character(name: str) -> dict:
     path = CHARACTERS_DIR / f"{name}.json"
+    if not path.exists():
+        path = CHARACTERS_DIR / f"{name.lower()}.json"
+    if not path.exists():
+        for f in CHARACTERS_DIR.glob("*.json"):
+            if f.stem.lower() == name.lower():
+                path = f
+                break
     if not path.exists():
         raise HTTPException(status_code=404, detail=f"Character {name} not found")
     with open(path, "r", encoding="utf-8") as f:
@@ -66,30 +112,40 @@ def list_characters() -> list:
             data = json.load(fh)
             chars.append({
                 "name": data.get("name", f.stem),
+                "filename": f.stem,
                 "avatar": data.get("avatar", "👤"),
                 "description": data.get("description", ""),
                 "greeting": data.get("greeting", ""),
             })
     return chars
 
-# === Memory System ===
+# === Memory System (Session-basiert) ===
-def get_short_term_memory(character: str, limit: int = 20) -> list:
+def get_short_term_memory(session_id: str, character: str, limit: int = 20) -> list:
-    """Holt die letzten N Nachrichten als Kurzzeitgedächtnis."""
     conn = sqlite3.connect(str(DB_PATH))
     cur = conn.execute(
-        "SELECT role, content FROM messages WHERE character = ? ORDER BY timestamp DESC LIMIT ?",
+        "SELECT role, content FROM messages WHERE session_id = ? AND character = ? ORDER BY timestamp DESC LIMIT ?",
-        (character, limit)
+        (session_id, character, limit)
     )
     rows = cur.fetchall()
     conn.close()
     return [{"role": r[0], "content": r[1]} for r in reversed(rows)]
 
-def get_long_term_memory(character: str) -> str:
+def get_full_history(session_id: str, character: str, limit: int = 200) -> list:
-    """Holt die letzten Zusammenfassungen als Langzeitgedächtnis."""
+    """Holt vollständigen Chat-Verlauf für die UI-Anzeige."""
     conn = sqlite3.connect(str(DB_PATH))
     cur = conn.execute(
-        "SELECT summary FROM summaries WHERE character = ? ORDER BY timestamp DESC LIMIT 3",
+        "SELECT role, content, timestamp FROM messages WHERE session_id = ? AND character = ? ORDER BY timestamp ASC LIMIT ?",
-        (character,)
+        (session_id, character, limit)
+    )
+    rows = cur.fetchall()
+    conn.close()
+    return [{"role": r[0], "content": r[1], "timestamp": r[2]} for r in rows]
+
+def get_long_term_memory(session_id: str, character: str) -> str:
+    conn = sqlite3.connect(str(DB_PATH))
+    cur = conn.execute(
+        "SELECT summary FROM summaries WHERE session_id = ? AND character = ? ORDER BY timestamp DESC LIMIT 3",
+        (session_id, character)
     )
     rows = cur.fetchall()
     conn.close()
@@ -97,32 +153,30 @@ def get_long_term_memory(character: str) -> str:
         return ""
     return "\n\n".join([r[0] for r in rows])
 
-def save_message(character: str, role: str, content: str):
+def save_message(session_id: str, character: str, role: str, content: str):
     conn = sqlite3.connect(str(DB_PATH))
     conn.execute(
-        "INSERT INTO messages (character, role, content, timestamp) VALUES (?, ?, ?, ?)",
+        "INSERT INTO messages (session_id, character, role, content, timestamp) VALUES (?, ?, ?, ?, ?)",
-        (character, role, content, time.time())
+        (session_id, character, role, content, time.time())
     )
     conn.commit()
     conn.close()
 
-def needs_summary(character: str) -> bool:
+def needs_summary(session_id: str, character: str) -> bool:
-    """Prüft ob eine neue Zusammenfassung nötig ist (alle 20 Nachrichten)."""
     conn = sqlite3.connect(str(DB_PATH))
     cur = conn.execute(
-        "SELECT COUNT(*) FROM messages WHERE character = ? AND summary IS NULL",
+        "SELECT COUNT(*) FROM messages WHERE session_id = ? AND character = ? AND summary IS NULL",
-        (character,)
+        (session_id, character)
     )
     count = cur.fetchone()[0]
     conn.close()
     return count >= 20
 
