feat(nvidia): add NIM billing origin header

tests/agent/test_auxiliary_client.py

@@ -2415,10 +2415,51 @@ def _clean_env(monkeypatch):
     """Strip provider env vars so each test starts clean."""
     for key in (
         "OPENROUTER_API_KEY", "OPENAI_BASE_URL", "OPENAI_API_KEY",
+        "NVIDIA_API_KEY", "NVIDIA_BASE_URL",
     ):
         monkeypatch.delenv(key, raising=False)
 
 
+class TestNvidiaBillingHeaders:
+    """NVIDIA NIM billing-origin headers are scoped to NVIDIA cloud."""
+
+    def test_resolve_provider_client_cloud_adds_billing_origin_header(self, monkeypatch):
+        monkeypatch.setenv("NVIDIA_API_KEY", "nvidia-key")
+        monkeypatch.delenv("NVIDIA_BASE_URL", raising=False)
+        mock_openai = MagicMock()
+        mock_openai.return_value = MagicMock(name="nvidia-client")
+
+        with patch("agent.auxiliary_client.OpenAI", mock_openai):
+            client, model = resolve_provider_client(
+                provider="nvidia",
+                model="nvidia/test-model",
+            )
+
+        assert client is not None
+        assert model == "nvidia/test-model"
+        call_kwargs = mock_openai.call_args[1]
+        headers = call_kwargs["default_headers"]
+        assert headers["X-BILLING-INVOKE-ORIGIN"] == "HermesAgent"
+
+    def test_resolve_provider_client_local_nim_skips_billing_origin_header(self, monkeypatch):
+        monkeypatch.setenv("NVIDIA_API_KEY", "nvidia-key")
+        monkeypatch.setenv("NVIDIA_BASE_URL", "http://localhost:8000/v1")
+        mock_openai = MagicMock()
+        mock_openai.return_value = MagicMock(name="nvidia-local-client")
+
+        with patch("agent.auxiliary_client.OpenAI", mock_openai):
+            client, model = resolve_provider_client(
+                provider="nvidia",
+                model="nvidia/test-model",
+            )
+
+        assert client is not None
+        assert model == "nvidia/test-model"
+        call_kwargs = mock_openai.call_args[1]
+        headers = call_kwargs.get("default_headers", {})
+        assert "X-BILLING-INVOKE-ORIGIN" not in headers
+
+
 class TestOpenRouterExplicitApiKey:
     """Test that explicit_api_key is correctly propagated to _try_openrouter()."""
 

tests/providers/test_provider_profiles.py

@@ -42,6 +42,10 @@ class TestNvidiaProfile:
         p = get_provider_profile("nvidia")
         assert "nvidia.com" in p.base_url
 
+    def test_billing_header_not_profile_wide(self):
+        p = get_provider_profile("nvidia")
+        assert p.default_headers == {}
+
 
 class TestKimiProfile:
     def test_temperature_omit(self):

tests/run_agent/test_provider_attribution_headers.py

@@ -3,6 +3,7 @@
 Mirrors the OpenRouter pattern for the Vercel AI Gateway so that
 referrerUrl / appName / User-Agent flow into gateway analytics.
 """
+from types import SimpleNamespace
 from unittest.mock import MagicMock, patch
 
 from run_agent import AIAgent
@@ -65,6 +66,73 @@ def test_routermint_base_url_applies_user_agent_header(mock_openai):
     assert headers["User-Agent"].startswith("HermesAgent/")
 
 
+@patch("run_agent.OpenAI")
+def test_nvidia_cloud_base_url_applies_billing_origin_header(mock_openai):
+    mock_openai.return_value = MagicMock()
+    agent = AIAgent(
+        api_key="test-key",
+        base_url="https://integrate.api.nvidia.com/v1",
+        model="nvidia/test-model",
+        provider="nvidia",
+        quiet_mode=True,
+        skip_context_files=True,
+        skip_memory=True,
+    )
+
+    assert agent._client_kwargs["default_headers"]["X-BILLING-INVOKE-ORIGIN"] == "HermesAgent"
+
+    agent._apply_client_headers_for_base_url("https://integrate.api.nvidia.com/v1")
+
+    headers = agent._client_kwargs["default_headers"]
+    assert headers["X-BILLING-INVOKE-ORIGIN"] == "HermesAgent"
+
+
+@patch("run_agent.OpenAI")
+def test_nvidia_local_base_url_does_not_apply_billing_origin_header(mock_openai):
+    mock_openai.return_value = MagicMock()
+    agent = AIAgent(
+        api_key="test-key",
+        base_url="https://integrate.api.nvidia.com/v1",
+        model="nvidia/test-model",
+        provider="nvidia",
+        quiet_mode=True,
+        skip_context_files=True,
+        skip_memory=True,
+    )
+    agent._client_kwargs["default_headers"] = {
+        "X-BILLING-INVOKE-ORIGIN": "HermesAgent",
+    }
+
+    agent._apply_client_headers_for_base_url("http://localhost:8000/v1")
+
+    assert "default_headers" not in agent._client_kwargs
+
+
+@patch("run_agent.OpenAI")
+def test_routed_client_preserves_openai_sdk_custom_headers(mock_openai):
+    mock_openai.return_value = MagicMock()
+    routed_client = SimpleNamespace(
+        api_key="test-key",
+        base_url="https://integrate.api.nvidia.com/v1",
+        _custom_headers={"X-BILLING-INVOKE-ORIGIN": "HermesAgent"},
+    )
+
+    with patch("agent.auxiliary_client.resolve_provider_client", return_value=(
+        routed_client,
+        "nvidia/test-model",
+    )):
+        agent = AIAgent(
+            provider="nvidia",
+            model="nvidia/test-model",
+            quiet_mode=True,
+            skip_context_files=True,
+            skip_memory=True,
+        )
+
+    headers = agent._client_kwargs["default_headers"]
+    assert headers["X-BILLING-INVOKE-ORIGIN"] == "HermesAgent"
+
+
 @patch("run_agent.OpenAI")
 def test_gmi_base_url_picks_up_profile_user_agent(mock_openai):
     """GMI declares User-Agent on its ProviderProfile.default_headers.