From ec4f13dd79e5a9a6e5d000c9d09d730bc1b632a1 Mon Sep 17 00:00:00 2001 From: jxxghp Date: Fri, 8 May 2026 10:52:30 +0800 Subject: [PATCH] feat(agent): merge MiniMax coding presets --- app/agent/llm/helper.py | 19 +- app/agent/llm/provider.py | 168 ++++++++++++++---- app/api/endpoints/llm.py | 6 + app/core/config.py | 2 + scripts/local_setup.py | 47 +++++ tests/test_llm_helper_testcall.py | 3 + tests/test_llm_provider_registry.py | 48 +++++ tests/test_local_setup_llm_provider_prompt.py | 61 +++++++ tests/test_system_llm_test.py | 9 +- 9 files changed, 328 insertions(+), 35 deletions(-) diff --git a/app/agent/llm/helper.py b/app/agent/llm/helper.py index 68eb2048..82e18fe8 100644 --- a/app/agent/llm/helper.py +++ b/app/agent/llm/helper.py @@ -449,14 +449,22 @@ class LLMHelper: 这主要用于单测 stub 环境以及极端的最小运行环境，正常生产路径仍优先 走 `LLMProviderManager.resolve_runtime()`。 """ + normalized_provider_name = str(provider_name or "").strip().lower() + if normalized_provider_name == "minimax-coding": + normalized_provider_name = "minimax" + api_key_value = api_key if api_key is not None else settings.LLM_API_KEY base_url_value = base_url if base_url is not None else settings.LLM_BASE_URL if not api_key_value: raise ValueError("未配置LLM API Key") - runtime_name = provider_name if provider_name in {"google", "deepseek"} else "openai_compatible" + runtime_name = ( + normalized_provider_name + if normalized_provider_name in {"google", "deepseek"} + else "openai_compatible" + ) return { - "provider_id": provider_name, + "provider_id": normalized_provider_name, "runtime": runtime_name, "model_id": model_name, "api_key": api_key_value, @@ -510,6 +518,7 @@ class LLMHelper: thinking_level: str | None = None, api_key: str | None = settings.LLM_API_KEY, base_url: str | None = settings.LLM_BASE_URL, + base_url_preset: str | None = settings.LLM_BASE_URL_PRESET, ): """ 获取LLM实例 @@ -539,6 +548,7 @@ class LLMHelper: model=model_name, api_key=api_key, base_url=base_url, + base_url_preset_id=base_url_preset, ) except Exception as err: logger.debug(f"LLM provider 目录不可用，回退到旧运行时逻辑: {err}") @@ -700,6 +710,7 @@ class LLMHelper: thinking_level: str | None = None, api_key: str | None = None, base_url: str | None = None, + base_url_preset: str | None = None, ) -> dict: """ 使用当前已保存配置执行一次最小 LLM 调用。 @@ -714,6 +725,7 @@ class LLMHelper: thinking_level=thinking_level, api_key=api_key, base_url=base_url, + base_url_preset=base_url_preset, ) try: response = await asyncio.wait_for(llm.ainvoke(prompt), timeout=timeout) @@ -743,6 +755,7 @@ class LLMHelper: provider: str, api_key: str | None = None, base_url: str | None = None, + base_url_preset: str | None = None, force_refresh: bool = False, ) -> List[dict[str, Any]]: """ @@ -759,6 +772,7 @@ class LLMHelper: provider_id=provider, api_key=api_key, base_url=base_url, + base_url_preset_id=base_url_preset, force_refresh=force_refresh, ) except Exception as err: @@ -776,6 +790,7 @@ class LLMHelper: LLMProviderManager().resolve_model_list_base_url( provider_id=provider, base_url=base_url, + base_url_preset_id=base_url_preset, ) or base_url ) diff --git a/app/agent/llm/provider.py b/app/agent/llm/provider.py index 7e2c6b8a..c92d4d73 100644 --- a/app/agent/llm/provider.py +++ b/app/agent/llm/provider.py @@ -48,6 +48,7 @@ class ProviderAuthMethod: class ProviderUrlPreset: """前端展示用的 Base URL 预设。""" + id: str label: str value: str model_list_base_url: Optional[str] = None @@ -418,11 +419,13 @@ class LLMProviderManager(metaclass=Singleton): models_dev_provider_id="siliconflow-cn", base_url_presets=( url_preset( + id="siliconflow-cn", label="中国大陆", value="https://api.siliconflow.cn/v1", models_dev_provider_id="siliconflow-cn", ), url_preset( + id="siliconflow-global", label="Global", value="https://api.siliconflow.com/v1", models_dev_provider_id="siliconflow", @@ -439,11 +442,13 @@ class LLMProviderManager(metaclass=Singleton): models_dev_provider_id="moonshotai-cn", base_url_presets=( url_preset( + id="moonshot-cn", label="中国站", value="https://api.moonshot.cn/v1", models_dev_provider_id="moonshotai-cn", ), url_preset( + id="moonshot-global", label="国际站", value="https://api.moonshot.ai/v1", models_dev_provider_id="moonshotai", @@ -469,11 +474,13 @@ class LLMProviderManager(metaclass=Singleton): models_dev_provider_id="zhipuai", base_url_presets=( url_preset( + id="zhipu-general", label="Token Plan / 通用 API", value="https://open.bigmodel.cn/api/paas/v4", models_dev_provider_id="zhipuai", ), url_preset( + id="zhipu-coding", label="Coding Plan", value="https://open.bigmodel.cn/api/coding/paas/v4", model_list_base_url="https://open.bigmodel.cn/api/paas/v4", @@ -490,11 +497,13 @@ class LLMProviderManager(metaclass=Singleton): sort_order=66, base_url_presets=( url_preset( + id="zai-general", label="Token Plan / 通用 API", value="https://api.z.ai/api/paas/v4", models_dev_provider_id="zai", ), url_preset( + id="zai-coding", label="Coding Plan", value="https://api.z.ai/api/coding/paas/v4", models_dev_provider_id="zai-coding-plan", @@ -511,22 +520,26 @@ class LLMProviderManager(metaclass=Singleton): models_dev_provider_id="alibaba-cn", base_url_presets=( url_preset( + id="alibaba-cn-general", label="中国内地 / 通用", value="https://dashscope.aliyuncs.com/compatible-mode/v1", models_dev_provider_id="alibaba-cn", ), url_preset( + id="alibaba-global-general", label="国际站 / 通用", value="https://dashscope-intl.aliyuncs.com/compatible-mode/v1", models_dev_provider_id="alibaba", ), url_preset( + id="alibaba-cn-coding", label="中国内地 / Coding Plan", value="https://coding.dashscope.aliyuncs.com/v1", model_list_base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", models_dev_provider_id="alibaba-coding-plan-cn", ), url_preset( + id="alibaba-global-coding", label="国际站 / Coding Plan", value="https://coding-intl.dashscope.aliyuncs.com/v1", model_list_base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1", @@ -543,10 +556,12 @@ class LLMProviderManager(metaclass=Singleton): default_base_url="https://qianfan.baidubce.com/v2", base_url_presets=( url_preset( + id="baidu-qianfan-general", label="通用 API", value="https://qianfan.baidubce.com/v2", ), url_preset( + id="baidu-qianfan-coding", label="Coding Plan", value="https://qianfan.baidubce.com/v2/coding", ), @@ -563,10 +578,12 @@ class LLMProviderManager(metaclass=Singleton): default_base_url="https://modelservice.jdcloud.com/v1", base_url_presets=( url_preset( + id="jdcloud-general", label="通用 API", value="https://modelservice.jdcloud.com/v1", ), url_preset( + id="jdcloud-coding", label="Coding Plan", value="https://modelservice.jdcloud.com/coding/openai/v1", ), @@ -593,11 +610,13 @@ class LLMProviderManager(metaclass=Singleton): default_base_url="https://tokenhub.tencentmaas.com/v1", base_url_presets=( url_preset( + id="tencent-tokenhub", label="TokenHub", value="https://tokenhub.tencentmaas.com/v1", models_dev_provider_id="tencent-tokenhub", ), url_preset( + id="tencent-coding", label="Coding Plan", value="https://api.lkeap.cloud.tencent.com/coding/v3", models_dev_provider_id="tencent-coding-plan", @@ -635,11 +654,13 @@ class LLMProviderManager(metaclass=Singleton): sort_order=115, base_url_presets=( url_preset( + id="opencode-zen", label="Zen", value="https://opencode.ai/zen/v1", models_dev_provider_id="opencode", ), url_preset( + id="opencode-go", label="Go", value="https://opencode.ai/zen/go/v1", models_dev_provider_id="opencode-go", @@ -656,39 +677,32 @@ class LLMProviderManager(metaclass=Singleton): models_dev_provider_id="minimax-cn", base_url_presets=( url_preset( + id="minimax-cn-general", label="中国内地 / 通用", value="https://api.minimaxi.com/anthropic/v1", models_dev_provider_id="minimax-cn", ), url_preset( + id="minimax-global-general", label="国际站 / 通用", value="https://api.minimax.io/anthropic/v1", models_dev_provider_id="minimax", ), - ), - api_key_hint="填写 MiniMax API Key，可在中国内地与国际站通用端点间切换。", - description="MiniMax Anthropic-compatible 通用端点。", - ), - anthropic_provider( - provider_id="minimax-coding", - name="MiniMax Coding Plan", - default_base_url="https://api.minimaxi.com/anthropic/v1", - sort_order=121, - models_dev_provider_id="minimax-cn-coding-plan", - base_url_presets=( url_preset( + id="minimax-cn-coding", label="中国内地 / Coding Plan", value="https://api.minimaxi.com/anthropic/v1", models_dev_provider_id="minimax-cn-coding-plan", ), url_preset( + id="minimax-global-coding", label="国际站 / Coding Plan", value="https://api.minimax.io/anthropic/v1", models_dev_provider_id="minimax-coding-plan", ), ), - api_key_hint="填写 MiniMax API Key，可在中国内地与国际站 Coding Plan 目录间切换。", - description="MiniMax Coding Plan Anthropic-compatible 端点。", + api_key_hint="填写 MiniMax API Key，可在中国内地、国际站、通用与 Coding Plan 目录间切换。", + description="MiniMax Anthropic-compatible 端点，支持通用与 Coding Plan 目录预设。", ), catalog_openai_provider( provider_id="xiaomi", @@ -697,21 +711,25 @@ class LLMProviderManager(metaclass=Singleton): sort_order=130, base_url_presets=( url_preset( + id="xiaomi-standard", label="标准端点", value="https://api.xiaomimimo.com/v1", models_dev_provider_id="xiaomi", ), url_preset( + id="xiaomi-token-plan-cn", label="Token Plan / 中国", value="https://token-plan-cn.xiaomimimo.com/v1", models_dev_provider_id="xiaomi-token-plan-cn", ), url_preset( + id="xiaomi-token-plan-sgp", label="Token Plan / 新加坡", value="https://token-plan-sgp.xiaomimimo.com/v1", models_dev_provider_id="xiaomi-token-plan-sgp", ), url_preset( + id="xiaomi-token-plan-ams", label="Token Plan / 欧洲", value="https://token-plan-ams.xiaomimimo.com/v1", models_dev_provider_id="xiaomi-token-plan-ams", @@ -932,11 +950,12 @@ class LLMProviderManager(metaclass=Singleton): async def _get_provider_async( self, provider_id: str, force_refresh: bool = False ) -> ProviderSpec: + normalized_provider_id = self._normalize_provider_id(provider_id) try: - return self.get_provider(provider_id) + return self.get_provider(normalized_provider_id) except LLMProviderError: await self.get_models_dev_data(force_refresh=force_refresh) - return self.get_provider(provider_id) + return self.get_provider(normalized_provider_id) def _serialize_provider(self, spec: ProviderSpec) -> dict[str, Any]: return { @@ -946,6 +965,7 @@ class LLMProviderManager(metaclass=Singleton): "default_base_url": self._default_base_url_for_provider(spec) or "", "base_url_presets": [ { + "id": preset.id, "label": preset.label, "value": self._sanitize_base_url(preset.value) or "", } @@ -989,7 +1009,7 @@ class LLMProviderManager(metaclass=Singleton): def get_provider(self, provider_id: str) -> ProviderSpec: """按 provider id 获取定义。""" - normalized = (provider_id or "").strip().lower() + normalized = self._normalize_provider_id(provider_id) for spec in self._provider_specs(): if spec.id == normalized: return spec @@ -1014,9 +1034,39 @@ class LLMProviderManager(metaclass=Singleton): return cls._sanitize_base_url(spec.base_url_presets[0].value) @classmethod - def _resolve_provider_model_list_base_url( - cls, spec: ProviderSpec, base_url: Optional[str] + def _normalize_provider_id(cls, provider_id: str) -> str: + normalized = (provider_id or "").strip().lower() + if normalized == "minimax-coding": + return "minimax" + return normalized + + @classmethod + def _normalize_base_url_preset_id( + cls, provider_id: str, base_url_preset_id: Optional[str] ) -> Optional[str]: + normalized_provider_id = cls._normalize_provider_id(provider_id) + normalized_preset_id = str(base_url_preset_id or "").strip().lower() or None + if not normalized_preset_id: + return None + if normalized_provider_id == "minimax" and normalized_preset_id == "minimax-coding": + return "minimax-cn-coding" + return normalized_preset_id + + @classmethod + def _resolve_provider_model_list_base_url( + cls, + spec: ProviderSpec, + base_url: Optional[str], + base_url_preset_id: Optional[str] = None, + ) -> Optional[str]: + normalized_preset_id = cls._normalize_base_url_preset_id(spec.id, base_url_preset_id) + if normalized_preset_id: + for preset in spec.base_url_presets: + if preset.id != normalized_preset_id: + continue + preset_value = cls._sanitize_base_url(preset.value) + return cls._sanitize_base_url(preset.model_list_base_url) or preset_value + normalized_base_url = cls._sanitize_base_url(base_url) if normalized_base_url: for preset in spec.base_url_presets: @@ -1037,8 +1087,18 @@ class LLMProviderManager(metaclass=Singleton): @classmethod def _resolve_provider_models_dev_provider_id( - cls, spec: ProviderSpec, base_url: Optional[str] + cls, + spec: ProviderSpec, + base_url: Optional[str], + base_url_preset_id: Optional[str] = None, ) -> Optional[str]: + normalized_preset_id = cls._normalize_base_url_preset_id(spec.id, base_url_preset_id) + if normalized_preset_id: + for preset in spec.base_url_presets: + if preset.id != normalized_preset_id: + continue + return preset.models_dev_provider_id or spec.models_dev_provider_id + normalized_base_url = cls._sanitize_base_url(base_url) if normalized_base_url: for preset in spec.base_url_presets: @@ -1058,10 +1118,17 @@ class LLMProviderManager(metaclass=Singleton): return spec.models_dev_provider_id def resolve_model_list_base_url( - self, provider_id: str, base_url: Optional[str] + self, + provider_id: str, + base_url: Optional[str], + base_url_preset_id: Optional[str] = None, ) -> Optional[str]: spec = self.get_provider(provider_id) - return self._resolve_provider_model_list_base_url(spec, base_url) + return self._resolve_provider_model_list_base_url( + spec, + base_url, + base_url_preset_id=base_url_preset_id, + ) @staticmethod def _httpx_proxy_key() -> str: @@ -1212,21 +1279,33 @@ class LLMProviderManager(metaclass=Singleton): raise LLMProviderError(f"获取 models.dev 数据失败: {err}") from err async def _models_dev_provider_payload( - self, provider_id: str, base_url: Optional[str] = None + self, + provider_id: str, + base_url: Optional[str] = None, + base_url_preset_id: Optional[str] = None, ) -> dict[str, Any]: spec = await self._get_provider_async(provider_id) models_dev_provider_id = self._resolve_provider_models_dev_provider_id( spec, base_url, + base_url_preset_id=base_url_preset_id, ) if not models_dev_provider_id: return {} return (await self.get_models_dev_data()).get(models_dev_provider_id, {}) or {} async def _models_dev_model( - self, provider_id: str, model_id: str, base_url: Optional[str] = None + self, + provider_id: str, + model_id: str, + base_url: Optional[str] = None, + base_url_preset_id: Optional[str] = None, ) -> dict[str, Any] | None: - payload = await self._models_dev_provider_payload(provider_id, base_url=base_url) + payload = await self._models_dev_provider_payload( + provider_id, + base_url=base_url, + base_url_preset_id=base_url_preset_id, + ) models = payload.get("models") if isinstance(payload, dict) else None if not isinstance(models, dict): return None @@ -1394,13 +1473,18 @@ class LLMProviderManager(metaclass=Singleton): provider_id: str, transport: str = "openai", base_url: Optional[str] = None, + base_url_preset_id: Optional[str] = None, ) -> list[dict[str, Any]]: """ 某些 provider 没有统一稳定的 models.list 行为， 因此优先读取 models.dev 目录；若未来 provider 暴露标准 models 接口， 再平滑补充实时刷新即可。 """ - payload = await self._models_dev_provider_payload(provider_id, base_url=base_url) + payload = await self._models_dev_provider_payload( + provider_id, + base_url=base_url, + base_url_preset_id=base_url_preset_id, + ) models = payload.get("models") if isinstance(payload, dict) else None if not isinstance(models, dict): raise LLMProviderError(f"{provider_id} 暂无可用模型目录") @@ -1551,11 +1635,16 @@ class LLMProviderManager(metaclass=Singleton): provider_id: str, api_key: Optional[str] = None, base_url: Optional[str] = None, + base_url_preset_id: Optional[str] = None, force_refresh: bool = False, ) -> list[dict[str, Any]]: """返回标准化后的模型目录。""" spec = await self._get_provider_async(provider_id, force_refresh=force_refresh) - if self._resolve_provider_models_dev_provider_id(spec, base_url): + if self._resolve_provider_models_dev_provider_id( + spec, + base_url, + base_url_preset_id=base_url_preset_id, + ): # 对依赖 models.dev 的 provider 主动刷新一次缓存，保证“刷新模型列表” # 在使用目录型 provider 时也能拿到最新参数。 if force_refresh: @@ -1565,6 +1654,7 @@ class LLMProviderManager(metaclass=Singleton): model=None, api_key=api_key, base_url=base_url, + base_url_preset_id=base_url_preset_id, ) if spec.model_list_strategy == "google": @@ -1582,6 +1672,7 @@ class LLMProviderManager(metaclass=Singleton): base_url=self._resolve_provider_model_list_base_url( spec, runtime["base_url"], + base_url_preset_id=base_url_preset_id, ), default_headers=runtime.get("default_headers"), ) @@ -1591,6 +1682,7 @@ class LLMProviderManager(metaclass=Singleton): provider_id=provider_id, transport="anthropic", base_url=base_url, + base_url_preset_id=base_url_preset_id, ) if spec.model_list_strategy == "models_dev_only": @@ -1598,6 +1690,7 @@ class LLMProviderManager(metaclass=Singleton): provider_id=provider_id, transport="openai", base_url=base_url, + base_url_preset_id=base_url_preset_id, ) # openai-compatible / deepseek 默认走官方 models 端点。 @@ -1607,6 +1700,7 @@ class LLMProviderManager(metaclass=Singleton): base_url=self._resolve_provider_model_list_base_url( spec, runtime["base_url"], + base_url_preset_id=base_url_preset_id, ), default_headers=runtime.get("default_headers"), ) @@ -1616,6 +1710,7 @@ class LLMProviderManager(metaclass=Singleton): provider_id: str, model_id: Optional[str], base_url: Optional[str] = None, + base_url_preset_id: Optional[str] = None, ) -> dict[str, Any] | None: if not model_id: return None @@ -1623,6 +1718,7 @@ class LLMProviderManager(metaclass=Singleton): provider_id, model_id, base_url=base_url, + base_url_preset_id=base_url_preset_id, ) if metadata: return metadata @@ -2079,13 +2175,19 @@ class LLMProviderManager(metaclass=Singleton): model: Optional[str], api_key: Optional[str] = None, base_url: Optional[str] = None, + base_url_preset_id: Optional[str] = None, ) -> dict[str, Any]: """ 解析 provider 运行时参数。 返回统一结构，供 `LLMHelper` 创建具体 LangChain 模型实例时使用。 """ - spec = await self._get_provider_async(provider_id) + normalized_provider_id = self._normalize_provider_id(provider_id) + normalized_base_url_preset_id = self._normalize_base_url_preset_id( + normalized_provider_id, + base_url_preset_id, + ) + spec = await self._get_provider_async(normalized_provider_id) normalized_api_key = str(api_key or "").strip() or None normalized_base_url = self._sanitize_base_url(base_url) model_record = None @@ -2095,9 +2197,10 @@ class LLMProviderManager(metaclass=Singleton): ( item for item in await self.list_models( - provider_id, + normalized_provider_id, api_key=api_key, base_url=base_url, + base_url_preset_id=normalized_base_url_preset_id, ) if item["id"] == model ), @@ -2108,21 +2211,22 @@ class LLMProviderManager(metaclass=Singleton): model_record = None result: dict[str, Any] = { - "provider_id": provider_id, + "provider_id": normalized_provider_id, "runtime": spec.runtime, "model_id": model, "model_record": model_record, "model_metadata": await self.resolve_model_metadata( - provider_id, + normalized_provider_id, model, base_url=base_url, + base_url_preset_id=normalized_base_url_preset_id, ), "default_headers": None, "use_responses_api": None, "auth_mode": "api_key", } - if provider_id == "chatgpt": + if normalized_provider_id == "chatgpt": auth = None try: auth = await self._resolve_chatgpt_oauth() @@ -2160,7 +2264,7 @@ class LLMProviderManager(metaclass=Singleton): raise LLMProviderAuthError("请提供 API Key 或完成 ChatGPT 授权") - if provider_id == "github-copilot": + if normalized_provider_id == "github-copilot": auth = self.get_saved_auth("github-copilot") if auth and auth.get("type") == "oauth": token = auth.get("refresh_token") or auth.get("access_token") diff --git a/app/api/endpoints/llm.py b/app/api/endpoints/llm.py index cd0cc525..e2763492 100644 --- a/app/api/endpoints/llm.py +++ b/app/api/endpoints/llm.py @@ -12,6 +12,7 @@ from app.agent.llm import ( LLMTestTimeout, render_auth_result_html, ) +from app.core.config import settings from app.db.models import User from app.db.user_oper import ( get_current_active_superuser_async, @@ -29,6 +30,7 @@ class LlmTestRequest(BaseModel): thinking_level: Optional[str] = None api_key: Optional[str] = None base_url: Optional[str] = None + base_url_preset: Optional[str] = None class LlmProviderAuthStartRequest(BaseModel): @@ -64,6 +66,7 @@ async def get_llm_models( provider: str, api_key: Optional[str] = None, base_url: Optional[str] = None, + base_url_preset: Optional[str] = None, force_refresh: Optional[bool] = False, _: User = Depends(get_current_active_user_async), ): @@ -76,6 +79,7 @@ async def get_llm_models( provider=provider, api_key=api_key, base_url=base_url, + base_url_preset=base_url_preset, force_refresh=bool(force_refresh), ) return schemas.Response( @@ -231,6 +235,7 @@ async def llm_test( thinking_level=settings.LLM_THINKING_LEVEL, api_key=settings.LLM_API_KEY, base_url=settings.LLM_BASE_URL, + base_url_preset=settings.LLM_BASE_URL_PRESET, ) if