-async def generate_summary(character: str):
+async def generate_summary(session_id: str, character: str):
-    """Erstellt eine Zusammenfassung der letzten 20 Nachrichten."""
     conn = sqlite3.connect(str(DB_PATH))
     cur = conn.execute(
-        "SELECT id, role, content FROM messages WHERE character = ? AND summary IS NULL ORDER BY timestamp ASC LIMIT 20",
+        "SELECT id, role, content FROM messages WHERE session_id = ? AND character = ? AND summary IS NULL ORDER BY timestamp ASC LIMIT 20",
-        (character,)
+        (session_id, character)
     )
     rows = cur.fetchall()
     if not rows:
@@ -143,35 +197,39 @@ Gespräch:
 Zusammenfassung (auf Deutsch, kurz):"""
 
     try:
-        async with httpx.AsyncClient() as client:
+        async with httpx.AsyncClient(timeout=60) as client:
             resp = await client.post(f"{OLLAMA_URL}/api/generate", json={
                 "model": DEFAULT_MODEL,
                 "prompt": prompt,
                 "stream": False,
+                "think": False,
                 "options": {"temperature": 0.3, "num_predict": 300}
             })
             summary = resp.json().get("response", "").strip()
     except:
         summary = "Zusammenfassung nicht verfügbar."
     
+    if not summary:
+        summary = "Zusammenfassung nicht verfügbar."
+    
     conn.execute(
-        "INSERT INTO summaries (character, summary, timestamp) VALUES (?, ?, ?)",
+        "INSERT INTO summaries (session_id, character, summary, timestamp) VALUES (?, ?, ?, ?)",
-        (character, summary, time.time())
+        (session_id, character, summary, time.time())
     )
     for r in rows:
         conn.execute("UPDATE messages SET summary = ? WHERE id = ?", (summary, r[0]))
     conn.commit()
     conn.close()
 
-def clear_memory(character: str):
+def clear_memory(session_id: str, character: str):
     conn = sqlite3.connect(str(DB_PATH))
-    conn.execute("DELETE FROM messages WHERE character = ?", (character,))
+    conn.execute("DELETE FROM messages WHERE session_id = ? AND character = ?", (session_id, character))
-    conn.execute("DELETE FROM summaries WHERE character = ?", (character,))
+    conn.execute("DELETE FROM summaries WHERE session_id = ? AND character = ?", (session_id, character))
     conn.commit()
     conn.close()
 
 # === Prompt Builder ===
-def build_prompt(character: dict, short_term: list, long_term: str, user_message: str) -> str:
+def build_prompt(character: dict, short_term: list, long_term: str, user_message: str) -> list:
     name = character.get("name", "AI")
     personality = character.get("personality", "")
     background = character.get("background", "")
@@ -200,7 +258,9 @@ REGELN:
 - Verwende keine Formulierungen wie "als KI" oder "als Sprachmodell"
 - Reagiere emotional angemessen
 - Sei nicht übermäßig freundlich — bleibe authentisch
-- Verwende deutsche Umgangssprache wo passend"""
+- Verwende deutsche Umgangssprache wo passend
+- Wiederhole dich nicht — variiere deine Ausdrücke
+- Wenn der Nutzer etwas sagt das du schon weißt (aus Erinnerungen), beziehe dich darauf"""
 
     if long_term:
         system += f"\n\nERINNERUNGEN (Langzeitgedächtnis):\n{long_term}"
@@ -221,48 +281,109 @@ templates = Jinja2Templates(directory=str(BASE_DIR / "templates"))
 app.mount("/static", StaticFiles(directory=str(BASE_DIR / "static")), name="static")
 
 @app.get("/", response_class=HTMLResponse)
-async def index(request: Request):
+async def index(request: Request, session_id: str = Cookie(None, alias=SESSION_COOKIE_NAME)):
+    sid = get_or_create_session(session_id)
     chars = list_characters()
-    return templates.TemplateResponse(request, "index.html", {"characters": chars})
+    response = templates.TemplateResponse(request, "index.html", {"characters": chars})
+    if not session_id or session_id != sid:
+        response.set_cookie(SESSION_COOKIE_NAME, sid, max_age=SESSION_DURATION, httponly=False, samesite="lax")
+    return response
 
 @app.get("/create", response_class=HTMLResponse)
 async def create_page(request: Request):
     return templates.TemplateResponse(request, "create.html", {})
 
