diff --git a/python/packages/ollama/agent_framework_ollama/_chat_client.py b/python/packages/ollama/agent_framework_ollama/_chat_client.py
index ea652aff941..4bda11a11a9 100644
--- a/python/packages/ollama/agent_framework_ollama/_chat_client.py
+++ b/python/packages/ollama/agent_framework_ollama/_chat_client.py
@@ -32,6 +32,7 @@
     Message,
     ResponseStream,
     UsageDetails,
+    validate_tool_mode,
 )
 from agent_framework._settings import load_settings
 from agent_framework._telemetry import mark_feature_used
@@ -97,7 +98,8 @@ class OllamaChatOptions(ChatOptions[ResponseModelT], Generic[ResponseModelT], to
             (converted to its JSON schema) for structured output.
 
         # Options not supported in Ollama:
-        tool_choice: Ollama only supports auto tool choice.
+        tool_choice: Only ``auto`` and ``none`` are supported (``none`` omits the tools).
+            ``required`` and ``allowed_tools`` raise ``ChatClientInvalidRequestException``.
         allow_multiple_tool_calls: Not configurable.
         user: Not supported.
         store: Not supported.
@@ -216,9 +218,6 @@ class OllamaChatOptions(ChatOptions[ResponseModelT], Generic[ResponseModelT], to
     """For thinking models: whether the model should think before responding."""
 
     # ChatOptions fields not supported in Ollama
-    tool_choice: None  # type: ignore[misc]
-    """Not supported. Ollama only supports auto tool choice."""
-
     allow_multiple_tool_calls: None  # type: ignore[misc]
     """Not supported. Not configurable in Ollama."""
 
@@ -414,7 +413,7 @@ def _prepare_options(self, messages: Sequence[Message], options: Mapping[str, An
             messages = prepend_instructions_to_messages(list(messages), instructions, role="system")
 
         # Keys to exclude from processing
-        exclude_keys = {"instructions", "tool_choice"}
+        exclude_keys = {"instructions", "tool_choice", "tools"}
 
         # Build run_options and model_options separately
         run_options: dict[str, Any] = {}
@@ -460,8 +459,21 @@ def _prepare_options(self, messages: Sequence[Message], options: Mapping[str, An
             run_options["model"] = self.model
 
         # tools
+        # Ollama has no tool_choice parameter. Only "auto" (send the tools) and "none" (don't send them)
+        # can be honored, so anything else is rejected instead of silently behaving like "auto".
+        tool_mode = validate_tool_mode(options.get("tool_choice"))
+        if tool_mode is not None:
+            if tool_mode.get("mode") == "required":
+                raise ChatClientInvalidRequestException(
+                    "Ollama does not support tool_choice mode 'required'. Use 'auto' or 'none'."
+                )
+            if "allowed_tools" in tool_mode:
+                raise ChatClientInvalidRequestException(
+                    "Ollama does not support tool_choice 'allowed_tools'. Pass only the tools you want to allow."
+                )
         tools = options.get("tools")
-        if tools is not None and (prepared_tools := self._prepare_tools_for_ollama(tools)):
+        tools_disabled = tool_mode is not None and tool_mode.get("mode") == "none"
+        if tools is not None and not tools_disabled and (prepared_tools := self._prepare_tools_for_ollama(tools)):
             run_options["tools"] = prepared_tools
 
         return run_options
diff --git a/python/packages/ollama/tests/test_ollama_chat_client.py b/python/packages/ollama/tests/test_ollama_chat_client.py
index be5c62218a3..90c66b18794 100644
--- a/python/packages/ollama/tests/test_ollama_chat_client.py
+++ b/python/packages/ollama/tests/test_ollama_chat_client.py
@@ -17,7 +17,12 @@
     chat_middleware,
     tool,
 )
-from agent_framework.exceptions import ChatClientException, ChatClientInvalidRequestException, SettingNotFoundError
+from agent_framework.exceptions import (
+    ChatClientException,
+    ChatClientInvalidRequestException,
+    ContentError,
+    SettingNotFoundError,
+)
 from ollama import AsyncClient
 from ollama._types import ChatResponse as OllamaChatResponse
 from ollama._types import Message as OllamaMessage
@@ -910,6 +915,80 @@ def test_prepare_options_single_stop_string_becomes_list(ollama_unit_test_env: d
     assert request["options"]["stop"] == ["END"]
 
