feat: config-driven multimodal detection, add_vision_model, streamline model presets

- is_multimodal_model now reads from vision_model.json instead of hardcoded patterns
- Add add_vision_model() with idempotent upsert (default/fallback) and field whitelist
- configure_new_model prompts for multimodal + chains configure_langsmith
- Trim ModelScope presets (14->11), LongCat presets (4->1), update GLM-5 to GLM-5.2
- Comprehensive tests for add_vision_model and multimodal config flow

Author: ScarletMercy

chcode/config.py

@@ -267,12 +267,29 @@ async def configure_new_model() -> dict | None:
     _merge_and_save_config(config)
     console.print(f"[green]模型配置已保存: {config['model']}[/green]")
 
+    # 多模态询问（仅手动配置入口触发；魔搭/LongCat 快捷配置各自已处理视觉）
+    if await confirm("该模型是否为多模态（视觉）模型？", default=False):
+        from chcode.vision_config import add_vision_model
+
+        try:
+            role = add_vision_model(config)
+            if role == "default":
+                console.print(f"[green]已加入视觉模型清单(默认): {config['model']}（图片将直接处理、vision 工具将停用）[/green]")
+            elif role == "fallback":
+                console.print(f"[green]已加入视觉模型清单(备用): {config['model']}（图片将直接处理、vision 工具将停用）[/green]")
+            else:
+                console.print("[yellow]未写入视觉配置（已存在相同配置）[/yellow]")
+        except Exception as e:
+            # 视觉配置失败不阻断主流程（模型本身已保存）
+            console.print(f"[yellow]视觉模型配置失败（不影响主配置）: {e}[/yellow]")
+
     await configure_tavily()
+    await configure_langsmith()
     return config
 
 
 async def _configure_modelscope_with_test() -> dict | None:
-    """魔搭快捷配置：收集 API Key → 测试连接 → 保存 12 个预定义模型。"""
+    """魔搭快捷配置：收集 API Key → 测试连接 → 保存预定义模型。"""
     from chcode.prompts import configure_modelscope
 
     ms_config = await configure_modelscope()
@@ -322,7 +339,7 @@ async def _configure_modelscope_with_test() -> dict | None:
 
 
 async def _configure_longcat_with_test() -> dict | None:
-    """LongCat 快捷配置：收集 API Key → 测试连接 → 保存 4 个预定义模型。"""
+    """LongCat 快捷配置：收集 API Key → 测试连接 → 保存预定义模型。"""
     lc_config = await configure_longcat()
     if lc_config is None:
         return None
@@ -442,19 +459,16 @@ def save_tavily_api_key(api_key: str) -> None:
     "deepseek-v4-pro": 1048576,
     "deepseek-v4-flash": 1048576,
     "glm-5.1": 200000,
+    "glm-5.2": 1048576,
     "glm-5": 200000,
     "glm-4.7": 200000,
     "minimax-m2": 204800,
     "minimax-m2.5": 200000,
     "kimi-k2": 262144,
-    "mimo-v2-flash": 262144,
     "qwen3.5-plus": 1048576,
     "qwen3.6-plus": 1048576,
     "qwen": 256000,
     "longcat-2.0-preview": 1048576,
-    "longcat-flash-chat": 262144,
-    "longcat-flash-thinking": 262144,
-    "longcat-flash-lite": 262144,
 }
 