 @app.get("/chat/{character_name}", response_class=HTMLResponse)
-async def chat_page(request: Request, character_name: str):
+async def chat_page(request: Request, character_name: str, session_id: str = Cookie(None, alias=SESSION_COOKIE_NAME)):
+    sid = get_or_create_session(session_id)
     char = load_character(character_name)
-    history = get_short_term_memory(character_name, limit=50)
+    # Fix: Load FULL history for display, not just short-term
-    long_term = get_long_term_memory(character_name)
+    history = get_full_history(sid, character_name, limit=200)
-    return templates.TemplateResponse(request, "chat.html", {
+    long_term = get_long_term_memory(sid, character_name)
+    response = templates.TemplateResponse(request, "chat.html", {
         "character": char,
         "history": history,
         "long_term": long_term,
+        "session_id": sid,
     })
+    if not session_id or session_id != sid:
+        response.set_cookie(SESSION_COOKIE_NAME, sid, max_age=SESSION_DURATION, httponly=False, samesite="lax")
+    return response
 
 @app.post("/api/chat/{character_name}")
-async def chat_api(character_name: str, request: Request):
+async def chat_api(character_name: str, request: Request, session_id: str = Cookie(None, alias=SESSION_COOKIE_NAME)):
+    sid = get_or_create_session(session_id)
     char = load_character(character_name)
     body = await request.json()
     user_message = body.get("message", "")
+    stream = body.get("stream", True)
     
     if not user_message.strip():
         return JSONResponse({"error": "Leere Nachricht"}, status_code=400)
     
-    save_message(character_name, "user", user_message)
+    save_message(sid, character_name, "user", user_message)
     
-    short_term = get_short_term_memory(character_name, limit=20)
+    short_term = get_short_term_memory(sid, character_name, limit=20)
-    long_term = get_long_term_memory(character_name)
+    long_term = get_long_term_memory(sid, character_name)
     messages = build_prompt(char, short_term, long_term, user_message)
     
     model = char.get("model") or DEFAULT_MODEL
     
+    if stream:
+        # Streaming via Server-Sent Events
+        async def stream_response():
+            full_response = ""
+            try:
+                async with httpx.AsyncClient(timeout=300) as client:
+                    async with client.stream("POST", f"{OLLAMA_URL}/api/chat", json={
+                        "model": model,
+                        "messages": messages,
+                        "stream": True,
+                        "think": False,
+                        "options": {
+                            "temperature": char.get("temperature", 0.8),
+                            "top_p": 0.9,
+                            "num_predict": 800,
+                            "repeat_penalty": 1.1,
+                        }
+                    }) as resp:
+                        async for line in resp.aiter_lines():
+                            if not line.strip():
+                                continue
+                            try:
+                                chunk = json.loads(line)
+                                content = chunk.get("message", {}).get("content", "")
+                                if content:
+                                    full_response += content
+                                    yield f"data: {json.dumps({'token': content})}\n\n"
+                                if chunk.get("done"):
+                                    break
+                            except json.JSONDecodeError:
+                                continue
+            except Exception as e:
+                full_response = f"*(Fehler: {e})*"
+                yield f"data: {json.dumps({'token': full_response})}\n\n"
+            
+            # Save complete response
+            if full_response.strip():
+                save_message(sid, character_name, "assistant", full_response.strip())
+            
+            # Check if summary needed
+            if needs_summary(sid, character_name):
+                asyncio.create_task(generate_summary(sid, character_name))
+            
+            yield f"data: {json.dumps({'done': True})}\n\n"
+        
+        return StreamingResponse(stream_response(), media_type="text/event-stream")
+    else:
+        # Non-streaming fallback
         try:
             async with httpx.AsyncClient(timeout=300) as client:
                 resp = await client.post(f"{OLLAMA_URL}/api/chat", json={
                     "model": model,
                     "messages": messages,
                     "stream": False,
+                    "think": False,
                     "options": {
                         "temperature": char.get("temperature", 0.8),
                         "top_p": 0.9,
@@ -272,18 +393,13 @@ async def chat_api(character_name: str, request: Request):
                 })
                 data = resp.json()
                 response_text = data.get("message", {}).get("content", "").strip()
-            # Some models (gemma4) put output in "thinking" field
-            if not response_text:
-                thinking = data.get("message", {}).get("thinking", "")
-                if thinking:
-                    response_text = thinking.strip()
         except Exception as e:
             response_text = f"*(Fehler: {e})*"
         
