route stuff through dispatch no direct main

litellm/__init__.py

@@ -1405,10 +1405,22 @@ from .images.main import *
 from .videos.main import *
 from .batch_completion.main import *
 from .rerank_api.main import *
-from .llms.anthropic.experimental_pass_through.messages.handler import *
 from .messages.dispatch import *
-from .responses.main import *
 from .responses.dispatch import *
+from .responses.main import (
+    acancel_responses,
+    acompact_responses,
+    adelete_responses,
+    aget_responses,
+    alist_input_items,
+    aresponses_api_with_mcp,
+    cancel_responses,
+    compact_responses,
+    delete_responses,
+    get_responses,
+    list_input_items,
+    mock_responses_api_response,
+)
 
 # Interactions API is available as litellm.interactions module
 # Usage: litellm.interactions.create(), litellm.interactions.get(), etc.

litellm/anthropic_interface/messages/__init__.py

@@ -13,10 +13,10 @@ This is an __init__.py file to allow the following interface
 from collections.abc import AsyncIterator, Coroutine, Iterator
 from typing import Any
 
-from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+from litellm.messages import (
     anthropic_messages as _async_anthropic_messages,
 )
-from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+from litellm.messages import (
     anthropic_messages_handler as _sync_anthropic_messages,
 )
 from litellm.types.llms.anthropic_messages.anthropic_response import (

litellm/chat_completions/dispatch.py

@@ -31,13 +31,15 @@ PythonAcompletion: TypeAlias = Callable[..., Awaitable[ChatResult]]
 
 def _python_completion() -> PythonCompletion:
     return cast(  # cast-ok: forward the original call shape through the Python @client decorator
-        PythonCompletion, main.completion
+        PythonCompletion,
+        main.completion,  # noqa: TID251  # dispatch boundary owns this Python fallback
     )
 
 
 def _python_acompletion() -> PythonAcompletion:
     return cast(  # cast-ok: forward the original call shape through the Python @client decorator
-        PythonAcompletion, main.acompletion
+        PythonAcompletion,
+        main.acompletion,  # noqa: TID251  # dispatch boundary owns this Python fallback
     )
 
 

litellm/llms/anthropic/experimental_pass_through/messages/handler.py

@@ -40,6 +40,8 @@ from ..utils import is_reasoning_auto_summary_enabled
 from .interceptors import get_messages_interceptors
 from .utils import AnthropicMessagesRequestUtils, mock_response
 
+__all__ = ("anthropic_messages", "anthropic_messages_handler")
+
 # Providers that are routed directly to the OpenAI Responses API instead of
 # going through chat/completions.
 _RESPONSES_API_PROVIDERS: Final = frozenset({"openai"})

litellm/llms/anthropic/experimental_pass_through/messages/interceptors/advisor.py

@@ -414,9 +414,7 @@ async def _call_messages_handler(
     Using the public function (decorated with @client) ensures logging, retries,
     and provider resolution all work correctly, identical to a direct user call.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
-        anthropic_messages,
-    )
+    from litellm.messages import anthropic_messages
 
     return await anthropic_messages(
         model=model,

litellm/main.py

@@ -5968,7 +5968,7 @@ def responses_with_retries(*args, **kwargs):
     except Exception as e:
         raise Exception(f"tenacity import failed please run `pip install tenacity`. Error{e}")
 
-    from litellm.responses.main import responses
+    from litellm.responses.dispatch import responses
 
     num_retries: Final = kwargs.pop("num_retries", 3)
     # reset retries in .responses()
@@ -5998,7 +5998,7 @@ async def aresponses_with_retries(*args, **kwargs):
     except Exception as e:
         raise Exception(f"tenacity import failed please run `pip install tenacity`. Error{e}")
 
-    from litellm.responses.main import aresponses
+    from litellm.responses.dispatch import aresponses
 
     num_retries: Final = kwargs.pop("num_retries", 3)
     kwargs["max_retries"] = 0

litellm/messages/dispatch.py

@@ -30,13 +30,15 @@ PythonAmessages: TypeAlias = Callable[..., Awaitable[MessagesResult]]
 
 def _python_messages() -> PythonMessages:
     return cast(  # cast-ok: forward the original call shape through the legacy handler
-        PythonMessages, main.anthropic_messages_handler
+        PythonMessages,
+        main.anthropic_messages_handler,  # noqa: TID251  # dispatch boundary owns this Python fallback
     )
 
 
 def _python_amessages() -> PythonAmessages:
     return cast(  # cast-ok: forward the original call shape through the Python @client decorator
-        PythonAmessages, main.anthropic_messages
+        PythonAmessages,
+        main.anthropic_messages,  # noqa: TID251  # dispatch boundary owns this Python fallback
     )
 
