fix(llm): support Volcengine Coding model aliases (#411)

* fix(llm): support Volcengine Coding model aliases

* refactor(llm): add Volcengine Coding provider

* style: format Volcengine Coding test

EvoScientist/config/onboard/constants.py

@@ -21,6 +21,7 @@ VALID_PROVIDERS: frozenset[str] = frozenset(
         "zhipu",
         "zhipu-code",
         "volcengine",
+        "volcengine-code",
         "dashscope",
         "dashscope-code",
         "deepseek",

EvoScientist/config/onboard/helpers.py

@@ -102,6 +102,11 @@ def _provider_key_info(config: EvoScientistConfig, provider: str):
             config.volcengine_api_key or os.environ.get("VOLCENGINE_API_KEY", ""),
             validate_volcengine_key,
         ),
+        "volcengine-code": (
+            "Volcengine Coding Plan",
+            config.volcengine_api_key or os.environ.get("VOLCENGINE_API_KEY", ""),
+            validate_volcengine_key,
+        ),
         "dashscope": (
             "DashScope",
             config.dashscope_api_key or os.environ.get("DASHSCOPE_API_KEY", ""),

EvoScientist/config/onboard/steps.py

@@ -278,6 +278,10 @@ def _step_provider(
             title="Volcengine (火山引擎 — Doubao models)",
             value="volcengine",
         ),
+        Choice(
+            title="Volcengine Coding Plan (火山引擎代码计划 — coding models)",
+            value="volcengine-code",
+        ),
         Choice(
             title="DashScope (阿里云 — Qwen models)",
             value="dashscope",

EvoScientist/config/onboard/validators.py

@@ -475,6 +475,9 @@ def validate_zhipu_key(api_key: str) -> tuple[bool, str]:
 def validate_volcengine_key(api_key: str) -> tuple[bool, str]:
     """Validate a Volcengine API key by making a test request.
 
+    Uses the general endpoint for validation; volcengine and volcengine-code
+    share the same API key and only differ in their runtime base URL.
+
     Returns:
         Tuple of (is_valid, message).
     """

EvoScientist/config/onboard/wizard.py

@@ -124,6 +124,7 @@ _PROVIDER_KEY_ATTR = {
     "zhipu": "zhipu_api_key",
     "zhipu-code": "zhipu_api_key",
     "volcengine": "volcengine_api_key",
+    "volcengine-code": "volcengine_api_key",
     "dashscope": "dashscope_api_key",
     "dashscope-code": "dashscope_api_key",
     "moonshot": "moonshot_api_key",

EvoScientist/llm/context_window.py

@@ -32,6 +32,8 @@ _KNOWN_MODEL_CONTEXT_WINDOWS: dict[str, int] = {
     # Zhipu GLM-5.2 — 1M context, an exception to the ``glm-5`` family (203K).
     # Matches OpenRouter ``z-ai/glm-5.2`` via split('/')[-1].
     "glm-5.2": 1_000_000,
+    # Volcengine Coding Plan's OpenAI-compatible alias for GLM-5.2.
+    "glm-5-2": 1_000_000,
     # Tencent Hunyuan HY3 — 262K context (OpenRouter ``tencent/hy3``).
     "hy3": 262_000,
 }

EvoScientist/llm/models.py

@@ -43,6 +43,7 @@ _SILICONFLOW_BASE_URL = "https://api.siliconflow.cn/v1"
 _ZHIPU_BASE_URL = "https://open.bigmodel.cn/api/paas/v4"
 _ZHIPU_CODE_BASE_URL = "https://open.bigmodel.cn/api/coding/paas/v4"
 _VOLCENGINE_BASE_URL = "https://ark.cn-beijing.volces.com/api/v3"
+_VOLCENGINE_CODE_BASE_URL = "https://ark.cn-beijing.volces.com/api/coding/v3"
 _DASHSCOPE_BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1"
 _DASHSCOPE_CODE_BASE_URL = "https://coding.dashscope.aliyuncs.com/v1"
 