-    save_message(character_name, "assistant", response_text)
+        save_message(sid, character_name, "assistant", response_text)
         
-    if needs_summary(character_name):
+        if needs_summary(sid, character_name):
-        asyncio.create_task(generate_summary(character_name))
+            asyncio.create_task(generate_summary(sid, character_name))
         
         return JSONResponse({"response": response_text, "character": character_name})
 
@@ -309,7 +425,7 @@ async def api_create_character(request: Request):
         "scenario": body.get("scenario", ""),
         "nsfw": body.get("nsfw", ""),
         "greeting": body.get("greeting", ""),
-        "model": body.get("model", DEFAULT_MODEL),
+        "model": body.get("model", ""),
         "temperature": body.get("temperature", 0.8),
     }
     
@@ -324,16 +440,25 @@ async def api_get_character(name: str):
     return JSONResponse(char)
 
 @app.post("/api/memory/clear/{character_name}")
-async def api_clear_memory(character_name: str):
+async def api_clear_memory(character_name: str, session_id: str = Cookie(None, alias=SESSION_COOKIE_NAME)):
-    clear_memory(character_name)
+    sid = get_or_create_session(session_id)
+    clear_memory(sid, character_name)
     return JSONResponse({"status": "ok"})
 
 @app.get("/api/memory/{character_name}")
-async def api_get_memory(character_name: str):
+async def api_get_memory(character_name: str, session_id: str = Cookie(None, alias=SESSION_COOKIE_NAME)):
-    short = get_short_term_memory(character_name, limit=100)
+    sid = get_or_create_session(session_id)
-    long = get_long_term_memory(character_name)
+    short = get_short_term_memory(sid, character_name, limit=100)
+    long = get_long_term_memory(sid, character_name)
     return JSONResponse({"short_term": short, "long_term": long})
 
+@app.get("/api/history/{character_name}")
+async def api_get_history(character_name: str, session_id: str = Cookie(None, alias=SESSION_COOKIE_NAME)):
+    """API endpoint to get full chat history — for reconnecting devices."""
+    sid = get_or_create_session(session_id)
+    history = get_full_history(sid, character_name, limit=200)
+    return JSONResponse({"history": history, "session_id": sid})
+
 @app.get("/api/models")
 async def api_list_models():
     try:

templates/chat.html

@@ -13,7 +13,6 @@
         }
         body { font-family: 'Segoe UI', system-ui, sans-serif; background: var(--bg); color: var(--text); height: 100vh; display: flex; flex-direction: column; }
         
-        /* Header */
         header { display: flex; align-items: center; gap: 14px; padding: 14px 20px; background: var(--surface); border-bottom: 1px solid var(--border); }
         .avatar { font-size: 2rem; width: 48px; height: 48px; display: flex; align-items: center; justify-content: center; background: var(--surface2); border-radius: 50%; }
         .char-name { font-size: 1.2rem; font-weight: 600; }
@@ -23,18 +22,18 @@
         .btn:hover { border-color: var(--accent); }
         .btn-danger:hover { border-color: var(--accent2); color: var(--accent2); }
         
-        /* Chat */
+        .chat-container { flex: 1; overflow-y: auto; padding: 20px; display: flex; flex-direction: column; gap: 12px; scroll-behavior: smooth; }
-        .chat-container { flex: 1; overflow-y: auto; padding: 20px; display: flex; flex-direction: column; gap: 12px; }
         .msg { max-width: 75%; padding: 12px 16px; border-radius: 16px; line-height: 1.6; font-size: 0.95rem; white-space: pre-wrap; word-break: break-word; }
         .msg-user { align-self: flex-end; background: var(--user); color: white; border-bottom-right-radius: 4px; }
         .msg-bot { align-self: flex-start; background: var(--surface); border: 1px solid var(--border); border-bottom-left-radius: 4px; }
         .msg-system { align-self: center; background: transparent; color: var(--muted); font-size: 0.85rem; font-style: italic; }
         .msg-greeting { align-self: flex-start; background: var(--surface); border: 1px solid var(--accent); border-bottom-left-radius: 4px; }
-        .typing { align-self: flex-start; color: var(--muted); font-size: 0.9rem; padding: 8px 16px; }
+        .typing { align-self: flex-start; color: var(--muted); font-size: 0.9rem; padding: 8px 16px; display: flex; gap: 4px; }
-        .typing::after { content: '●●●'; animation: blink 1.2s infinite; letter-spacing: 3px; }
+        .typing span { width: 8px; height: 8px; border-radius: 50%; background: var(--accent); animation: blink 1.2s infinite; }
+        .typing span:nth-child(2) { animation-delay: 0.2s; }
+        .typing span:nth-child(3) { animation-delay: 0.4s; }
         @keyframes blink { 0%,100% { opacity: 0.2; } 50% { opacity: 1; } }
         