 

litellm/ocr/dispatch.py

@@ -43,10 +43,12 @@ def _public_request(name: str, args: tuple[object, ...], kwargs: Mapping[str, ob
 
 
 _PYTHON_OCR: Final = cast(  # cast-ok: forward the original call shape through the Python @client decorator
-    Callable[..., OCRResponse | Coroutine[object, object, OCRResponse]], main.ocr
+    Callable[..., OCRResponse | Coroutine[object, object, OCRResponse]],
+    main.ocr,  # noqa: TID251  # dispatch boundary owns this Python fallback
 )
 _PYTHON_AOCR: Final = cast(  # cast-ok: forward the original call shape through the Python @client decorator
-    Callable[..., Awaitable[OCRResponse]], main.aocr
+    Callable[..., Awaitable[OCRResponse]],
+    main.aocr,  # noqa: TID251  # dispatch boundary owns this Python fallback
 )
 
 

litellm/responses/dispatch.py

@@ -24,13 +24,15 @@ PythonAresponses: TypeAlias = Callable[..., Awaitable[ResponsesResult]]
 
 def _python_responses() -> PythonResponses:
     return cast(  # cast-ok: forward the original call shape through the Python @client decorator
-        PythonResponses, main.responses
+        PythonResponses,
+        main.responses,  # noqa: TID251  # dispatch boundary owns this Python fallback
     )
 
 
 def _python_aresponses() -> PythonAresponses:
     return cast(  # cast-ok: forward the original call shape through the Python @client decorator
-        PythonAresponses, main.aresponses
+        PythonAresponses,
+        main.aresponses,  # noqa: TID251  # dispatch boundary owns this Python fallback
     )
 
 

litellm/responses/file_search/emulated_handler.py

@@ -390,7 +390,7 @@ def _synthesize_responses_api_response(
 
 
 async def _call_aresponses(input, model, tools, **kwargs):  # pragma: no cover – thin wrapper for patching in tests
-    from litellm.responses.main import aresponses
+    from litellm.responses.main import aresponses  # noqa: TID251  # inner call must not re-enter file-search emulation
 
     return await aresponses(input=input, model=model, tools=tools, **kwargs)
 

litellm/responses/main.py

@@ -67,6 +67,23 @@ else:
 
 from .streaming_iterator import BaseResponsesAPIStreamingIterator
 
+__all__ = (
+    "acancel_responses",
+    "acompact_responses",
+    "adelete_responses",
+    "aget_responses",
+    "alist_input_items",
+    "aresponses",
+    "aresponses_api_with_mcp",
+    "cancel_responses",
+    "compact_responses",
+    "delete_responses",
+    "get_responses",
+    "list_input_items",
+    "mock_responses_api_response",
+    "responses",
+)
+
 ####### ENVIRONMENT VARIABLES ###################
 # Initialize any necessary instances or variables here
 base_llm_http_handler = BaseLLMHTTPHandler()

litellm/responses/mcp/litellm_proxy_mcp_handler.py

@@ -16,7 +16,7 @@ from litellm.proxy._experimental.mcp_server.utils import (
     split_server_prefix_from_name,
     strip_known_server_prefix,
 )
-from litellm.responses.main import aresponses
+from litellm.responses.main import aresponses  # noqa: TID251  # inner call must skip the MCP gateway that invoked it
 from litellm.responses.streaming_iterator import BaseResponsesAPIStreamingIterator
 from litellm.types.llms.openai import (
     ResponseInputParam,

litellm/responses/mcp/mcp_streaming_iterator.py

@@ -609,7 +609,7 @@ class MCPEnhancedStreamingIterator(BaseResponsesAPIStreamingIterator):
         """Create the initial response iterator by making the first LLM call"""
         try:
             # Import the core aresponses function that doesn't have MCP logic
-            from litellm.responses.main import aresponses
+            from litellm.responses.main import aresponses  # noqa: TID251  # core call without MCP logic
 
