Add MiMo OpenAI-compatible guidance and LiteLLM fallback pricing hook

Agent-Logs-Url: https://github.com/ZhuLinsen/daily_stock_analysis/sessions/d13f1d78-733b-44f8-bde8-9cac1b6d4b6c

Co-authored-by: ted-davis227 <254451857+ted-davis227@users.noreply.github.com>

Author: Copilot

.env.example

@@ -211,6 +211,15 @@ GEMINI_API_KEY=
 # LLM_MINIMAX_MODELS=MiniMax-M2.7,MiniMax-M2.7-highspeed
 # LITELLM_MODEL=openai/MiniMax-M2.7
 #
+# 小米 MiMo（OpenAI Compatible）
+# Base URL 与模型名请以 MiMo 官方文档/控制台为准。
+# LLM_CHANNELS=mimo
+# LLM_MIMO_PROTOCOL=openai
+# LLM_MIMO_BASE_URL=
+# LLM_MIMO_API_KEY=sk-xxx
+# LLM_MIMO_MODELS=mimo-xxx
+# LITELLM_MODEL=openai/mimo-xxx
+#
 # 火山方舟 / 豆包（OpenAI Compatible，按实际开通地域调整 endpoint）
 # 来源：火山方舟在线推理 / Responses API 文档 https://www.volcengine.com/docs/82379/2121998
 # LiteLLM 通过 openai/<model> 路由到兼容 Chat Completions 的 Base URL。

docs/CHANGELOG.md

@@ -40,6 +40,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/).
 - [修复] 修正分析报告 API 构建策略点位时数值字段未归一为字符串的问题，避免策略价格触发响应 DTO 类型校验失败。
 - [修复] Docker 启动入口自动修复 `data` / `logs` / `reports` 挂载目录权限并降权运行，文档化的 Compose `exec` 手动命令显式使用 `dsa` 用户，避免普通部署需要手动 `chown` / `chmod`。
 - [修复] Web 首页大盘复盘结果改由主内容滚动区承载，避免 loading 切换到长结果后下方报告区域被截断或无法继续滚动。
+- [修复] 为 OpenAI-compatible 未知模型注册兜底计费信息，避免 cost 计算异常影响返回正文，并补充 MiMo 渠道示例配置。
 