-        /* Input */
         .input-area { padding: 16px 20px; background: var(--surface); border-top: 1px solid var(--border); }
         .input-row { display: flex; gap: 10px; max-width: 1200px; margin: 0 auto; }
         textarea { flex: 1; background: var(--surface2); border: 1px solid var(--border); border-radius: 12px; padding: 12px 16px; color: var(--text); font-size: 0.95rem; font-family: inherit; resize: none; outline: none; transition: border 0.15s; max-height: 120px; }
@@ -44,7 +43,6 @@
         .send-btn:hover { opacity: 0.9; }
         .send-btn:disabled { opacity: 0.4; cursor: not-allowed; }
         
-        /* Memory Panel */
         .memory-panel { position: fixed; top: 0; right: -400px; width: 380px; height: 100vh; background: var(--surface); border-left: 1px solid var(--border); padding: 20px; overflow-y: auto; transition: right 0.3s; z-index: 100; }
         .memory-panel.open { right: 0; }
         .memory-panel h3 { margin-bottom: 12px; font-size: 1rem; color: var(--accent); }
@@ -54,6 +52,8 @@
         .memory-overlay { position: fixed; top: 0; left: 0; width: 100vw; height: 100vh; background: rgba(0,0,0,0.5); z-index: 99; display: none; }
         .memory-overlay.show { display: block; }
         
+        .session-info { font-size: 0.75rem; color: var(--muted); padding: 4px 8px; background: var(--surface2); border-radius: 6px; }
+        
         @media (max-width: 768px) {
             .msg { max-width: 90%; }
         }
@@ -68,6 +68,7 @@
             <div class="char-desc">{{ character.description[:60] }}{% if character.description|length > 60 %}...{% endif %}</div>
         </div>
         <div class="header-actions">
+            <span class="session-info" id="sessionInfo">Session: ...</span>
             <button class="btn" onclick="toggleMemory()">🧠 Gedächtnis</button>
             <button class="btn btn-danger" onclick="clearMemory()">🗑️ Vergessen</button>
         </div>
@@ -89,7 +90,6 @@
         </div>
     </div>
 
-    <!-- Memory Panel -->
     <div class="memory-overlay" id="overlay" onclick="toggleMemory()"></div>
     <div class="memory-panel" id="memoryPanel">
         <h3>🧠 Gedächtnis — {{ character.name }}</h3>
@@ -101,11 +101,16 @@
 
     <script>
         const charName = "{{ character.name }}";
+        const charFilename = "{{ character.filename if character.filename else character_name }}";
         const chat = document.getElementById('chat');
         const input = document.getElementById('input');
         const sendBtn = document.getElementById('send');
         let isSending = false;
 
+        // Show session ID (short)
+        const sessionId = "{{ session_id }}";
+        document.getElementById('sessionInfo').textContent = 'Session: ' + sessionId.substring(0, 8) + '...';
+
         // Auto-scroll
         chat.scrollTop = chat.scrollHeight;
 
@@ -140,27 +145,60 @@
             // Typing indicator
             const typing = document.createElement('div');
             typing.className = 'typing';
-            typing.textContent = '';
+            typing.innerHTML = '<span></span><span></span><span></span>';
             chat.appendChild(typing);
+            chat.scrollTop = chat.scrollHeight;
 