             # Make the initial response API call - but avoid the MCP wrapper
             params: Final[dict[str, object]] = self.original_request_params.copy()
@@ -773,7 +773,7 @@ class MCPEnhancedStreamingIterator(BaseResponsesAPIStreamingIterator):
             self.base_iterator = None
             return
 
-        from litellm.responses.main import aresponses
+        from litellm.responses.main import aresponses  # noqa: TID251  # follow-up call without MCP logic
         from litellm.responses.mcp.litellm_proxy_mcp_handler import (
             LiteLLM_Proxy_MCP_Handler,
         )

ruff-strict.toml

@@ -57,3 +57,15 @@ max-args = 5
 "typing_extensions.TypeGuard".msg = "Same as typing.TypeGuard."
 "typing.TypeIs".msg = "Unverified narrowing (the body is trusted). Parse into a concrete type instead."
 "typing_extensions.TypeIs".msg = "Same as typing.TypeIs."
+# Dispatched public entry points: import them from their dispatch module so every
+# supported call path selects Rust or Python in one place. Only the dispatch
+# modules and internal recursive calls may reach the Python implementation
+# directly, each with a `# noqa: TID251  # <reason>`.
+"litellm.responses.main.responses".msg = "Import litellm.responses.dispatch.responses so the call routes through dispatch."
+"litellm.responses.main.aresponses".msg = "Import litellm.responses.dispatch.aresponses so the call routes through dispatch."
+"litellm.llms.anthropic.experimental_pass_through.messages.handler.anthropic_messages".msg = "Import litellm.messages.anthropic_messages so the call routes through dispatch."
+"litellm.llms.anthropic.experimental_pass_through.messages.handler.anthropic_messages_handler".msg = "Import litellm.messages.anthropic_messages_handler so the call routes through dispatch."
+"litellm.ocr.main.ocr".msg = "Import litellm.ocr.dispatch.ocr so the call routes through dispatch."
+"litellm.ocr.main.aocr".msg = "Import litellm.ocr.dispatch.aocr so the call routes through dispatch."
+"litellm.main.completion".msg = "Import litellm.completion so the call routes through dispatch."
+"litellm.main.acompletion".msg = "Import litellm.acompletion so the call routes through dispatch."

tests/test_litellm/chat_completions/test_dispatch.py

@@ -1,5 +1,5 @@
 import inspect
-from collections.abc import Callable, Mapping
+from collections.abc import Awaitable, Callable, Mapping
 from typing import Final, cast  # noqa: TID251  # narrows legacy callable signatures for inspect
 
 import pytest
@@ -10,9 +10,12 @@ from litellm.chat_completions.dispatch import (
     _ADISPATCH,  # pyright: ignore[reportPrivateUsage]  # tests configured dispatch
     _DISPATCH,  # pyright: ignore[reportPrivateUsage]  # tests configured dispatch
 )
+from litellm.rust_bridge import catalog
 from litellm.rust_bridge.bindings import NativeBinding
 from litellm.rust_bridge.catalog import Route, Rule
 from litellm.rust_bridge.chat_completions.entrypoints import (
+    NATIVE_ACOMPLETION,
+    NATIVE_COMPLETION,
     LiteLLMChatCompletionsRequest,
     NativeAcompletion,
     NativeCompletion,
@@ -219,3 +222,50 @@ def test_binding_errors_delegate_to_python(args: tuple[object, ...], kwargs: Map
         is response
     )
     assert captured == [(args, kwargs)]
+
+
+def test_public_completion_routes_through_dispatch(monkeypatch: pytest.MonkeyPatch) -> None:
+    captured: Final[list[LiteLLMChatCompletionsRequest]] = []
+    expected: Final = ModelResponse()
+
+    def native(
+        request: LiteLLMChatCompletionsRequest,
+        args: tuple[object, ...],
+        kwargs: Mapping[str, object],
+    ) -> ModelResponse:
+        captured.append(request)
+        return expected
+
+    NATIVE_COMPLETION.override(native)
+    monkeypatch.setattr(catalog, "RULES", RUST_RULES)
+    public_completion: Final = cast(Callable[..., ModelResponse], litellm.completion)
+    try:
+        result: Final = public_completion(model="gpt-4o", messages=MESSAGES)
+    finally:
+        NATIVE_COMPLETION.reset()
+    assert result is expected
+    assert [request.model for request in captured] == ["gpt-4o"]
+
+
+@pytest.mark.asyncio
+async def test_public_acompletion_routes_through_dispatch(monkeypatch: pytest.MonkeyPatch) -> None:
+    captured: Final[list[LiteLLMChatCompletionsRequest]] = []
+    expected: Final = ModelResponse()
+
+    async def native(
+        request: LiteLLMChatCompletionsRequest,
+        args: tuple[object, ...],
+        kwargs: Mapping[str, object],
+    ) -> ModelResponse:
+        captured.append(request)
+        return expected
+
+    NATIVE_ACOMPLETION.override(native)
+    monkeypatch.setattr(catalog, "RULES", RUST_RULES)
+    public_acompletion: Final = cast(Callable[..., Awaitable[ModelResponse]], litellm.acompletion)
+    try:
+        result: Final = await public_acompletion(model="gpt-4o", messages=MESSAGES)
+    finally:
+        NATIVE_ACOMPLETION.reset()
+    assert result is expected
+    assert [request.model for request in captured] == ["gpt-4o"]

