fix(rust-bridge): keep proxy /v1/messages calls on Rust when a cache or shared session is configured

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/messages/dispatch.py

@@ -104,7 +104,6 @@ _FEATURES: Final[detectors.Detectors] = MappingProxyType(
     {
         **detectors.COMMON,
         Feature.SAFEGUARDS_CLASSIFIER: route_detectors.safeguards_classifier,
-        Feature.RESPONSE_CACHE: detectors.response_cache,
         Feature.MODIFY_PARAMS: detectors.modify_params,
         Feature.WORKLOAD_IDENTITY: route_detectors.anthropic_workload_identity,
         Feature.CACHE_CONTROL_INJECTION: detectors.cache_control_injection,

litellm/rust_bridge/catalog.py

@@ -169,7 +169,6 @@ MESSAGES_PYTHON_FEATURES: Final = COMMON_PYTHON_FEATURES | {
     Feature.PRE_REQUEST_HOOK,
     Feature.WEBSEARCH_INTERCEPTION,
     Feature.MESSAGES_INTERCEPTOR,
-    Feature.RESPONSE_CACHE,
     Feature.MODIFY_PARAMS,
     Feature.WORKLOAD_IDENTITY,
     Feature.CACHE_CONTROL_INJECTION,
@@ -208,7 +207,7 @@ RULES: Final[Rules] = (
     RouteRule(Route.OCR, Rollout.RUST_REQUIRED),
     RouteRule(Route.MESSAGES, Rollout.PYTHON_ONLY, targets=frozenset({Target.CHAT_COMPLETIONS, Target.RESPONSES})),
     RouteRule(Route.MESSAGES, Rollout.PYTHON_ONLY, features=MESSAGES_PYTHON_FEATURES),
-    RouteRule(Route.MESSAGES, Rollout.RUST_OPT_IN, providers=frozenset({"anthropic"})),
+    RouteRule(Route.MESSAGES, Rollout.RUST_OPT_IN, providers=frozenset({"anthropic", "deepseek", "vertex_ai"})),
     RouteRule(Route.MESSAGES, Rollout.PYTHON_ONLY),
     RouteRule(Route.RESPONSES, Rollout.PYTHON_ONLY, features=RESPONSES_PYTHON_FEATURES),
     RouteRule(Route.RESPONSES, Rollout.PYTHON_ONLY),

litellm/rust_bridge/detectors.py

@@ -36,8 +36,8 @@ def mock_response(request: NativeCall) -> bool:
 
 
 def caller_client(request: NativeCall) -> bool:
-    """Python sends through the caller's HTTP client or shared session."""
-    return present(request, "client") or present(request, "shared_session")
+    """Python sends through the caller's HTTP client. The proxy's shared session is only a pool native replaces."""
+    return present(request, "client")
 
 
 def extra_body(request: NativeCall) -> bool:

tests/test_litellm_rust/messages/test_dispatch.py

@@ -16,7 +16,7 @@ from litellm.messages.dispatch import (
 )
 from litellm.proxy import proxy_server
 from litellm.rust_bridge.bindings import NativeBinding
-from litellm.rust_bridge.catalog import Feature, Route, RouteRule, Rules, Target
+from litellm.rust_bridge.catalog import MESSAGES_PYTHON_FEATURES, Feature, Route, RouteRule, Rules, Target
 from litellm.rust_bridge.configuration import Rollout
 from litellm.rust_bridge.messages.entrypoints import NATIVE_AMESSAGES, NATIVE_MESSAGES, NativeAmessages, NativeMessages
 from litellm.rust_bridge.public_call import NativeCall
@@ -228,6 +228,25 @@ def test_only_requests_that_pass_through_to_a_messages_api_reach_a_messages_targ
     assert result is (native_result if passes_through else python_result)
 