 ## [3.16.0] - 2026-05-10
 

docs/llm-providers.md

@@ -70,6 +70,7 @@ OpenAI-compatible Base URL 只填到服务商兼容入口，不额外拼接 `/ch
 | 通义千问 / DashScope | `dashscope` | `openai` | `https://dashscope.aliyuncs.com/compatible-mode/v1` | `qwen3.6-plus,qwen3.6-flash` |
 | 智谱 GLM | `zhipu` | `openai` | `https://open.bigmodel.cn/api/paas/v4` | `glm-5.1,glm-4.7-flash` |
 | MiniMax | `minimax` | `openai` | `https://api.minimax.io/v1` | `MiniMax-M2.7,MiniMax-M2.7-highspeed` |
+| 小米 MiMo | `mimo` | `openai` | 官方控制台提供 | 官方文档/控制台为准 |
 | 火山方舟 / 豆包 | `volcengine` | `openai` | `https://ark.cn-beijing.volces.com/api/v3` | `doubao-seed-1-6-251015,doubao-seed-1-6-thinking-251015` |
 | 硅基流动 / SiliconFlow | `siliconflow` | `openai` | `https://api.siliconflow.cn/v1` | `deepseek-ai/DeepSeek-V3.2,Qwen/Qwen3-235B-A22B-Thinking-2507` |
 | OpenRouter | `openrouter` | `openai` | `https://openrouter.ai/api/v1` | `~anthropic/claude-sonnet-latest,~openai/gpt-latest` |
@@ -88,6 +89,7 @@ OpenAI-compatible Base URL 只填到服务商兼容入口，不额外拼接 `/ch
 | 通义千问 / DashScope | [文本生成](https://help.aliyun.com/zh/model-studio/text-generation-model/) | 百炼推荐 `qwen3.6-plus`，确认效果后可用 `qwen3.6-flash` 降低成本。 |
 | 智谱 GLM | [模型概览](https://docs.bigmodel.cn/cn/guide/start/model-overview)、[GLM-5.1](https://docs.bigmodel.cn/cn/guide/models/text/glm-5.1) | `glm-5.1` 是当前旗舰；`glm-4.7-flash` 作为轻量/免费模型示例。 |
 | MiniMax | [OpenAI API 兼容](https://platform.minimax.io/docs/api-reference/text-chat)、[获取模型列表](https://platform.minimax.io/docs/api-reference/models/openai/list-models) | 官方 OpenAI-compatible Base URL 为 `https://api.minimax.io/v1`，并列出 `MiniMax-M2.7`、`MiniMax-M2.7-highspeed`。中国区 Coding 工具场景可能使用 `.com`/Anthropic 专用入口，以控制台为准。 |
+| 小米 MiMo | 官方文档 / 控制台 | 当前按 OpenAI-compatible 方式接入，Base URL、模型名与权限以 MiMo 官方文档/控制台为准。 |
 | 火山方舟 / 豆包 | [在线推理（常规）](https://www.volcengine.com/docs/82379/2121998)、[模型列表](https://www.volcengine.com/docs/82379/1949118) | 官方示例使用 `https://ark.cn-beijing.volces.com/api/v3` 与 `doubao-seed-1-6-251015`；如使用 Coding Plan，请改用其专用 Base URL 和模型名，不要套用本表的在线推理模板。 |
 | SiliconFlow | [模型列表](https://docs.siliconflow.cn/quickstart/models)、[获取模型列表 API](https://docs.siliconflow.cn/cn/api-reference/models/get-model-list) | 平台模型实时更新且 `/models` 需要 API Key；模板只给常见新模型示例，保存前建议在 Web 设置页点击「获取模型」确认账号可见性。 |
 | OpenRouter | [Models API](https://openrouter.ai/docs/api/api-reference/models/get-models) | OpenRouter 支持 `~anthropic/claude-sonnet-latest`、`~openai/gpt-latest` 等 latest router alias；2026-05-03 的一次手动 live smoke 以 Claude Sonnet latest 作为默认示例通过，GPT latest 保留为可按账号权限切换的备选。 |

src/agent/llm_adapter.py

@@ -11,7 +11,7 @@
 import time
 import uuid
 from dataclasses import dataclass, field
-from typing import Any, Dict, List, Optional
+from typing import Any, Dict, Iterable, List, Optional, Tuple
 
 import litellm
 from litellm import Router
@@ -101,6 +101,28 @@ class LLMResponse:
     },
 }
 
+_FALLBACK_MODEL_PRICING: Dict[str, Any] = {
+    "supports_function_calling": True,
+    "supports_vision": False,
+    "supports_audio_input": False,
+    "supports_audio_output": False,
+    "context_window": 100000,
+    "max_tokens": 10000,
+    "input_cost_per_token": 0.0,
+    "output_cost_per_token": 0.0,
+}
+_FALLBACK_MODEL_PRICING_REGISTERED: set[str] = set()
+
+
+def _split_provider_model(model: str) -> Tuple[str, str]:
+    normalized = (model or "").strip()
+    if not normalized:
+        return "", ""
+    if "/" in normalized:
+        provider, remainder = normalized.split("/", 1)
+        return provider.lower(), remainder.strip()
+    return "openai", normalized
+
 