tests/test_litellm/messages/test_dispatch.py

@@ -1,5 +1,5 @@
 import inspect
-from collections.abc import Callable, Mapping
+from collections.abc import Awaitable, Callable, Mapping
 from typing import Final, cast  # noqa: TID251  # narrows legacy callable signatures for inspect
 
 import pytest
@@ -10,10 +10,13 @@ from litellm.messages.dispatch import (
     _ADISPATCH,  # pyright: ignore[reportPrivateUsage]  # tests configured dispatch
     _DISPATCH,  # pyright: ignore[reportPrivateUsage]  # tests configured dispatch
 )
+from litellm.rust_bridge import catalog
 from litellm.rust_bridge.bindings import NativeBinding
 from litellm.rust_bridge.catalog import Route, Rule, Rules
 from litellm.rust_bridge.configuration import Rollout
 from litellm.rust_bridge.messages.entrypoints import (
+    NATIVE_AMESSAGES,
+    NATIVE_MESSAGES,
     LiteLLMMessagesRequest,
     NativeAmessages,
     NativeMessages,
@@ -235,3 +238,50 @@ def test_binding_errors_delegate_to_python(args: tuple[object, ...], kwargs: Map
     )
     assert result is expected
     assert captured == [(args, kwargs)]
+
+
+def test_anthropic_create_routes_through_dispatch(monkeypatch: pytest.MonkeyPatch) -> None:
+    captured: Final[list[LiteLLMMessagesRequest]] = []
+    expected: Final = response()
+
+    def native(
+        request: LiteLLMMessagesRequest,
+        args: tuple[object, ...],
+        kwargs: Mapping[str, object],
+    ) -> AnthropicMessagesResponse:
+        captured.append(request)
+        return expected
+
+    NATIVE_MESSAGES.override(native)
+    monkeypatch.setattr(catalog, "RULES", RUST_RULES)
+    public_create: Final = cast(Callable[..., AnthropicMessagesResponse], litellm.anthropic.create)
+    try:
+        result: Final = public_create(max_tokens=16, messages=MESSAGES, model="claude-sonnet-4-5")
+    finally:
+        NATIVE_MESSAGES.reset()
+    assert result is expected
+    assert [request.model for request in captured] == ["claude-sonnet-4-5"]
+
+
+@pytest.mark.asyncio
+async def test_anthropic_acreate_routes_through_dispatch(monkeypatch: pytest.MonkeyPatch) -> None:
+    captured: Final[list[LiteLLMMessagesRequest]] = []
+    expected: Final = response()
+
+    async def native(
+        request: LiteLLMMessagesRequest,
+        args: tuple[object, ...],
+        kwargs: Mapping[str, object],
+    ) -> AnthropicMessagesResponse:
+        captured.append(request)
+        return expected
+
+    NATIVE_AMESSAGES.override(native)
+    monkeypatch.setattr(catalog, "RULES", RUST_RULES)
+    public_acreate: Final = cast(Callable[..., Awaitable[AnthropicMessagesResponse]], litellm.anthropic.acreate)
+    try:
+        result: Final = await public_acreate(max_tokens=16, messages=MESSAGES, model="claude-sonnet-4-5")
+    finally:
+        NATIVE_AMESSAGES.reset()
+    assert result is expected
+    assert [request.model for request in captured] == ["claude-sonnet-4-5"]