not payload.provider: @@ -262,6 +267,7 @@ async def llm_test( thinking_level=payload.thinking_level, api_key=payload.api_key, base_url=payload.base_url, + base_url_preset=payload.base_url_preset, ) if not result.get("reply_preview"): return schemas.Response( diff --git a/app/core/config.py b/app/core/config.py index 2d2f22e6..24316400 100644 --- a/app/core/config.py +++ b/app/core/config.py @@ -515,6 +515,8 @@ class ConfigModel(BaseModel): LLM_API_KEY: Optional[str] = None # LLM基础URL（用于自定义API端点） LLM_BASE_URL: Optional[str] = "https://api.deepseek.com" + # LLM Base URL 预设标识，用于区分同一 Base URL 下的不同模型目录 + LLM_BASE_URL_PRESET: Optional[str] = None # LLM最大上下文Token数量（K） LLM_MAX_CONTEXT_TOKENS: int = 64 # LLM温度参数 diff --git a/scripts/local_setup.py b/scripts/local_setup.py index 727a5fe6..281a05b4 100644 --- a/scripts/local_setup.py +++ b/scripts/local_setup.py @@ -1214,9 +1214,15 @@ def _llm_provider_defaults( default_base_url = str(provider_meta.get("default_base_url") or "").strip() if default_base_url: defaults["base_url"] = default_base_url + base_url_presets = provider_meta.get("base_url_presets") or [] + if isinstance(base_url_presets, list) and base_url_presets: + preset_id = str((base_url_presets[0] or {}).get("id") or "").strip() + if preset_id: + defaults["base_url_preset"] = preset_id defaults.setdefault("model", _env_default("LLM_MODEL", "")) defaults.setdefault("base_url", _env_default("LLM_BASE_URL", "")) + defaults.setdefault("base_url_preset", _env_default("LLM_BASE_URL_PRESET", "")) return defaults @@ -1245,11 +1251,13 @@ def _load_llm_models_inner(payload: dict[str, Any]) -> list[dict[str, Any]]: provider_module = _load_llm_provider_module() api_key = str(payload.get("api_key") or "").strip() or None base_url = str(payload.get("base_url") or "").strip() or None + base_url_preset = str(payload.get("base_url_preset") or "").strip() or None models = asyncio.run( provider_module.LLMProviderManager().list_models( provider_id=provider, api_key=api_key, base_url=base_url, + base_url_preset_id=base_url_preset, force_refresh=False, ) ) @@ -1261,12 +1269,14 @@ def _load_llm_models( provider: str, api_key: Optional[str], base_url: Optional[str], + base_url_preset: Optional[str], runtime_python: Optional[Path] = None, ) -> list[dict[str, Any]]: payload = { "provider": str(provider or "").strip().lower(), "api_key": str(api_key or "").strip(), "base_url": str(base_url or "").strip(), + "base_url_preset": str(base_url_preset or "").strip(), } try: return _load_llm_models_inner(payload) @@ -1795,6 +1805,9 @@ def _collect_agent_config( defaults = _llm_provider_defaults(provider, provider_definitions) current_model = _env_default("LLM_MODEL", defaults["model"]) current_base_url = _env_default("LLM_BASE_URL", defaults["base_url"]) + current_base_url_preset = _env_default( + "LLM_BASE_URL_PRESET", defaults.get("base_url_preset", "") + ) api_key_label = str(provider_meta.get("api_key_label") or "API Key").strip() or "API Key" api_key_hint = str(provider_meta.get("api_key_hint") or "").strip() requires_base_url = bool(provider_meta.get("requires_base_url")) @@ -1838,8 +1851,41 @@ def _collect_agent_config( "是否启用图片输入支持", default=_env_bool("LLM_SUPPORT_IMAGE_INPUT", True), ), + "LLM_BASE_URL_PRESET": current_base_url_preset, } + base_url_presets = provider_meta.get("base_url_presets") or [] + if isinstance(base_url_presets, list): + duplicate_value_presets = [] + normalized_current_base_url = current_base_url.strip() + for item in base_url_presets: + if not isinstance(item, dict): + continue + preset_value = str(item.get("value") or "").strip() + preset_id = str(item.get("id") or "").strip() + if not preset_id or preset_value != normalized_current_base_url: + continue + duplicate_value_presets.append(item) + + if len(duplicate_value_presets) > 1: + choices: dict[str, str] = {} + default_preset = current_base_url_preset + if not default_preset or default_preset not in { + str(item.get("id") or "").strip() for item in duplicate_value_presets + }: + default_preset = str((duplicate_value_presets[0] or {}).get("id") or "").strip() + for item in duplicate_value_presets: + preset_id = str(item.get("id") or "").strip() + preset_label = str(item.get("label") or preset_id).strip() + if preset_id: + choices[preset_id] = preset_label + if choices: + config["LLM_BASE_URL_PRESET"] = _prompt_choice( + "LLM Base URL 预设", + choices=choices, + default=default_preset, + ) + config["LLM_BASE_URL"] = _prompt_text( base_url_label, default=current_base_url, @@ -1849,6 +1895,7 @@ def _collect_agent_config( provider=provider, api_key=config["LLM_API_KEY"], base_url=config["LLM_BASE_URL"], + base_url_preset=config["LLM_BASE_URL_PRESET"], runtime_python=runtime_python, ) config["LLM_MODEL"] = _prompt_model_choice(models, default=current_model) diff --git a/tests/test_llm_helper_testcall.py b/tests/test_llm_helper_testcall.py index 4747d77d..d31fb544 100644 --- a/tests/test_llm_helper_testcall.py +++ b/tests/test_llm_helper_testcall.py @@ -38,6 +38,7 @@ _stub_module( LLM_MODEL="global-model", LLM_API_KEY="global-key", LLM_BASE_URL="https://global.example.com", + LLM_BASE_URL_PRESET=None, LLM_THINKING_LEVEL=None, LLM_TEMPERATURE=0.1, LLM_MAX_CONTEXT_TOKENS=64, @@ -76,6 +77,7 @@ class LlmHelperTestCallTest(unittest.TestCase): model="deepseek-chat", api_key="sk-test", base_url="https://api.deepseek.com", + base_url_preset="deepseek-default", ) ) @@ -86,6 +88,7 @@ class LlmHelperTestCallTest(unittest.TestCase): thinking_level=None, api_key="sk-test", base_url="https://api.deepseek.com", + base_url_preset="deepseek-default", ) self.assertEqual(result["provider"], "deepseek") self.assertEqual(result["model"], "deepseek-chat") diff --git a/tests/test_llm_provider_registry.py b/tests/test_llm_provider_registry.py index 026dff1d..69d90174 100644 --- a/tests/test_llm_provider_registry.py +++ b/tests/test_llm_provider_registry.py @@ -246,6 +246,54 @@ class LlmProviderRegistryTest(unittest.TestCase): self.assertIsNone(provider.models_dev_provider_id) self.assertFalse(provider.supports_model_refresh) + def test_builtin_minimax_provider_merges_general_and_coding_presets(self): + manager = LLMProviderManager() + + provider = manager.get_provider("minimax") + serialized = manager.list_providers() + minimax_payload = next(item for item in serialized if item["id"] == "minimax") + + self.assertEqual(provider.name, "MiniMax") + self.assertEqual(provider.runtime, "anthropic_compatible") + self.assertEqual( + tuple((preset.id, preset.label, preset.value) for preset in provider.base_url_presets), + ( + ("minimax-cn-general", "中国内地 / 通用", "https://api.minimaxi.com/anthropic/v1"), + ("minimax-global-general", "国际站 / 通用", "https://api.minimax.io/anthropic/v1"), + ("minimax-cn-coding", "中国内地 / Coding Plan", "https://api.minimaxi.com/anthropic/v1"), + ("minimax-global-coding", "国际站 / Coding Plan", "https://api.minimax.io/anthropic/v1"), + ), + ) + self.assertEqual( + tuple(item["id"] for item in minimax_payload["base_url_presets"]), + ( + "minimax-cn-general", + "minimax-global-general", + "minimax-cn-coding", + "minimax-global-coding", + ), + ) + + def test_minimax_coding_alias_resolves_to_minimax_provider(self): + manager = LLMProviderManager() + + provider = manager.get_provider("minimax-coding") + + self.assertEqual(provider.id, "minimax") + + def test_resolve_models_dev_provider_id_prefers_minimax_preset_id(self): + manager = LLMProviderManager() + provider = manager.get_provider("minimax") + + self.assertEqual( + manager._resolve_provider_models_dev_provider_id( + provider, + base_url="https://api.minimaxi.com/anthropic/v1", + base_url_preset_id="minimax-cn-coding", + ), + "minimax-cn-coding-plan", + ) + if __name__ == "__main__": unittest.main() diff --git a/tests/test_local_setup_llm_provider_prompt.py b/tests/test_local_setup_llm_provider_prompt.py index 21ae491c..8481dba1 100644 --- a/tests/test_local_setup_llm_provider_prompt.py +++ b/tests/test_local_setup_llm_provider_prompt.py @@ -68,6 +68,7 @@ class LocalSetupLlmProviderPromptTests(unittest.TestCase): provider="frogbot", api_key="sk-frog", base_url="https://override.example.com/v1", + base_url_preset="", runtime_python=Path("/tmp/runtime-python"), ) model_prompt.assert_called_once_with(models, default="") @@ -205,6 +206,66 @@ class