 
+def test_proxy_cache_and_shared_session_keep_messages_native(monkeypatch: pytest.MonkeyPatch) -> None:
+    model: Final = "anthropic/claude-sonnet-4-5"
+    python_result: Final = response(model)
+    native_result: Final = response(model)
+    monkeypatch.setattr(litellm, "cache", Cache())
+    result: Final = _DISPATCH.run(
+        (16, MESSAGES, model),
+        {"shared_session": object()},
+        python=lambda *call_args, **call_kwargs: python_result,
+        binding=messages_binding(lambda request: native_result),
+        native=lambda hook, request, call_args, call_kwargs: hook(request),
+        rules=(
+            RouteRule(Route.MESSAGES, Rollout.PYTHON_ONLY, features=MESSAGES_PYTHON_FEATURES),
+            RouteRule(Route.MESSAGES, Rollout.RUST_REQUIRED),
+        ),
+    )
+    assert result is native_result
+
+
 DANGEROUS_TOOL_USE: Final = [{"type": "dangerous_tool_use", "classifier_context": {"permission_mode": "auto"}}]
 CLASSIFIER_SETTINGS: Final = {"safeguards_classifier_model": "gpt-oss-120b"}
 
@@ -457,7 +476,6 @@ ADVISOR_TOOL: Final = {"type": ANTHROPIC_ADVISOR_TOOL_TYPE, "name": "advisor", "
         (Feature.CALLER_CLIENT, "anthropic/claude-sonnet-4-5", {"client": object()}, {}),
         (Feature.TIMEOUT_VARIANTS, "anthropic/claude-sonnet-4-5", {"request_timeout": 5}, {}),
         (Feature.AGENTIC_LOOP_HOOK, "anthropic/claude-sonnet-4-5", {}, {"callbacks": [_Agentic()]}),
-        (Feature.RESPONSE_CACHE, "anthropic/claude-sonnet-4-5", {}, {"cache": Cache()}),
         (Feature.TLS_POLICY, "anthropic/claude-sonnet-4-5", {}, {"ssl_security_level": "DEFAULT@SECLEVEL=1"}),
         (Feature.MODIFY_PARAMS, "anthropic/claude-sonnet-4-5", {}, {"modify_params": True}),
         (

tests/test_litellm_rust/responses/test_dispatch.py

@@ -260,7 +260,6 @@ ENCODED_ID: Final = ResponsesAPIRequestUtils._build_encrypted_item_id("deploymen
         (Feature.MOCK_RESPONSE, {"mock_response": "hi"}, {}),
         (Feature.PROXY_MCP_TOOLS, {"tools": [{"type": "mcp", "server_label": "g", "server_url": "litellm_proxy"}]}, {}),
         (Feature.CALLER_CLIENT, {"client": object()}, {}),
-        pytest.param(Feature.CALLER_CLIENT, {"shared_session": object()}, {}, id="shared-session"),
         pytest.param(Feature.EXTRA_BODY, {"extra_body": {"stream": True}}, {}, id="extra-body-stream"),
         pytest.param(
             Feature.PROVIDER_AFFINITY,

tests/unit/rust_bridge/test_catalog.py

@@ -128,3 +128,9 @@ def test_empty_constraints_match_nothing() -> None:
     assert catalog.decision(
         RouteContext(Route.OCR, features=frozenset({Feature.SAFEGUARDS_CLASSIFIER})), rules
     ) is Decision.PYTHON
+
+
+@pytest.mark.parametrize("provider", ("anthropic", "deepseek", "vertex_ai"))
+def test_messages_passthrough_providers_opt_in_to_rust(provider: str) -> None:
+    context: Final = RouteContext(Route.MESSAGES, provider=provider, model="claude-sonnet-4-6", target=Target.MESSAGES)
+    assert catalog.rollout(context) is Rollout.RUST_OPT_IN

tests/unit/rust_bridge/test_detectors.py

@@ -92,7 +92,7 @@ def _wrap_nested_level(value: object, depth: int) -> object:
         (detectors.prompt_management, {"model": "litellm_agent/gpt-5"}, True),
         (detectors.prompt_management, {"model": "gpt-5"}, False),
         (detectors.caller_client, {"client": object()}, True),
-        (detectors.caller_client, {"shared_session": object()}, True),
+        (detectors.caller_client, {"shared_session": object()}, False),
         (detectors.caller_client, {"client": None, "shared_session": None}, False),
         (detectors.caller_client, {}, False),
         (detectors.extra_body, {"extra_body": {"stream": True}}, True),