@@ -114,6 +115,7 @@ _OPENAI_ROUTED_PROVIDERS: dict[str, tuple[str | None, str]] = {
     "zhipu": (_ZHIPU_BASE_URL, "ZHIPU_API_KEY"),
     "zhipu-code": (_ZHIPU_CODE_BASE_URL, "ZHIPU_API_KEY"),
     "volcengine": (_VOLCENGINE_BASE_URL, "VOLCENGINE_API_KEY"),
+    "volcengine-code": (_VOLCENGINE_CODE_BASE_URL, "VOLCENGINE_API_KEY"),
     "dashscope": (_DASHSCOPE_BASE_URL, "DASHSCOPE_API_KEY"),
     "dashscope-code": (_DASHSCOPE_CODE_BASE_URL, "DASHSCOPE_API_KEY"),
     "requesty": (_REQUESTY_BASE_URL, "REQUESTY_API_KEY"),
@@ -283,6 +285,10 @@ _MODEL_ENTRIES: list[tuple[str, str, str]] = [
     ("qwen3.5-122b", "qwen/qwen3.5-122b-a10b", "openrouter"),
     ("deepseek-v4-pro", "deepseek/deepseek-v4-pro", "openrouter"),
     ("deepseek-v4-flash", "deepseek/deepseek-v4-flash", "openrouter"),
+    # Volcengine Coding Plan (火山引擎代码计划 — coding-only endpoint)
+    # Listed before Zhipu so simple GLM lookups keep their existing default.
+    ("glm-5.2", "glm-5-2", "volcengine-code"),
+    ("kimi-k2.5", "kimi-k2-5", "volcengine-code"),
     # Zhipu CodePlan (智谱代码计划 — coding-only endpoint)
     ("glm-5.2", "glm-5.2", "zhipu-code"),
     ("glm-5.1", "glm-5.1", "zhipu-code"),

tests/test_context_window.py

@@ -202,3 +202,8 @@ def test_lookup_is_case_insensitive():
     # GLM-5 capitalized variant
     glm = SimpleNamespace(model_name="Pro/zai-org/GLM-5", profile=None)
     assert get_context_window(glm) == 203_000
+
+
+def test_volcengine_glm_5_2_alias_keeps_exact_context_window():
+    model = SimpleNamespace(model_name="glm-5-2", profile=None)
+    assert get_context_window(model) == 1_000_000

tests/test_llm.py

@@ -41,6 +41,7 @@ class TestModelsRegistry:
         assert "zhipu" in providers
         assert "zhipu-code" in providers
         assert "volcengine" in providers
+        assert "volcengine-code" in providers
         assert "dashscope" in providers
         assert "dashscope-code" in providers
         assert "deepseek" in providers
@@ -62,6 +63,7 @@ class TestModelsRegistry:
             "zhipu",
             "zhipu-code",
             "volcengine",
+            "volcengine-code",
             "dashscope",
             "dashscope-code",
             "custom-openai",
@@ -1128,6 +1130,28 @@ class TestThirdPartyRouting:
         assert call_kwargs["base_url"] == "https://ark.cn-beijing.volces.com/api/v3"
         assert call_kwargs["api_key"] == "ve-key-123"
 
+    @pytest.mark.parametrize(
+        ("configured_model", "api_model"),
+        [("glm-5.2", "glm-5-2"), ("kimi-k2.5", "kimi-k2-5")],
+    )
+    @patch("EvoScientist.llm.models.init_chat_model")
+    def test_volcengine_code_routes_through_openai(
+        self, mock_init, configured_model, api_model, monkeypatch
+    ):
+        """Volcengine Coding Plan uses its endpoint, IDs, and vendor API key."""
+        mock_init.return_value = "mock_model"
+        monkeypatch.setenv("VOLCENGINE_API_KEY", "ve-code-key-123")
+
+        get_chat_model(configured_model, provider="volcengine-code")
+
+        call_kwargs = mock_init.call_args[1]
+        assert call_kwargs["model_provider"] == "openai"
+        assert call_kwargs["model"] == api_model
+        assert (
+            call_kwargs["base_url"] == "https://ark.cn-beijing.volces.com/api/coding/v3"
+        )
+        assert call_kwargs["api_key"] == "ve-code-key-123"
+
     @patch("EvoScientist.llm.models.init_chat_model")
     def test_dashscope_routes_through_openai(self, mock_init, monkeypatch):
         """DashScope provider should route through OpenAI with correct base_url."""