 def _model_matches(model: str, entries: List[str]) -> bool:
     """Check if model name matches any entry (exact or prefix with version suffix)."""
@@ -176,7 +198,6 @@ def _register_custom_model_pricing() -> None:
                 logger.debug(f"Registered custom pricing for {model_name}")
             except Exception as e:
                 logger.debug(f"Model {model_name} may already be registered or pricing error: {e}")
-
     def _has_channel_config(self) -> bool:
         """Check if multi-channel config (channels / YAML) is active."""
         return bool(self._config.llm_model_list) and not all(
@@ -566,3 +587,30 @@ def _parse_litellm_response(self, response: Any, model: str) -> LLMResponse:
             model=model,
             raw=response,
         )
+
+
+def register_fallback_model_pricing(models: Iterable[str]) -> None:
+    """Register zero-cost pricing for unknown OpenAI-compatible models."""
+    if not models:
+        return
+    LLMToolAdapter._register_custom_model_pricing()
+    register = getattr(litellm, "register_model", None)
+    if not callable(register):
+        return
+    cost_map = getattr(litellm, "model_cost", {})
+    if not isinstance(cost_map, dict):
+        cost_map = {}
+    for model in models:
+        provider, wire_model = _split_provider_model(str(model))
+        if provider != "openai":
+            continue
+        if not wire_model or wire_model.startswith("__legacy_"):
+            continue
+        if wire_model in cost_map or wire_model in _FALLBACK_MODEL_PRICING_REGISTERED:
+            continue
+        try:
+            register({wire_model: dict(_FALLBACK_MODEL_PRICING)})
+            _FALLBACK_MODEL_PRICING_REGISTERED.add(wire_model)
+            logger.debug("Registered fallback pricing for %s", wire_model)
+        except Exception as exc:
+            logger.debug("Fallback pricing registration skipped for %s: %s", wire_model, exc)

src/analyzer.py

@@ -22,7 +22,7 @@
 from json_repair import repair_json
 from litellm import Router
 
-from src.agent.llm_adapter import get_thinking_extra_body
+from src.agent.llm_adapter import get_thinking_extra_body, register_fallback_model_pricing
 from src.agent.skills.defaults import CORE_TRADING_SKILL_POLICY_ZH
 from src.config import (
     Config,
@@ -31,6 +31,7 @@
     get_config,
     get_configured_llm_models,
     normalize_litellm_temperature,
+    resolve_litellm_wire_model,
     resolve_news_window_days,
 )
 from src.storage import persist_llm_usage
@@ -1884,6 +1885,8 @@ def _dispatch_litellm_completion(
         router_model_names: set[str],
     ) -> Any:
         """Dispatch a LiteLLM completion through router or direct fallback."""
+        wire_model = resolve_litellm_wire_model(model, config.llm_model_list)
+        register_fallback_model_pricing([wire_model])
         effective_kwargs = dict(call_kwargs)
         if use_channel_router and self._router and model in router_model_names:
             return self._router.completion(**effective_kwargs)

src/services/system_config_service.py

@@ -724,12 +724,10 @@ def test_llm_channel(
 
         try:
             import litellm
-            from src.agent.llm_adapter import LLMToolAdapter
+            from src.agent.llm_adapter import register_fallback_model_pricing
 
-            # Register custom model pricing for MiniMax models not in LiteLLM's built-in list
-            # This must be done before litellm.completion() to prevent cost calculation errors
-            # Reuses the registration logic from LLMToolAdapter to avoid code duplication
-            LLMToolAdapter._register_custom_model_pricing()
+            # Register fallback pricing for OpenAI-compatible models to prevent cost calculation errors
+            register_fallback_model_pricing([resolved_model])
 
             started_at = time.perf_counter()
             response = litellm.completion(**call_kwargs)

tests/test_litellm_fallback_pricing.py

@@ -0,0 +1,29 @@
+# -*- coding: utf-8 -*-
+"""Tests for fallback LiteLLM pricing registration."""
+
+import unittest
+from unittest.mock import patch
+
+try:
+    import litellm  # noqa: F401
+except ModuleNotFoundError:
+    from tests.litellm_stub import ensure_litellm_stub
+
+    ensure_litellm_stub()
+
+from src.agent import llm_adapter
+
+
+class LiteLLMFallbackPricingTestCase(unittest.TestCase):
+    def test_register_fallback_pricing_registers_openai_model(self) -> None:
+        registered = []
+
+        def _register(payload):
+            registered.append(payload)
+
+        with patch.object(llm_adapter.litellm, "register_model", side_effect=_register, create=True):
+            with patch.object(llm_adapter.litellm, "model_cost", {}, create=True):
+                llm_adapter._FALLBACK_MODEL_PRICING_REGISTERED.clear()
+                llm_adapter.register_fallback_model_pricing(["openai/mimo-alpha"])
+
+        self.assertTrue(any("mimo-alpha" in payload for payload in registered))