tests/test_litellm/ocr/test_dispatch.py

@@ -1,18 +1,26 @@
-from collections.abc import Mapping
-from typing import Final
+from collections.abc import Awaitable, Callable, Mapping
+from typing import Final, cast  # noqa: TID251  # narrows legacy callable signatures for inspect
 
 import httpx
 import pytest
 
+import litellm
 from litellm.llms.base_llm.ocr.transformation import OCRResponse
 from litellm.ocr.dispatch import (
     _ADISPATCH,  # pyright: ignore[reportPrivateUsage]  # tests configured dispatch
     _DISPATCH,  # pyright: ignore[reportPrivateUsage]  # tests configured dispatch
 )
+from litellm.rust_bridge import catalog
 from litellm.rust_bridge.bindings import NativeBinding
 from litellm.rust_bridge.catalog import Route, Rule, Rules
 from litellm.rust_bridge.configuration import Rollout
-from litellm.rust_bridge.ocr.entrypoints import LiteLLMOcrRequest, NativeAocr, NativeOcr
+from litellm.rust_bridge.ocr.entrypoints import (
+    NATIVE_AOCR,
+    NATIVE_OCR,
+    LiteLLMOcrRequest,
+    NativeAocr,
+    NativeOcr,
+)
 
 PYTHON_RULES: Final[Rules] = (Rule(Route.OCR, Rollout.PYTHON_ONLY),)
 RUST_RULES: Final[Rules] = (Rule(Route.OCR, Rollout.RUST_REQUIRED),)
@@ -324,3 +332,58 @@ async def test_aocr_parser_errors_before_python_or_native(
             native=lambda hook, request, call_args, call_kwargs: hook(request, call_args, call_kwargs),
             rules=RUST_RULES,
         )
+
+
+def test_public_ocr_routes_through_dispatch(monkeypatch: pytest.MonkeyPatch) -> None:
+    document: Final[Mapping[str, object]] = {
+        "type": "document_url",
+        "document_url": "https://example.invalid/document.pdf",
+    }
+    captured: Final[list[LiteLLMOcrRequest]] = []
+    expected: Final = response()
+
+    def native(
+        request: LiteLLMOcrRequest,
+        args: tuple[object, ...],
+        kwargs: Mapping[str, object],
+    ) -> OCRResponse:
+        captured.append(request)
+        return expected
+
+    NATIVE_OCR.override(native)
+    monkeypatch.setattr(catalog, "RULES", RUST_RULES)
+    public_ocr: Final = cast(Callable[..., OCRResponse], litellm.ocr)
+    try:
+        result: Final = public_ocr(model="mistral/mistral-ocr-latest", document=document)
+    finally:
+        NATIVE_OCR.reset()
+    assert result is expected
+    assert [request.model for request in captured] == ["mistral/mistral-ocr-latest"]
+
+
+@pytest.mark.asyncio
+async def test_public_aocr_routes_through_dispatch(monkeypatch: pytest.MonkeyPatch) -> None:
+    document: Final[Mapping[str, object]] = {
+        "type": "document_url",
+        "document_url": "https://example.invalid/document.pdf",
+    }
+    captured: Final[list[LiteLLMOcrRequest]] = []
+    expected: Final = response()
+
+    async def native(
+        request: LiteLLMOcrRequest,
+        args: tuple[object, ...],
+        kwargs: Mapping[str, object],
+    ) -> OCRResponse:
+        captured.append(request)
+        return expected
+
+    NATIVE_AOCR.override(native)
+    monkeypatch.setattr(catalog, "RULES", RUST_RULES)
+    public_aocr: Final = cast(Callable[..., Awaitable[OCRResponse]], litellm.aocr)
+    try:
+        result: Final = await public_aocr(model="mistral/mistral-ocr-latest", document=document)
+    finally:
+        NATIVE_AOCR.reset()
+    assert result is expected
+    assert [request.model for request in captured] == ["mistral/mistral-ocr-latest"]

