Python: [BREAKING] Refactor middleware layering and split Anthropic raw client (#4746)

* [BREAKING] Refactor middleware layering and raw clients

Reorder chat client layers so function invocation wraps chat middleware, and chat middleware stays outside telemetry while still running for each inner model call. Add middleware pipeline caching, refresh docs and samples, and split Anthropic into raw and public clients to match the standard layering model.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Tighten typing ignores in ancillary modules

Add targeted typing ignores in workflow visualization and lab modules so pyright stays clean alongside the middleware refactor work.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Fix categorize_middleware to unpack tuple/Sequence and use relative MRO assertions

- Broaden isinstance check in categorize_middleware from list to Sequence
  so tuples and other Sequence types are properly unpacked instead of
  being appended as a single item.
- Replace fragile hardcoded MRO index assertions in anthropic test with
  relative ordering via mro.index().
- Add regression tests for categorize_middleware with tuple, list, and
  None inputs.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Fix middleware string decomposition, add middleware param to FunctionInvocationLayer, and add tests (#4710)

- Guard categorize_middleware Sequence check against str/bytes to prevent
  character-by-character decomposition of accidentally passed strings
- Add explicit middleware parameter to FunctionInvocationLayer.get_response
  and merge it into client_kwargs before categorization, fixing the
  inconsistency where only OpenAIChatClient supported this parameter
- Add assertions that RawAnthropicClient does not inherit convenience layers
- Add chat middleware cache test with non-empty base middleware
- Add tests for single unwrapped middleware item and string input

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Apply pre-commit auto-fixes

* Apply pre-commit auto-fixes

* Address review feedback for #4710: review comment fixes

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <copilot@github.com>

Author: 3 people

python/packages/ag-ui/agent_framework_ag_ui/_client.py

@@ -111,8 +111,8 @@ def _map_update(update: ChatResponseUpdate) -> ChatResponseUpdate:
 
 @_apply_server_function_call_unwrap
 class AGUIChatClient(
-    ChatMiddlewareLayer[AGUIChatOptionsT],
     FunctionInvocationLayer[AGUIChatOptionsT],
+    ChatMiddlewareLayer[AGUIChatOptionsT],
     ChatTelemetryLayer[AGUIChatOptionsT],
     BaseChatClient[AGUIChatOptionsT],
     Generic[AGUIChatOptionsT],

python/packages/ag-ui/tests/ag_ui/conftest.py

@@ -45,16 +45,16 @@ def pytest_configure() -> None:
 
 
 class StreamingChatClientStub(
-    ChatMiddlewareLayer[OptionsCoT],
     FunctionInvocationLayer[OptionsCoT],
+    ChatMiddlewareLayer[OptionsCoT],
     ChatTelemetryLayer[OptionsCoT],
     BaseChatClient[OptionsCoT],
     Generic[OptionsCoT],
 ):
     """Typed streaming stub that satisfies SupportsChatGetResponse."""
 
     def __init__(self, stream_fn: StreamFn, response_fn: ResponseFn | None = None) -> None:
-        super().__init__(function_middleware=[])
+        super().__init__(middleware=[])
         self._stream_fn = stream_fn
         self._response_fn = response_fn
         self.last_session: AgentSession | None = None

python/packages/anthropic/agent_framework_anthropic/__init__.py

@@ -2,7 +2,7 @@
 
 import importlib.metadata
 
-from ._chat_client import AnthropicChatOptions, AnthropicClient
+from ._chat_client import AnthropicChatOptions, AnthropicClient, RawAnthropicClient
 
 try:
     __version__ = importlib.metadata.version(__name__)
@@ -12,5 +12,6 @@
 __all__ = [
     "AnthropicChatOptions",
     "AnthropicClient",
+    "RawAnthropicClient",
     "__version__",
 ]

python/packages/anthropic/agent_framework_anthropic/_chat_client.py

@@ -68,6 +68,7 @@
 __all__ = [
     "AnthropicChatOptions",
     "AnthropicClient",
+    "RawAnthropicClient",
     "ThinkingConfig",
 ]
 