-            try {
+            // Streaming via EventSource (fetch with streaming)
-                const resp = await fetch(`/api/chat/${encodeURIComponent(charName)}`, {
-                    method: 'POST',
-                    headers: {'Content-Type': 'application/json'},
-                    body: JSON.stringify({message: msg})
-                });
-                const data = await resp.json();
-                typing.remove();
             const botDiv = document.createElement('div');
             botDiv.className = 'msg msg-bot';
-                botDiv.textContent = data.response;
             chat.appendChild(botDiv);
+
+            try {
+                const resp = await fetch(`/api/chat/${encodeURIComponent(charFilename)}`, {
+                    method: 'POST',
+                    headers: {'Content-Type': 'application/json'},
+                    body: JSON.stringify({message: msg, stream: true})
+                });
+
+                const reader = resp.body.getReader();
+                const decoder = new TextDecoder();
+                let buffer = '';
+                let fullResponse = '';
+
+                while (true) {
+                    const {done, value} = await reader.read();
+                    if (done) break;
+                    buffer += decoder.decode(value, {stream: true});
+                    
+                    const lines = buffer.split('\n');
+                    buffer = lines.pop();
+                    
+                    for (const line of lines) {
+                        if (!line.startsWith('data: ')) continue;
+                        try {
+                            const data = JSON.parse(line.slice(6));
+                            if (data.token) {
+                                fullResponse += data.token;
+                                botDiv.textContent = fullResponse;
                                 chat.scrollTop = chat.scrollHeight;
+                            }
+                            if (data.done) {
+                                // Streaming complete
+                            }
+                        } catch(e) {}
+                    }
+                }
+                typing.remove();
+                if (!fullResponse.trim()) {
+                    botDiv.textContent = '*(Keine Antwort)*';
+                }
             } catch(e) {
                 typing.remove();
+                botDiv.remove();
                 const errDiv = document.createElement('div');
                 errDiv.className = 'msg msg-system';
-                errDiv.textContent = '⚠️ Verbindungsfehler';
+                errDiv.textContent = '⚠️ Verbindungsfehler: ' + e.message;
                 chat.appendChild(errDiv);
             }
 
@@ -184,7 +222,7 @@
 
         async function loadMemory() {
             try {
-                const resp = await fetch(`/api/memory/${encodeURIComponent(charName)}`);
+                const resp = await fetch(`/api/memory/${encodeURIComponent(charFilename)}`);
                 const data = await resp.json();
                 const shortDiv = document.getElementById('shortTerm');
                 shortDiv.innerHTML = '';
@@ -205,7 +243,7 @@
 
         async function clearMemory() {
             if (!confirm(`Gedächtnis von ${charName} wirklich löschen?`)) return;
-            await fetch(`/api/memory/clear/${encodeURIComponent(charName)}`, {method: 'POST'});
+            await fetch(`/api/memory/clear/${encodeURIComponent(charFilename)}`, {method: 'POST'});
             chat.innerHTML = '';
             if ("{{ character.greeting }}") {
                 const g = document.createElement('div');
@@ -214,6 +252,42 @@
                 chat.appendChild(g);
             }
         }
+
+        // Load history on reconnect (fixes the "conversation is somewhere else" bug)
+        async function loadHistory() {
+            try {
+                const resp = await fetch(`/api/history/${encodeURIComponent(charFilename)}`);
+                const data = await resp.json();
+                if (data.history && data.history.length > 0) {
+                    // Only reload if we have more messages than currently shown
+                    const currentMsgs = chat.querySelectorAll('.msg').length;
+                    if (data.history.length > currentMsgs || currentMsgs === 0) {
+                        chat.innerHTML = '';
+                        if ("{{ character.greeting }}") {
+                            const g = document.createElement('div');
+                            g.className = 'msg msg-greeting';
+                            g.textContent = "{{ character.greeting }}";
+                            chat.appendChild(g);
+                        }
+                        data.history.forEach(h => {
+                            const div = document.createElement('div');
+                            div.className = `msg ${h.role === 'user' ? 'msg-user' : 'msg-bot'}`;
+                            div.textContent = h.content;
+                            chat.appendChild(div);
+                        });
+                        chat.scrollTop = chat.scrollHeight;
+                    }
+                }
+            } catch(e) {
+                console.error('History load failed:', e);
+            }
+        }
+
+        // Load history on page load (fixes reconnect issue)
+        // Only if we don't already have history from server-side rendering
+        {% if not history %}
+        loadHistory();
+        {% endif %}
     </script>
 </body>
 </html>

templates/index.html

@@ -44,7 +44,7 @@
         <p class="subtitle">Lokaler RP-Chatbot · Wähle einen Charakter oder erstelle einen neuen</p>
         <div class="char-grid">
             {% for c in characters %}
-            <div class="char-card" onclick="location.href='/chat/{{ c.name | urlencode }}'">
+            <div class="char-card" onclick="location.href='/chat/{{ c.filename | urlencode }}'">
                 <div class="char-avatar">{{ c.avatar }}</div>
                 <div class="char-info">
                     <h3>{{ c.name }}</h3>