fix(chat): route OpenClaw chat through Ollama lane

apps/api/tests/test_chat_manager_ollama_routing.py

@@ -0,0 +1,105 @@
+from __future__ import annotations
+
+from pathlib import Path
+from types import SimpleNamespace
+from typing import Any
+
+import pytest
+
+from src.services import chat_manager as chat_module
+from src.services.chat_manager import ChatManager
+
+
+class _FakeResponse:
+    def __init__(self, content: str = "老闆，系統目前穩定。") -> None:
+        self._content = content
+
+    def raise_for_status(self) -> None:
+        return None
+
+    def json(self) -> dict[str, Any]:
+        return {
+            "message": {"content": self._content},
+            "prompt_eval_count": 11,
+            "eval_count": 13,
+        }
+
+
+class _FakeAsyncClient:
+    posted: list[tuple[str, dict[str, Any]]] = []
+
+    def __init__(self, *args: Any, **kwargs: Any) -> None:
+        self.args = args
+        self.kwargs = kwargs
+
+    async def __aenter__(self) -> _FakeAsyncClient:
+        return self
+
+    async def __aexit__(self, *args: Any) -> None:
+        return None
+
+    async def post(self, url: str, *, json: dict[str, Any]) -> _FakeResponse:
+        self.posted.append((url, json))
+        return _FakeResponse()
+
+
+def _settings() -> SimpleNamespace:
+    return SimpleNamespace(OPENCLAW_DEFAULT_MODEL="qwen3:14b")
+
+
+@pytest.fixture(autouse=True)
+def _reset_fake_client() -> None:
+    _FakeAsyncClient.posted = []
+
+
+@pytest.mark.asyncio
+async def test_openclaw_chat_uses_ollama_interactive_lane(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(chat_module.httpx, "AsyncClient", _FakeAsyncClient)
+    monkeypatch.setattr(chat_module, "get_settings", _settings)
+    monkeypatch.setattr(
+        chat_module,
+        "resolve_ollama_endpoint",
+        lambda workload_type: "http://gcp-a:11435",
+    )
+
+    result = await ChatManager()._call_openclaw("system context", "幫我看狀態")
+
+    assert result is not None
+    assert "qwen3:14b" in result
+    assert "免費" in result
+
+    assert len(_FakeAsyncClient.posted) == 1
+    url, payload = _FakeAsyncClient.posted[0]
+    assert url == "http://gcp-a:11435/api/chat"
+    assert payload["model"] == "qwen3:14b"
+    assert payload["messages"][0]["role"] == "system"
+    assert payload["messages"][1] == {"role": "user", "content": "幫我看狀態"}
+
+
+@pytest.mark.asyncio
+async def test_nemoclaw_chat_uses_resolved_interactive_lane(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(chat_module.httpx, "AsyncClient", _FakeAsyncClient)
+    monkeypatch.setattr(
+        chat_module,
+        "resolve_ollama_endpoint",
+        lambda workload_type: "http://gcp-a:11435",
+    )
+
+    result = await ChatManager()._call_nemotron("system context", "補充觀點")
+
+    assert result is not None
+    url, payload = _FakeAsyncClient.posted[0]
+    assert url == "http://gcp-a:11435/api/chat"
+    assert payload["model"] == "deepseek-r1:14b"
+
+
+def test_chat_manager_has_no_direct_gemini_generation_path() -> None:
+    source_path = Path(chat_module.__file__).resolve()
+    source = source_path.read_text(encoding="utf-8")
+
+    assert "generativelanguage.googleapis.com" not in source
+    assert "GEMINI_API_KEY" not in source