 _DEFAULT_CONTEXT_WINDOW = 204800

chcode/prompts.py

@@ -109,17 +109,14 @@ async def select_or_custom(
 
 # 每个模型只需声明差异字段，base_url / stream_usage 由生成器统一填充
 _MODELSCOPE_MODELS: list[dict] = [
-    {"model": "ZhipuAI/GLM-5", "temperature": 1.0, "top_p": 0.95},
+    {"model": "ZhipuAI/GLM-5.2", "temperature": 1.0, "top_p": 0.95},
     {"model": "Qwen/Qwen3-235B-A22B-Thinking-2507", "temperature": 0.6, "top_p": 0.95, "extra_body": {"top_k": 20}},
     {"model": "Qwen/Qwen3-235B-A22B-Instruct-2507", "temperature": 0.7, "top_p": 0.8, "extra_body": {"top_k": 20}},
     {"model": "Qwen/Qwen3.5-397B-A17B", "temperature": 0.6, "top_p": 0.95, "extra_body": {"top_k": 20, "repetition_penalty": 1.0}},
     {"model": "deepseek-ai/DeepSeek-V3.2", "temperature": 1.0, "top_p": 0.95},
     {"model": "MiniMax/MiniMax-M2.5", "temperature": 1.0, "top_p": 0.95, "extra_body": {"top_k": 40}},
     {"model": "moonshotai/Kimi-K2.5", "temperature": 1.0, "top_p": 0.95},
     {"model": "ZhipuAI/GLM-5.1", "temperature": 1.0, "top_p": 0.95},
-    {"model": "Qwen/Qwen3-Coder-480B-A35B-Instruct", "temperature": 0.7, "top_p": 0.8, "extra_body": {"top_k": 20, "repetition_penalty": 1.05}},
-    {"model": "XiaomiMiMo/MiMo-V2-Flash", "temperature": 0.3, "top_p": 0.95},
-    {"model": "deepseek-ai/DeepSeek-R1-0528", "temperature": 0.6, "top_p": 0.95},
     {"model": "Qwen/Qwen3-Next-80B-A3B-Thinking", "temperature": 0.6, "top_p": 0.95, "extra_body": {"top_k": 20}},
     {"model": "deepseek-ai/DeepSeek-V4-Pro", "temperature": 1.0, "top_p": 1.0},
     {"model": "deepseek-ai/DeepSeek-V4-Flash", "temperature": 1.0, "top_p": 1.0},
@@ -435,7 +432,7 @@ async def model_config_form(
 
 
 async def configure_modelscope() -> dict | None:
-    """魔搭快捷配置 — 只需 API Key，自动生成 12 个预定义模型。"""
+    """魔搭快捷配置 — 只需 API Key，自动生成预定义模型配置。"""
     # 收集 API Key
     env_choices = [
         f"{var} ({desc})"
@@ -474,32 +471,11 @@ async def configure_modelscope() -> dict | None:
         "top_p": 0.95,
         "stream_usage": True,
     },
-    {
-        "model": "LongCat-Flash-Chat",
-        "base_url": LONGCAT_BASE_URL,
-        "temperature": 1.0,
-        "top_p": 0.95,
-        "stream_usage": True,
-    },
-    {
-        "model": "LongCat-Flash-Thinking",
-        "base_url": LONGCAT_BASE_URL,
-        "temperature": 0.6,
-        "top_p": 0.95,
-        "stream_usage": True,
-    },
-    {
-        "model": "LongCat-Flash-Lite",
-        "base_url": LONGCAT_BASE_URL,
-        "temperature": 1.0,
-        "top_p": 0.95,
-        "stream_usage": True,
-    },
 ]
 
 
 async def configure_longcat() -> dict | None:
-    """LongCat 快捷配置 — 只需 API Key，自动生成 4 个预定义模型。"""
+    """LongCat 快捷配置 — 只需 API Key，自动生成预定义模型配置。"""
     env_choices = [
         f"{var} ({desc})"
         for var, desc in API_KEY_ENV_VARS

chcode/utils/multimodal.py

@@ -17,39 +17,34 @@
 
 from langchain_core.messages import HumanMessage
 
-# ─── Multimodal model patterns ──────────────────────────────────
-
-# Models whose short name (after /) or full name matches these patterns
-# are considered multimodal (native vision capability).
-MULTIMODAL_MODEL_PATTERNS: list[str] = [
-    # Kimi K2.5 series
-    "Kimi-K2",
-    # Qwen3 VL (dedicated vision-language)
-    "Qwen3-VL",
-    # Qwen3.5 MoE models with vision
-    "Qwen3.5-397B",
-    "Qwen3.5-122B",
-    "Qwen3.5-35B",
-    "Qwen3.5-27B",
-    # Intern-S1 series
-    "Intern-S1",
-]
+# ─── Multimodal model detection ────────────────────────────────
 
 
 def is_multimodal_model(model_name: str) -> bool:
     """Check if a model name indicates native multimodal (vision) capability.
 
-    Handles both short names (e.g., "Kimi-K2.5") and full names
-    (e.g., "moonshotai/Kimi-K2.5"). Case-insensitive.
+    判断依据:该模型是否在视觉模型清单(vision_model.json)中。取清单里每个模型的
+    短名(斜杠后部分),用短名子串匹配 model_name 的短名或全名,大小写不敏感。
+    清单为空时返回 False(没有视觉模型配置)。
     """
     if not model_name:
         return False
+    from chcode.vision_config import get_vision_default_model, get_vision_fallback_models
+
+    names: list[str] = []
+    default = get_vision_default_model()
+    if default and default.get("model"):
+        names.append(default["model"])
+    names.extend(m.get("model", "") for m in get_vision_fallback_models())
+    if not names:
+        return False
+
     short_name = model_name.split("/")[-1]
     lower_name = model_name.lower()
     lower_short = short_name.lower()
-    for pattern in MULTIMODAL_MODEL_PATTERNS:
-        lower_pattern = pattern.lower()
-        if lower_pattern in lower_short or lower_pattern in lower_name:
+    for vm in names:
+        vshort = vm.split("/")[-1].lower()
+        if vshort and (vshort in lower_short or vshort in lower_name):
             return True
     return False
 

chcode/vision_config.py

@@ -162,6 +162,73 @@ def auto_configure_vision() -> dict | None:
     return data["default"]
 
 
+# 视觉相关字段白名单（与消费侧 tools.py 视觉请求所读字段对齐）
+_VISION_FIELDS = ("model", "base_url", "api_key", "temperature", "top_p", "stream_usage")
+
+
+def _vision_equal(existing: dict, vision_cfg: dict) -> bool:
+    """现有视觉条目与新配置在白名单字段上是否一致（用于幂等判断）。"""
+    if not isinstance(existing, dict):
+        return False
+    return all(existing.get(k) == vision_cfg.get(k) for k in _VISION_FIELDS)
+
+
+def add_vision_model(config: dict) -> str | None:
+    """把一个模型配置加入视觉模型列表；同名模型就地更新。
+
+    - 同名已在 default → 更新 default
+    - 同名已在 fallback → 更新该 fallback 条目
+    - 新模型：有有效 default（api_key 非空）→ 加入 fallback；否则设为 default
+    - 内容完全一致时跳过写入（幂等）；缺 model/api_key 返回 None。
+
+    Returns: "default" | "fallback" 表示实际落入/更新的角色；None 表示未写入。
+    """
+    model = (config or {}).get("model", "")
+    api_key = (config or {}).get("api_key", "")
+    if not model or not api_key:
+        return None
+
+    # 只保留视觉相关字段，丢弃 extra_body/stop_sequences 等文本侧字段
+    vision_cfg = {k: config[k] for k in _VISION_FIELDS if k in config}
+
+    data = copy.deepcopy(load_vision_json())
+    existing_default = data.get("default", {})
+    existing_fallback: dict = dict(data.get("fallback", {}))
+
+    # 同名已在 default → 就地更新
+    if existing_default.get("model") == model:
+        if _vision_equal(existing_default, vision_cfg):
+            return None
+        data["default"] = vision_cfg
+        data["fallback"] = existing_fallback
+        save_vision_json(data)
+        return "default"
+
+    # 同名已在 fallback → 就地更新
+    if model in existing_fallback:
+        if _vision_equal(existing_fallback[model], vision_cfg):
+            return None
+        existing_fallback[model] = vision_cfg
+        data["default"] = existing_default
+        data["fallback"] = existing_fallback
+        save_vision_json(data)
+        return "fallback"
+
+    # 新模型：有有效 default → 加入 fallback
+    if existing_default and existing_default.get("api_key"):
+        existing_fallback[model] = vision_cfg
+        data["default"] = existing_default
+        data["fallback"] = existing_fallback
+        save_vision_json(data)
+        return "fallback"
+
+    # 无有效 default → 设为 default（保留已存在的 fallback）
+    data["default"] = vision_cfg
+    data["fallback"] = existing_fallback
+    save_vision_json(data)
+    return "default"
+
+
 async def configure_vision_interactive() -> dict | None:
     """交互式配置视觉模型（/vision 命令调用）"""
     ensure_config_dir()