tests/test_litellm/responses/test_dispatch.py

@@ -1,19 +1,23 @@
 import inspect
-from collections.abc import Callable, Mapping
+from collections.abc import Awaitable, Callable, Mapping
 from typing import Final, cast  # noqa: TID251  # narrows legacy callable signatures for inspect
 
 import pytest
 
 import litellm
+from litellm.responses import dispatch as responses_dispatch
 from litellm.responses import main as python_responses
 from litellm.responses.dispatch import (
     _ADISPATCH,  # pyright: ignore[reportPrivateUsage]  # tests configured dispatch
     _DISPATCH,  # pyright: ignore[reportPrivateUsage]  # tests configured dispatch
 )
+from litellm.rust_bridge import catalog
 from litellm.rust_bridge.bindings import NativeBinding
 from litellm.rust_bridge.catalog import Route, Rule
 from litellm.rust_bridge.configuration import Rollout
 from litellm.rust_bridge.responses.entrypoints import (
+    NATIVE_ARESPONSES,
+    NATIVE_RESPONSES,
     LiteLLMResponsesRequest,
     NativeAresponses,
     NativeResponses,
@@ -253,3 +257,66 @@ def test_binding_errors_delegate_unchanged_to_python(
         is response
     )
     assert captured == [(args, kwargs)]
+
+
+def test_public_responses_routes_through_dispatch(monkeypatch: pytest.MonkeyPatch) -> None:
+    captured: Final[list[LiteLLMResponsesRequest]] = []
+    expected: Final = _response()
+
+    def native(
+        request: LiteLLMResponsesRequest,
+        args: tuple[object, ...],
+        kwargs: Mapping[str, object],
+    ) -> ResponsesAPIResponse:
+        captured.append(request)
+        return expected
+
+    NATIVE_RESPONSES.override(native)
+    monkeypatch.setattr(catalog, "RULES", RUST_RULES)
+    public_responses: Final = cast(Callable[..., ResponsesAPIResponse], litellm.responses)
+    try:
+        result: Final = public_responses(input=INPUT, model="gpt-4o")
+    finally:
+        NATIVE_RESPONSES.reset()
+    assert result is expected
+    assert [request.model for request in captured] == ["gpt-4o"]
+
+
+@pytest.mark.asyncio
+async def test_public_aresponses_routes_through_dispatch(monkeypatch: pytest.MonkeyPatch) -> None:
+    captured: Final[list[LiteLLMResponsesRequest]] = []
+    expected: Final = _response()
+
+    async def native(
+        request: LiteLLMResponsesRequest,
+        args: tuple[object, ...],
+        kwargs: Mapping[str, object],
+    ) -> ResponsesAPIResponse:
+        captured.append(request)
+        return expected
+
+    NATIVE_ARESPONSES.override(native)
+    monkeypatch.setattr(catalog, "RULES", RUST_RULES)
+    public_aresponses: Final = cast(Callable[..., Awaitable[ResponsesAPIResponse]], litellm.aresponses)
+    try:
+        result: Final = await public_aresponses(input=INPUT, model="gpt-4o")
+    finally:
+        NATIVE_ARESPONSES.reset()
+    assert result is expected
+    assert [request.model for request in captured] == ["gpt-4o"]
+
+
+def test_responses_with_retries_uses_the_dispatch_entrypoint(monkeypatch: pytest.MonkeyPatch) -> None:
+    calls: Final[list[Mapping[str, object]]] = []
+    expected: Final = _response()
+
+    def dispatch_responses(*args: object, **kwargs: object) -> ResponsesAPIResponse:  # kwargs-ok: records call shape
+        calls.append(kwargs)
+        return expected
+
+    monkeypatch.setattr(responses_dispatch, "responses", dispatch_responses)
+    retry: Final = cast(Callable[..., ResponsesAPIResponse], litellm.responses_with_retries)
+    result: Final = retry(input=INPUT, model="gpt-4o", num_retries=1)
+    assert result is expected
+    assert calls[0]["num_retries"] == 0
+    assert calls[0]["max_retries"] == 0