@@ -210,14 +211,24 @@ class AnthropicSettings(TypedDict, total=False):
     chat_model_id: str | None
 
 
-class AnthropicClient(
-    ChatMiddlewareLayer[AnthropicOptionsT],
-    FunctionInvocationLayer[AnthropicOptionsT],
-    ChatTelemetryLayer[AnthropicOptionsT],
+class RawAnthropicClient(
     BaseChatClient[AnthropicOptionsT],
     Generic[AnthropicOptionsT],
 ):
-    """Anthropic Chat client with middleware, telemetry, and function invocation support."""
+    """Raw Anthropic chat client without middleware, telemetry, or function invocation support.
+
+    Warning:
+        **This class should not normally be used directly.** It does not include middleware,
+        telemetry, or function invocation support that you most likely need. If you do use it,
+        you should consider which additional layers to apply. There is a defined ordering that
+        you should follow:
+
+        1. **FunctionInvocationLayer** - Owns the tool/function calling loop and routes function middleware
+        2. **ChatMiddlewareLayer** - Applies chat middleware per model call and stays outside telemetry
+        3. **ChatTelemetryLayer** - Must stay inside chat middleware for correct per-call telemetry
+
+        Use ``AnthropicClient`` instead for a fully-featured client with all layers applied.
+    """
 
     OTEL_PROVIDER_NAME: ClassVar[str] = "anthropic"  # type: ignore[reportIncompatibleVariableOverride, misc]
 
@@ -229,12 +240,10 @@ def __init__(
         anthropic_client: AsyncAnthropic | None = None,
         additional_beta_flags: list[str] | None = None,
         additional_properties: dict[str, Any] | None = None,
-        middleware: Sequence[ChatAndFunctionMiddlewareTypes] | None = None,
-        function_invocation_configuration: FunctionInvocationConfiguration | None = None,
         env_file_path: str | None = None,
         env_file_encoding: str | None = None,
     ) -> None:
-        """Initialize an Anthropic Agent client.
+        """Initialize a raw Anthropic client.
 
         Keyword Args:
             api_key: The Anthropic API key to use for authentication.
@@ -245,37 +254,35 @@ def __init__(
             additional_beta_flags: Additional beta flags to enable on the client.
                 Default flags are: "mcp-client-2025-04-04", "code-execution-2025-08-25".
             additional_properties: Additional properties stored on the client instance.
-            middleware: Optional middleware to apply to the client.
-            function_invocation_configuration: Optional function invocation configuration override.
             env_file_path: Path to environment file for loading settings.
             env_file_encoding: Encoding of the environment file.
 
         Examples:
             .. code-block:: python
 
-                from agent_framework.anthropic import AnthropicClient
+                from agent_framework.anthropic import RawAnthropicClient
                 from azure.identity.aio import DefaultAzureCredential
 
                 # Using environment variables
                 # Set ANTHROPIC_API_KEY=your_anthropic_api_key
                 # ANTHROPIC_CHAT_MODEL_ID=claude-sonnet-4-5-20250929
 
                 # Or passing parameters directly
-                client = AnthropicClient(
+                client = RawAnthropicClient(
                     model_id="claude-sonnet-4-5-20250929",
                     api_key="your_anthropic_api_key",
                 )
 
                 # Or loading from a .env file
-                client = AnthropicClient(env_file_path="path/to/.env")
+                client = RawAnthropicClient(env_file_path="path/to/.env")
 
                 # Or passing in an existing client
                 from anthropic import AsyncAnthropic
 
                 anthropic_client = AsyncAnthropic(
                     api_key="your_anthropic_api_key", base_url="https://custom-anthropic-endpoint.com"
                 )
-                client = AnthropicClient(
+                client = RawAnthropicClient(
                     model_id="claude-sonnet-4-5-20250929",
                     anthropic_client=anthropic_client,
                 )
@@ -289,7 +296,7 @@ class MyOptions(AnthropicChatOptions, total=False):
                     my_custom_option: str
 
