fix(models): keep LiteLLM chat-completions tool schema when STRIX_API_TYPE=responses

Add test for API type override settings in LlmSettings.

docs/advanced/configuration.mdx

@@ -19,6 +19,12 @@ Configure Strix using environment variables or a config file.
   Custom API base URL. Also accepts `OPENAI_API_BASE`, `LITELLM_BASE_URL`, or `OLLAMA_API_BASE`.
 </ParamField>
 
+<ParamField path="STRIX_API_TYPE" type="string">
+  Select the OpenAI API path for the model: `responses` or `chat_completions`.
+  By default, a custom `LLM_API_BASE` uses chat completions. Set this variable
+  when your gateway requires the other API. Also accepts `STRIX_FORCE_API`.
+</ParamField>
+
 <ParamField path="LLM_EXTRA_HEADERS" type="string">
   Extra HTTP headers sent on every LLM request, as a JSON object (e.g.
   `{"X-Feature-Key":"value","X-Tenant":"acme"}`). Useful for OpenAI-compatible

strix/config/models.py

@@ -632,9 +632,11 @@ def configure_sdk_model_defaults(settings: Settings) -> None:
     if llm.api_base:
         os.environ["OPENAI_BASE_URL"] = llm.api_base
         _configure_litellm_default("api_base", llm.api_base)
-        set_default_openai_api("chat_completions")
-    else:
-        set_default_openai_api("responses")
+    api_type = llm.api_type
+    if api_type is None:
+        api_type = "chat_completions" if llm.api_base else "responses"
+
+    set_default_openai_api(api_type)
     _configure_extra_headers(llm)
 
 
@@ -809,6 +811,8 @@ def uses_chat_completions_tool_schema(model_name: str, settings: Settings) -> bo
     model = model_name.strip().lower()
     if "/" in model and not model.startswith("openai/"):
         return True
+    if settings.llm.api_type is not None:
+        return settings.llm.api_type == "chat_completions"
     if settings.llm.api_base:
         return True
     return not model_supports_reasoning(model_name)

strix/config/settings.py

@@ -9,6 +9,7 @@ from pydantic_settings import BaseSettings, SettingsConfigDict
 
 
 ReasoningEffort = Literal["none", "minimal", "low", "medium", "high", "xhigh", "max"]
+ApiType = Literal["responses", "chat_completions"]
 
 DEFAULT_MAX_TURNS = 500
 
@@ -23,6 +24,11 @@ class LlmSettings(BaseSettings):
     model_config = _BASE_CONFIG
 
     model: str | None = Field(default=None, alias="STRIX_LLM")
+    api_type: ApiType | None = Field(
+        default=None,
+        validation_alias=AliasChoices("STRIX_API_TYPE", "STRIX_FORCE_API"),
+        description="Force 'responses' or 'chat_completions' API path",
+    )
     api_key: str | None = Field(
         default=None,
         validation_alias=AliasChoices("LLM_API_KEY", "OPENAI_API_KEY"),

tests/test_models.py

@@ -2,20 +2,27 @@
 
 from __future__ import annotations
 
+import litellm
 import pytest
 from agents.extensions.models.litellm_model import LitellmModel
 from agents.model_settings import ModelSettings
+from agents.models import _openai_shared
+from agents.models.openai_chatcompletions import OpenAIChatCompletionsModel
+from agents.models.openai_responses import OpenAIResponsesModel
 
 from strix.config.models import (
     RECOMMENDED_MODEL_NAMES,
     StrixProvider,
     _NonStreamingModel,
     _TurnGuardModel,
+    configure_sdk_model_defaults,
     is_recommended_or_frontier_model,
     request_timeout_extra_args,
     routes_through_litellm,
     supports_strict_tool_schemas,
+    uses_chat_completions_tool_schema,
 )
+from strix.config.settings import Settings
 
 
 @pytest.mark.parametrize("model_name", RECOMMENDED_MODEL_NAMES)
@@ -168,3 +175,46 @@ def test_routes_through_litellm_matches_the_provider(
     while isinstance(model, _NonStreamingModel | _TurnGuardModel):
         model = model._inner
     assert isinstance(model, LitellmModel) is litellm
+
+
+def test_api_type_override_settings(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setenv("STRIX_LLM", "gpt-4")
+    monkeypatch.setenv("STRIX_API_TYPE", "chat_completions")
+    assert uses_chat_completions_tool_schema("gpt-4", Settings()) is True
+    monkeypatch.setenv("STRIX_LLM", "openai/gpt-4")
+    monkeypatch.setenv("STRIX_API_TYPE", "responses")
+    assert uses_chat_completions_tool_schema("openai/gpt-4", Settings()) is False
+    monkeypatch.setenv("STRIX_LLM", "anthropic/claude-sonnet-4-5")
+    assert uses_chat_completions_tool_schema("anthropic/claude-sonnet-4-5", Settings()) is True
+
+
+@pytest.mark.parametrize(
+    ("api_type", "expected"),
+    [
+        (None, OpenAIChatCompletionsModel),
+        ("chat_completions", OpenAIChatCompletionsModel),
+        ("responses", OpenAIResponsesModel),
+    ],
+)
+def test_api_type_overrides_the_api_base_route(
+    monkeypatch: pytest.MonkeyPatch, api_type: str | None, expected: type
+) -> None:
+    """``LLM_API_BASE`` defaults to chat completions. ``STRIX_API_TYPE`` must win."""
+    monkeypatch.setattr(_openai_shared, "_use_responses_by_default", True)
+    monkeypatch.setattr(_openai_shared, "_default_openai_client", None)
+    monkeypatch.setattr(_openai_shared, "_default_openai_key", None)
+    monkeypatch.setattr(litellm, "api_key", None)
+    monkeypatch.setattr(litellm, "api_base", None)
+    monkeypatch.setenv("OPENAI_API_KEY", "test-key")
+    monkeypatch.setenv("OPENAI_BASE_URL", "")
+    monkeypatch.setenv("STRIX_LLM", "gpt-5")
+    monkeypatch.setenv("LLM_API_KEY", "test-key")
+    monkeypatch.setenv("LLM_API_BASE", "https://gateway.example/v1")
+    monkeypatch.delenv("STRIX_API_TYPE", raising=False)
+    if api_type is not None:
+        monkeypatch.setenv("STRIX_API_TYPE", api_type)
+    configure_sdk_model_defaults(Settings())
+    model = StrixProvider().get_model("gpt-5")
+    while isinstance(model, _NonStreamingModel | _TurnGuardModel):
+        model = model._inner
+    assert isinstance(model, expected)