diff --git a/src/llamacpp_ha/forwarder.py b/src/llamacpp_ha/forwarder.py
index 4bcdf55..df46564 100644
--- a/src/llamacpp_ha/forwarder.py
+++ b/src/llamacpp_ha/forwarder.py
@@ -126,11 +126,13 @@ async def forward_best_effort(
     request: Request,
     registry_backends: list[BackendState],
     session: aiohttp.ClientSession,
+    preferred_url: str | None = None,
 ) -> Response:
     """Forward without slot gating to any live backend (catch-all paths).
 
     Backends are tried in round-robin order; on connection failure the next
-    backend is attempted until all live backends are exhausted.
+    backend is attempted until all live backends are exhausted.  When
+    preferred_url is given (session affinity hint) that backend is tried first.
     """
     if not registry_backends:
         return Response(content="No live backends", status_code=503)
@@ -139,6 +141,11 @@ async def forward_best_effort(
     start = next(_best_effort_counter) % n
     ordered = registry_backends[start:] + registry_backends[:start]
 
+    if preferred_url:
+        preferred = [b for b in ordered if b.url == preferred_url]
+        rest = [b for b in ordered if b.url != preferred_url]
+        ordered = preferred + rest
+
     path = request.url.path
     query = request.url.query
     body = await request.body()
diff --git a/src/llamacpp_ha/proxy.py b/src/llamacpp_ha/proxy.py
index db3f6b5..4f29407 100644
--- a/src/llamacpp_ha/proxy.py
+++ b/src/llamacpp_ha/proxy.py
@@ -232,13 +232,19 @@ async def _catch_all(
     *,
     http: _HttpSession,
     registry: BackendRegistry,
-    stats: ProxyStats,
+    session_store: SessionStore,
 ) -> Response:
     if http.client is None:
         return Response(content="Proxy not ready", status_code=503)
-    stats.increment_requests()
+    # Non-inference paths are not counted as "requests served".
+    # If the client has a session cookie, prefer the backend that is already
+    # holding its KV-cache so web-UI API calls stay consistent with the chat.
+    session_id = _session_id_from(request)
+    preferred_url: str | None = None
+    if session_id:
+        preferred_url = await session_store.get_preferred_backend(session_id)
     live = registry.get_all_live_backends()
-    return await forward_best_effort(request, live, http.client)
+    return await forward_best_effort(request, live, http.client, preferred_url=preferred_url)
 
 
 async def _inference_endpoint(
@@ -394,7 +400,7 @@ def create_app(config: ProxyConfig) -> FastAPI:
         )
 
     async def catch_all_handler(request: Request, full_path: str) -> Response:  # noqa: ARG001
-        return await _catch_all(request, http=http, registry=registry, stats=stats)
+        return await _catch_all(request, http=http, registry=registry, session_store=session_store)
 
     app.add_api_route("/v1/models", list_models_handler, methods=["GET"])
     app.add_api_route("/health", health_handler, methods=["GET"])
diff --git a/tests/test_forwarder.py b/tests/test_forwarder.py
index 7f4fd5d..cfe1ec0 100644
--- a/tests/test_forwarder.py
+++ b/tests/test_forwarder.py
@@ -324,6 +324,45 @@ class TestForwardBestEffort(unittest.IsolatedAsyncioTestCase):
         response = await forward_best_effort(req, [state1, state2], session)
         self.assertEqual(response.status_code, 502)
 
+    async def test_preferred_url_tried_first(self):
+        """When preferred_url is given, that backend is attempted before others."""
+        state1 = _make_state("http://b1")
+        state2 = _make_state("http://b2")
+
+        calls: list[str] = []
+
+        def fake_iter(method, url, **kwargs):
+            calls.append(url.split("//")[1].split("/")[0])  # extract host
+            ctx, _ = _mock_aiohttp_response(status=200, body=b"ok")
+            return ctx
+
+        session = MagicMock()
+        session.request = MagicMock(side_effect=fake_iter)
+        req = _make_request({})
+
+        await forward_best_effort(req, [state1, state2], session, preferred_url="http://b2")
+        self.assertEqual(calls[0], "b2")
+
+    async def test_preferred_url_fallback_on_failure(self):
+        """If preferred backend fails, the next one is still tried."""
+        state1 = _make_state("http://b1")
+        state2 = _make_state("http://b2")
+
+        ctx_fail = MagicMock()
+        ctx_fail.__aenter__ = AsyncMock(side_effect=Exception("down"))
+        ctx_fail.__aexit__ = AsyncMock(return_value=False)
+        ctx_ok, _ = _mock_aiohttp_response(status=200, body=b"ok")
+
+        def pick(method, url, **kwargs):
+            return ctx_fail if "b2" in url else ctx_ok
+
+        session = MagicMock()
+        session.request = MagicMock(side_effect=pick)
+        req = _make_request({})
+
+        response = await forward_best_effort(req, [state1, state2], session, preferred_url="http://b2")
+        self.assertEqual(response.status_code, 200)
+
     async def test_status_code_and_body_preserved(self):
         state = _make_state("http://b1")
         ctx, _ = _mock_aiohttp_response(status=404, body=b"not found")