LocalSetupLlmProviderPromptTests(unittest.TestCase): "https://modelservice.jdcloud.com/v1", ) + def test_collect_agent_config_prompts_for_duplicate_base_url_presets(self): + module = load_local_setup_module() + + provider_definitions = [ + { + "id": "minimax", + "name": "MiniMax", + "default_base_url": "https://api.minimaxi.com/anthropic/v1", + "api_key_label": "API Key", + "base_url_presets": [ + { + "id": "minimax-cn-general", + "label": "中国内地 / 通用", + "value": "https://api.minimaxi.com/anthropic/v1", + }, + { + "id": "minimax-cn-coding", + "label": "中国内地 / Coding Plan", + "value": "https://api.minimaxi.com/anthropic/v1", + }, + ], + } + ] + + with patch.object(module, "print_step"), patch.object( + module, "_prompt_yes_no", side_effect=[True, False, True] + ), patch.object( + module, "_load_llm_provider_definitions", return_value=provider_definitions + ), patch.object( + module, "_prompt_provider_choice", return_value="minimax" + ), patch.object( + module, "_prompt_text", side_effect=["https://api.minimaxi.com/anthropic/v1"] + ), patch.object( + module, "_prompt_secret_text", return_value="sk-minimax" + ), patch.object( + module, "_load_llm_models", return_value=[] + ) as load_models, patch.object( + module, "_prompt_model_choice", return_value="MiniMax-M1" + ), patch.object( + module, "read_env_value", return_value=None + ), patch.object( + module, "_env_default", side_effect=lambda key, default="": default + ), patch.object( + module, "_env_bool", side_effect=lambda key, default: default + ), patch.object( + module, "_env_llm_thinking_level_default", return_value="auto" + ), patch.object( + module, "_prompt_choice", side_effect=["auto", "minimax-cn-coding"] + ): + config = module._collect_agent_config() + + self.assertEqual(config["LLM_BASE_URL_PRESET"], "minimax-cn-coding") + load_models.assert_called_once_with( + provider="minimax", + api_key="sk-minimax", + base_url="https://api.minimaxi.com/anthropic/v1", + base_url_preset="minimax-cn-coding", + runtime_python=None, + ) + if __name__ == "__main__": unittest.main() diff --git a/tests/test_system_llm_test.py b/tests/test_system_llm_test.py index b9866d8b..7f75aa3f 100644 --- a/tests/test_system_llm_test.py +++ b/tests/test_system_llm_test.py @@ -77,7 +77,7 @@ _stub_module("app.utils.crypto", HashUtils=_Dummy) _stub_module("app.utils.http", RequestUtils=_Dummy, AsyncRequestUtils=_Dummy) _stub_module("version", APP_VERSION="test") -from app.api.endpoints import system as system_endpoint +from app.api.endpoints import llm as system_endpoint class LlmTestEndpointTest(unittest.TestCase): @@ -124,6 +124,8 @@ class LlmTestEndpointTest(unittest.TestCase): system_endpoint.settings, "LLM_API_KEY", "sk-test" ), patch.object( system_endpoint.settings, "LLM_BASE_URL", "https://api.deepseek.com" + ), patch.object( + system_endpoint.settings, "LLM_BASE_URL_PRESET", "deepseek-default" ), patch.object( system_endpoint.LLMHelper, "test_current_settings", @@ -138,6 +140,7 @@ class LlmTestEndpointTest(unittest.TestCase): thinking_level="max", api_key="sk-test", base_url="https://api.deepseek.com", + base_url_preset="deepseek-default", ) self.assertTrue(resp.success) self.assertEqual(resp.data["provider"], "deepseek") @@ -161,6 +164,7 @@ class LlmTestEndpointTest(unittest.TestCase): thinking_level="high", api_key="sk-live", base_url="https://example.com/v1", + base_url_preset="openai-default", ) with patch.object(system_endpoint.settings, "AI_AGENT_ENABLE", False), patch.object( @@ -183,6 +187,7 @@ class LlmTestEndpointTest(unittest.TestCase): thinking_level="high", api_key="sk-live", base_url="https://example.com/v1", + base_url_preset="openai-default", ) self.assertTrue(resp.success) self.assertEqual(resp.data["provider"], "openai") @@ -203,6 +208,7 @@ class LlmTestEndpointTest(unittest.TestCase): model="deepseek-v4-pro", api_key="sk-live", base_url="https://api.deepseek.com", + base_url_preset="deepseek-default", ) with patch.object(system_endpoint.settings, "AI_AGENT_ENABLE", False), patch.object( @@ -219,6 +225,7 @@ class LlmTestEndpointTest(unittest.TestCase): thinking_level=None, api_key="sk-live", base_url="https://api.deepseek.com", + base_url_preset="deepseek-default", ) self.assertTrue(resp.success)