Primary-text index for personas, faster retrieval, OpenAI model options

- Contemporary and Kafka personas search only "Die Verwandlung" (own index and tool description); per-persona subtitle
- New contemporary portrait and avatar alt text
- ChatGPT: gpt-6.1-sol, reasoning_effort low via Responses API; default in the model menu
- GPT-Realtime-2.1 Mini as third option (realtime_chat.py, WebSocket, text only)
- Load .env for OPENAI_API_KEY
- Reranker in fp16 on GPU with 25 candidates; reranker, embeddings and BM25 shared and built at server start
- BM25 stem cache (first search ~31s -> ~1.7s)
- Timing output per LLM call and search

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

server.py

@@ -18,7 +18,10 @@ from personas import DEFAULT_PERSONA, PERSONAS, PUBLIC_FIELDS
 load_dotenv()
 
 OLLAMA_MODEL = "gemma4:12b"
-OPENAI_MODEL = "gpt-6-luna"
+OPENAI_MODEL = "gpt-6.1-sol"
+REALTIME_MODEL = "gpt-realtime-2.1-mini"  # OpenAI Realtime API, text only (realtime_chat.py)
+# Model name per provider, for the console trace
+MODEL_NAMES = {"ollama": OLLAMA_MODEL, "openai": OPENAI_MODEL, "realtime": REALTIME_MODEL}
 # One RAG instance per (LLM provider, vector store), for on-the-fly provider switching.
 # Each persona names its vector store in personas.py; personas sharing one share the instance.
 rags = {}
@@ -30,16 +33,23 @@ for vectorstore in sorted({p["vectorstore"] for p in PERSONAS.values()}):
         openai_model=OPENAI_MODEL,
     )
     if os.environ.get("OPENAI_API_KEY"):
-        try:
-            rags[("openai", vectorstore)] = LocalRAG(
-                vectorstore_path=vectorstore,
-                llm_provider="openai",
-                ollama_model=OLLAMA_MODEL,
-                openai_model=OPENAI_MODEL,
-            )
-        except Exception as e:
-            print(f"OpenAI RAG not available for {vectorstore}: {e}")
-openai_available = any(provider == "openai" for provider, _ in rags)
+        for provider in ("openai", "realtime"):
+            try:
+                rags[(provider, vectorstore)] = LocalRAG(
+                    vectorstore_path=vectorstore,
+                    llm_provider=provider,
+                    ollama_model=OLLAMA_MODEL,
+                    openai_model=OPENAI_MODEL,
+                    realtime_model=REALTIME_MODEL,
+                )
+            except Exception as e:
+                print(f"{provider} RAG not available for {vectorstore}: {e}")
+available = {provider for provider, _ in rags}
+openai_available = "openai" in available
+# Load the reranker and build the BM25 indexes at startup (shared copies) so the first search
+# is not delayed
+for rag in rags.values():
+    rag.preload()
 
 app = FastAPI(title="Local RAG Chat", version="1.0.0")
 # Images for the chat GUI (researcher portrait, avatar)
@@ -54,7 +64,7 @@ class ChatMessage(BaseModel):
 class ChatRequest(BaseModel):
     message: str
     history: list[ChatMessage] = []  # previous turns for conversation context
-    llm_provider: str = "ollama"  # "ollama" | "openai"
+    llm_provider: str = "ollama"  # "ollama" | "openai" | "realtime"
     persona: str = DEFAULT_PERSONA  # key in personas.PERSONAS
 
 
@@ -83,7 +93,7 @@ def chat_view():
 
 def _get_rag(provider: str, persona: str):
     """Return the RAG instance for the provider and the persona's vector store.
-    Fall back to Ollama if OpenAI is unavailable."""
+    Fall back to Ollama if the provider is unavailable."""
     vectorstore = PERSONAS[persona]["vectorstore"]
     return rags.get((provider, vectorstore)) or rags[("ollama", vectorstore)]
 
@@ -93,7 +103,7 @@ def chat(request: ChatRequest):
     """Handle a chat message and return the RAG answer."""
     if not request.message or not request.message.strip():
         return ChatResponse(answer="", error="Message cannot be empty")
-    if request.llm_provider == "openai" and not openai_available:
+    if request.llm_provider in ("openai", "realtime") and request.llm_provider not in available:
         return ChatResponse(answer="", error="OpenAI not configured. Set OPENAI_API_KEY.")
     if request.persona not in PERSONAS:
         return ChatResponse(answer="", error=f"Unknown persona: {request.persona}")
@@ -120,8 +130,8 @@ def chat(request: ChatRequest):
                 print(f"  [{i + 1}] {chunk.get('source', '')} p.{chunk.get('page', '?')} s={chunk.get('score')} | {preview!r}")
         else:
             print(f"\n[RAG] No corpus search for this message")
-        provider_label = "OpenAI" if request.llm_provider == "openai" else "Ollama"
-        model_name = OPENAI_MODEL if request.llm_provider == "openai" else OLLAMA_MODEL
+        provider_label = {"openai": "OpenAI", "realtime": "OpenAI Realtime"}.get(request.llm_provider, "Ollama")
+        model_name = MODEL_NAMES.get(request.llm_provider, OLLAMA_MODEL)
         print(f"[RAG] LLM response ({provider_label} / {model_name}, {request.persona}):\n{answer}")
 
         return ChatResponse(answer=answer, retrieved=retrieved, searches=searches)
@@ -139,7 +149,7 @@ def health():
 @app.get("/api/providers")
 def providers():
     """Return which LLM providers are available."""
-    return {"ollama": True, "openai": openai_available}
+    return {"ollama": True, "openai": openai_available, "realtime": "realtime" in available}
 
 
 @app.get("/api/personas")