 
-                client: AnthropicClient[MyOptions] = AnthropicClient(model_id="claude-sonnet-4-5-20250929")
+                client: RawAnthropicClient[MyOptions] = RawAnthropicClient(model_id="claude-sonnet-4-5-20250929")
                 response = await client.get_response("Hello", options={"my_custom_option": "value"})
 
         """
@@ -320,8 +327,6 @@ class MyOptions(AnthropicChatOptions, total=False):
         # Initialize parent
         super().__init__(
             additional_properties=additional_properties,
-            middleware=middleware,
-            function_invocation_configuration=function_invocation_configuration,
         )
 
         # Initialize instance variables
@@ -1376,3 +1381,95 @@ def service_url(self) -> str:
             The service URL for the chat client, or None if not set.
         """
         return str(self.anthropic_client.base_url)
+
+
+class AnthropicClient(
+    FunctionInvocationLayer[AnthropicOptionsT],
+    ChatMiddlewareLayer[AnthropicOptionsT],
+    ChatTelemetryLayer[AnthropicOptionsT],
+    RawAnthropicClient[AnthropicOptionsT],
+    Generic[AnthropicOptionsT],
+):
+    """Anthropic chat client with middleware, telemetry, and function invocation support."""
+
+    def __init__(
+        self,
+        *,
+        api_key: str | None = None,
+        model_id: str | None = None,
+        anthropic_client: AsyncAnthropic | None = None,
+        additional_beta_flags: list[str] | None = None,
+        additional_properties: dict[str, Any] | None = None,
+        middleware: Sequence[ChatAndFunctionMiddlewareTypes] | None = None,
+        function_invocation_configuration: FunctionInvocationConfiguration | None = None,
+        env_file_path: str | None = None,
+        env_file_encoding: str | None = None,
+    ) -> None:
+        """Initialize an Anthropic client.
+
+        Keyword Args:
+            api_key: The Anthropic API key to use for authentication.
+            model_id: The ID of the model to use.
+            anthropic_client: An existing Anthropic client to use. If not provided, one will be created.
+                This can be used to further configure the client before passing it in.
+                For instance if you need to set a different base_url for testing or private deployments.
+            additional_beta_flags: Additional beta flags to enable on the client.
+                Default flags are: "mcp-client-2025-04-04", "code-execution-2025-08-25".
+            additional_properties: Additional properties stored on the client instance.
+            middleware: Optional middleware to apply to the client.
+            function_invocation_configuration: Optional function invocation configuration override.
+            env_file_path: Path to environment file for loading settings.
+            env_file_encoding: Encoding of the environment file.
+
+        Examples:
+            .. code-block:: python
+
+                from agent_framework.anthropic import AnthropicClient
+
+                # Using environment variables
+                # Set ANTHROPIC_API_KEY=your_anthropic_api_key
+                # ANTHROPIC_CHAT_MODEL_ID=claude-sonnet-4-5-20250929
+
+                # Or passing parameters directly
+                client = AnthropicClient(
+                    model_id="claude-sonnet-4-5-20250929",
+                    api_key="your_anthropic_api_key",
+                )
+
+                # Or loading from a .env file
+                client = AnthropicClient(env_file_path="path/to/.env")
+
+                # Or passing in an existing client
+                from anthropic import AsyncAnthropic
+
+                anthropic_client = AsyncAnthropic(
+                    api_key="your_anthropic_api_key", base_url="https://custom-anthropic-endpoint.com"
+                )
+                client = AnthropicClient(
+                    model_id="claude-sonnet-4-5-20250929",
+                    anthropic_client=anthropic_client,
+                )
+
+                # Using custom ChatOptions with type safety:
+                from typing import TypedDict
+                from agent_framework.anthropic import AnthropicChatOptions
+
+
+                class MyOptions(AnthropicChatOptions, total=False):
+                    my_custom_option: str
+
+
+                client: AnthropicClient[MyOptions] = AnthropicClient(model_id="claude-sonnet-4-5-20250929")
+                response = await client.get_response("Hello", options={"my_custom_option": "value"})
+        """
+        super().__init__(
+            api_key=api_key,
+            model_id=model_id,
+            anthropic_client=anthropic_client,
+            additional_beta_flags=additional_beta_flags,
+            additional_properties=additional_properties,
+            middleware=middleware,
+            function_invocation_configuration=function_invocation_configuration,
+            env_file_path=env_file_path,
+            env_file_encoding=env_file_encoding,
+        )

python/packages/anthropic/tests/test_anthropic_client.py

@@ -6,15 +6,18 @@
 
 import pytest
 from agent_framework import (
+    ChatMiddlewareLayer,
     ChatOptions,
     ChatResponseUpdate,
     Content,
+    FunctionInvocationLayer,
     Message,
     SupportsChatGetResponse,
     tool,
 )
 from agent_framework._settings import load_settings
 from agent_framework._tools import SHELL_TOOL_KIND_VALUE
+from agent_framework.observability import ChatTelemetryLayer
 from anthropic.types.beta import (
     BetaMessage,
     BetaTextBlock,
@@ -23,7 +26,7 @@
 )
 from pydantic import BaseModel, Field
 
-from agent_framework_anthropic import AnthropicClient
+from agent_framework_anthropic import AnthropicClient, RawAnthropicClient
 from agent_framework_anthropic._chat_client import AnthropicSettings
 
 # Test constants
@@ -64,6 +67,8 @@ def create_test_anthropic_client(
     client.additional_beta_flags = []
     client.chat_middleware = []
     client.function_middleware = []
+    client._cached_chat_middleware_pipeline = None
+    client._cached_function_middleware_pipeline = None
     client.function_invocation_configuration = normalize_function_invocation_configuration(None)
 
     return client
@@ -117,6 +122,19 @@ def test_anthropic_client_init_with_client(mock_anthropic_client: MagicMock) ->
     assert isinstance(client, SupportsChatGetResponse)
 
 
+def test_anthropic_client_wraps_raw_client_with_standard_layer_order() -> None:
+    """Test AnthropicClient composes the standard public layer stack around the raw client."""
+    assert issubclass(AnthropicClient, RawAnthropicClient)
+    mro = AnthropicClient.__mro__
+    assert mro.index(FunctionInvocationLayer) < mro.index(ChatMiddlewareLayer)
+    assert mro.index(ChatMiddlewareLayer) < mro.index(ChatTelemetryLayer)
+    assert mro.index(ChatTelemetryLayer) < mro.index(RawAnthropicClient)
+    # RawAnthropicClient must not include the convenience layers
+    assert not issubclass(RawAnthropicClient, FunctionInvocationLayer)
+    assert not issubclass(RawAnthropicClient, ChatMiddlewareLayer)
+    assert not issubclass(RawAnthropicClient, ChatTelemetryLayer)
+
+
 def test_anthropic_client_init_auto_create_client(
     anthropic_unit_test_env: dict[str, str],
 ) -> None:

python/packages/azure-ai/agent_framework_azure_ai/_chat_client.py

@@ -206,8 +206,8 @@ class AzureAIAgentOptions(ChatOptions, total=False):
 
 
 class AzureAIAgentClient(
-    ChatMiddlewareLayer[AzureAIAgentOptionsT],
     FunctionInvocationLayer[AzureAIAgentOptionsT],
+    ChatMiddlewareLayer[AzureAIAgentOptionsT],
     ChatTelemetryLayer[AzureAIAgentOptionsT],
     BaseChatClient[AzureAIAgentOptionsT],
     Generic[AzureAIAgentOptionsT],

python/packages/azure-ai/agent_framework_azure_ai/_client.py

@@ -97,9 +97,9 @@ class RawAzureAIClient(RawOpenAIResponsesClient[AzureAIClientOptionsT], Generic[
         you should consider which additional layers to apply. There is a defined ordering that
         you should follow:
 
-        1. **ChatMiddlewareLayer** - Should be applied first as it also prepares function middleware
-        2. **FunctionInvocationLayer** - Handles tool/function calling loop
-        3. **ChatTelemetryLayer** - Must be inside the function calling loop for correct per-call telemetry
+        1. **FunctionInvocationLayer** - Owns the tool/function calling loop and routes function middleware
+        2. **ChatMiddlewareLayer** - Applies chat middleware per model call and stays outside telemetry
+        3. **ChatTelemetryLayer** - Must stay inside chat middleware for correct per-call telemetry
 
         Use ``AzureAIClient`` instead for a fully-featured client with all layers applied.
     """
@@ -1214,8 +1214,8 @@ def as_agent(
 
 
 class AzureAIClient(
-    ChatMiddlewareLayer[AzureAIClientOptionsT],
     FunctionInvocationLayer[AzureAIClientOptionsT],
+    ChatMiddlewareLayer[AzureAIClientOptionsT],
     ChatTelemetryLayer[AzureAIClientOptionsT],
     RawAzureAIClient[AzureAIClientOptionsT],
     Generic[AzureAIClientOptionsT],

python/packages/azure-ai/tests/test_azure_ai_agent_client.py

@@ -87,6 +87,8 @@ def create_test_azure_ai_chat_client(
     client.middleware = None
     client.chat_middleware = []
     client.function_middleware = []
+    client._cached_chat_middleware_pipeline = None
+    client._cached_function_middleware_pipeline = None
     client.otel_provider_name = "azure.ai"
     client.function_invocation_configuration = {
         "enabled": True,
@@ -151,6 +153,10 @@ def test_azure_ai_chat_client_init_auto_create_client(
     chat_client.agent_name = None
     chat_client.additional_properties = {}
     chat_client.middleware = None
+    chat_client.chat_middleware = []
+    chat_client.function_middleware = []
+    chat_client._cached_chat_middleware_pipeline = None
+    chat_client._cached_function_middleware_pipeline = None
 
     assert chat_client.agents_client is mock_agents_client
     assert chat_client.agent_id is None

python/packages/bedrock/agent_framework_bedrock/_chat_client.py

@@ -216,8 +216,8 @@ class BedrockSettings(TypedDict, total=False):
 
 
 class BedrockChatClient(
-    ChatMiddlewareLayer[BedrockChatOptionsT],
     FunctionInvocationLayer[BedrockChatOptionsT],
+    ChatMiddlewareLayer[BedrockChatOptionsT],
     ChatTelemetryLayer[BedrockChatOptionsT],
     BaseChatClient[BedrockChatOptionsT],
     Generic[BedrockChatOptionsT],

python/packages/core/agent_framework/_clients.py

@@ -966,16 +966,7 @@ def _apply_get_response_docstrings() -> None:
     from .observability import ChatTelemetryLayer
 
     apply_layered_docstring(ChatTelemetryLayer.get_response, BaseChatClient.get_response)
-    apply_layered_docstring(
-        FunctionInvocationLayer.get_response,
-        ChatTelemetryLayer.get_response,
-        extra_keyword_args={
-            "function_middleware": """
-                Optional per-call function middleware.
-                When omitted, middleware configured on the client or forwarded from higher layers is used.
-            """,
-        },
-    )
+    apply_layered_docstring(FunctionInvocationLayer.get_response, ChatTelemetryLayer.get_response)
     apply_layered_docstring(
         ChatMiddlewareLayer.get_response,
         FunctionInvocationLayer.get_response,