 
+@pytest.mark.parametrize("tool_choice", ["none", {"mode": "none"}])
+def test_prepare_options_tool_choice_none_omits_tools(ollama_unit_test_env: dict[str, str], tool_choice: Any) -> None:
+    """Ollama has no tool_choice parameter, so "none" is honored by not offering the tools."""
+    client = OllamaChatClient()
+    messages = [Message(role="user", contents=[Content.from_text(text="hello")])]
+
+    request = client._prepare_options(messages, {"tools": [hello_world], "tool_choice": tool_choice})
+
+    assert "tools" not in request
+    assert "tool_choice" not in request
+
+
+@pytest.mark.parametrize("tool_choice", [None, "auto", {"mode": "auto"}])
+def test_prepare_options_auto_tool_choice_keeps_tools(ollama_unit_test_env: dict[str, str], tool_choice: Any) -> None:
+    """No tool_choice or "auto" sends the tools to Ollama."""
+    client = OllamaChatClient()
+    messages = [Message(role="user", contents=[Content.from_text(text="hello")])]
+
+    request = client._prepare_options(messages, {"tools": [hello_world], "tool_choice": tool_choice})
+
+    assert request["tools"] == [hello_world.to_json_schema_spec()]
+    assert "tool_choice" not in request
+
+
+@patch.object(AsyncClient, "chat", new_callable=AsyncMock)
+async def test_cmc_function_invocation_limit_final_request_omits_tools(
+    mock_chat: AsyncMock,
+    ollama_unit_test_env: dict[str, str],
+    chat_history: list[Message],
+    mock_chat_completion_tool_call: OllamaChatResponse,
+    mock_chat_completion_response: OllamaChatResponse,
+) -> None:
+    """After the function invocation limit, the final request must not offer tools to Ollama."""
+    mock_chat.side_effect = [mock_chat_completion_tool_call, mock_chat_completion_response]
+    chat_history.append(Message(contents=["hello world"], role="user"))
+
+    ollama_client = OllamaChatClient()
+    ollama_client.function_invocation_configuration["max_iterations"] = 1
+    result = await ollama_client.get_response(messages=chat_history, options={"tools": [hello_world]})
+
+    assert mock_chat.call_count == 2
+    assert "tools" in mock_chat.call_args_list[0].kwargs
+    assert "tools" not in mock_chat.call_args_list[1].kwargs
+    assert result.text == "test"
+
+
+def test_prepare_options_invalid_tool_choice_raises(ollama_unit_test_env: dict[str, str]) -> None:
+    """An invalid tool_choice is rejected instead of being silently ignored."""
+    client = OllamaChatClient()
+    messages = [Message(role="user", contents=[Content.from_text(text="hello")])]
+
+    with pytest.raises(ContentError):
+        client._prepare_options(messages, {"tools": [hello_world], "tool_choice": "bogus"})
+
+
+@pytest.mark.parametrize(
+    "tool_choice",
+    [
+        "required",
+        {"mode": "required"},
+        {"mode": "required", "required_function_name": "hello_world"},
+        {"mode": "auto", "allowed_tools": ["hello_world"]},
+        {"mode": "required", "allowed_tools": ["hello_world"]},
+    ],
+)
+def test_prepare_options_unsupported_tool_choice_raises(ollama_unit_test_env: dict[str, str], tool_choice: Any) -> None:
+    """Ollama can't enforce "required" or "allowed_tools", so they raise instead of acting like "auto"."""
+    client = OllamaChatClient()
+    messages = [Message(role="user", contents=[Content.from_text(text="hello")])]
+
+    with pytest.raises(ChatClientInvalidRequestException):
+        client._prepare_options(messages, {"tools": [hello_world], "tool_choice": tool_choice})
+
+
 async def test_cmc_streaming_closes_sdk_stream_when_consumer_stops_early(ollama_unit_test_env: dict[str, str]) -> None:
     """Stopping early must close the Ollama SDK stream instead of leaving it to garbage collection."""
     closed = False