diff --git a/backend/app/api/tickets.py b/backend/app/api/tickets.py
index bb4f2f7..db470ec 100644
--- a/backend/app/api/tickets.py
+++ b/backend/app/api/tickets.py
@@ -6,6 +6,8 @@ Returns data in a format aligned with the frontend TicketData interface:
 
 from __future__ import annotations
 
+from collections import Counter
+
 import httpx
 from litestar import get
 from litestar.exceptions import NotFoundException
@@ -51,6 +53,53 @@ def _transform_ticket(t: dict) -> dict:
     }
 
 
+@get("/api/tickets/summary")
+async def get_tickets_summary() -> dict:
+    """GET /api/tickets/summary -- Aggregate ticket statistics.
+
+    Returns:
+        {
+            "total": int,
+            "by_status":   {"pending": N, "processing": N, "resolved": N},
+            "by_priority": {"P0": N, "P1": N, "P2": N, "P3": N}
+        }
+    """
+    # Fetch a large page to get representative counts.
+    # For a production system this would be a dedicated aggregation API;
+    # the Gongdan API only exposes list endpoints so we aggregate client-side.
+    url = f"{settings.gongdan_api_base}/api/tickets"
+    params = {"page": 1, "pageSize": 100}
+
+    async with httpx.AsyncClient(timeout=15) as client:
+        resp = await client.get(url, params=params, headers=_gongdan_headers())
+        resp.raise_for_status()
+        data = resp.json()
+
+    tickets = data.get("tickets", [])
+
+    status_counter: Counter[str] = Counter()
+    priority_counter: Counter[str] = Counter()
+
+    for t in tickets:
+        status_counter[_map_status(t.get("status", ""))] += 1
+        priority_counter[_map_priority(t.get("priority", ""))] += 1
+
+    return {
+        "total": len(tickets),
+        "by_status": {
+            "pending": status_counter.get("pending", 0),
+            "processing": status_counter.get("processing", 0),
+            "resolved": status_counter.get("resolved", 0),
+        },
+        "by_priority": {
+            "P0": priority_counter.get("P0", 0),
+            "P1": priority_counter.get("P1", 0),
+            "P2": priority_counter.get("P2", 0),
+            "P3": priority_counter.get("P3", 0),
+        },
+    }
+
+
 @get("/api/tickets")
 async def list_tickets(page: int = 1, page_size: int = 20) -> list[dict]:
     """GET /api/tickets — List tickets from Gongdan, formatted for frontend."""
diff --git a/backend/app/main.py b/backend/app/main.py
index 8fc2f3b..4ee0dc0 100644
--- a/backend/app/main.py
+++ b/backend/app/main.py
@@ -22,10 +22,12 @@ from app.api.conversations import (
     update_conversation,
 )
 from app.api.health import health_check
-from app.api.tickets import get_ticket, list_tickets
+from app.api.tickets import get_ticket, get_tickets_summary, list_tickets
 from app.cache.redis import close_redis
+from app.storage.blob import close_blob_client
 from app.store.memory import close_checkpointer
 from app.store.postgres import create_tables, dispose_engine
+from app.tasks.bus import close_service_bus
 
 cors_config = CORSConfig(
     allow_origins=["*"],
@@ -42,6 +44,8 @@ async def lifespan(app: Litestar) -> AsyncGenerator[None, None]:
     yield
     await close_checkpointer()
     await close_redis()
+    await close_blob_client()
+    await close_service_bus()
     await dispose_engine()
 
 
@@ -54,6 +58,7 @@ app = Litestar(
         create_conversation,
         update_conversation,
         delete_conversation,
+        get_tickets_summary,
         list_tickets,
         get_ticket,
     ],
diff --git a/backend/app/tools/search.py b/backend/app/tools/search.py
index 97abe02..f99f2cd 100644
--- a/backend/app/tools/search.py
+++ b/backend/app/tools/search.py
@@ -13,6 +13,7 @@ All results are cached in Redis with TTL=300s.
 from __future__ import annotations
 
 import asyncio
+import contextvars
 import logging
 from typing import Any
 
@@ -196,16 +197,18 @@ async def _search_pro(query: str) -> str:
 
 # ── The LangChain tool exposed to the ReAct agent ──────────────────────
 
-# The model context is injected via the tool's config at call time.
-# We store a thread-local-like mapping so the tool knows which model
-# strategy to use. The chat handler sets this before invoking the graph.
-_current_model: str = "flash"
+# Use contextvars to safely pass the model strategy per-request in an
+# async concurrent environment.  Each asyncio Task (i.e. each SSE
+# request handler) gets its own copy, so concurrent requests never
+# overwrite each other's value.
+_current_model: contextvars.ContextVar[str] = contextvars.ContextVar(
+    "search_model", default="flash"
+)
 
 
 def set_search_model(model: str) -> None:
-    """Set the search strategy model for the current request."""
-    global _current_model
-    _current_model = model
+    """Set the search strategy model for the current request context."""
+    _current_model.set(model)
 
 
 @tool
@@ -219,7 +222,7 @@ async def web_search(query: str) -> str:
     Args:
         query: The search query in natural language.
     """
-    model = _current_model
+    model = _current_model.get()
 
     # Check Redis cache first
     cached = await get_cached_search(query, model)