feat: refactor ChatSessionManager to be application-specific and integrate httpx2 for websocket transport

app/api/chat.py

@@ -100,7 +100,7 @@ async def chat_websocket(websocket: WebSocket) -> None:
     await websocket.accept()
 
     agent: ChatAgent = websocket.app.chat_agent
-    session_manager: ChatSessionManager = ChatSessionManager()
+    session_manager: ChatSessionManager = websocket.app.state.chat_sessions
     session = await session_manager.create()
 
     await websocket.send_json(Connected(session_id=session.id).model_dump(mode="json"))

app/main.py

@@ -22,6 +22,7 @@ from app.middleware.profiler import ProfilingMiddleware
 from app.redis import get_redis
 from app.services.auth import AuthBearer
 from app.services.chat_agent import build_chat_agent
+from app.services.chat_session import ChatSessionManager
 from app.services.scheduler import scheduler_lifespan
 
 templates = Jinja2Templates(directory=Path(__file__).parent.parent / "templates")
@@ -32,11 +33,9 @@ async def lifespan(app: FastAPI):
     app.logger = get_logger()
     app.redis = await get_redis()
     global_settings.postgres_url.unicode_string()
-    # Chat service: initialize the pluggable model-client adapter. The session
-    # manager is now a singleton and will be auto-instantiated on first access.
-    # See app/services/chat_agent.py to swap the local stub for a real model
-    # client (OpenAI, Ollama, etc.).
+    # Initialize the chat agent and per-application session registry.
     app.chat_agent = build_chat_agent(global_settings.chat)
+    app.state.chat_sessions = ChatSessionManager()
     try:
         # app.postgres_pool = await asyncpg.create_pool(
         #     dsn=postgres_dsn,

app/services/chat_session.py

@@ -13,7 +13,6 @@ from uuid import UUID, uuid4
 import attrs
 
 from app.schemas.chat import ChatMessage
-from app.utils.singleton import SingletonMetaNoArgs
 
 
 @attrs.define(slots=True)
@@ -28,11 +27,11 @@ class ChatSession:
         return list(self.messages)
 
 
-class ChatSessionManager(metaclass=SingletonMetaNoArgs):
+class ChatSessionManager:
     """Tracks active chat sessions keyed by their opaque session id.
 
-    Implemented as a singleton to ensure exactly one instance per application,
-    maintaining a consistent registry of all active websocket chat sessions.
+    Each application lifespan owns its own manager, maintaining a consistent
+    registry of active websocket chat sessions for that application.
     """
 
     def __init__(self) -> None: