fix(llm): show all enabled chat models in multimodal sidecar selector

The vision/video sidecar dropdown filtered candidates through the built-in
capability heuristics, so provider-compatible models whose custom names match
no known prefix (and carry no declared modalities) were hidden and could not be
selected as a sidecar — even when they natively support the modality.

- listByType now returns every enabled chat model, annotating each row with a
  transient modalityCapable flag and sorting known-capable rows first, instead
  of hard-filtering recognized models only.
- MultimodalRouter honours an explicitly configured sidecar model instead of
  dropping it when the heuristics don't recognize it; a wrong pick degrades
  gracefully through the caption path rather than silently disabling routing.
- ModelPicker gains an optional capability badge; the sidecar UI tags
  recognized vision models while keeping every enabled model selectable.

mateclaw-server/src/main/java/vip/mate/llm/model/ModelConfigEntity.java

@@ -77,6 +77,16 @@ public class ModelConfigEntity {
      */
     private String modalities;
 
+    /**
+     * Transient, request-scoped flag set by {@link vip.mate.llm.service.ModelConfigService#listByType}
+     * when a modality filter is supplied: {@code true} when this row's declared or
+     * heuristically-resolved capabilities cover the requested modality. Lets the
+     * sidecar selector list every enabled chat model while still highlighting the
+     * ones already known to support the modality. Never persisted.
+     */
+    @TableField(exist = false)
+    private Boolean modalityCapable;
+
     @TableField(fill = FieldFill.INSERT)
     private LocalDateTime createTime;
 

mateclaw-server/src/main/java/vip/mate/llm/routing/MultimodalRouter.java

@@ -125,11 +125,19 @@ public class MultimodalRouter {
     }
 
     /**
-     * Resolve the configured sidecar model for a modality. Returns null when:
+     * Resolve the configured sidecar model for a modality. Returns null only when:
      *   - the setting is empty / blank;
-     *   - the referenced row no longer exists or has been disabled;
-     *   - the row's resolved capability set does not actually contain the modality.
+     *   - the referenced row no longer exists or has been disabled.
      * The caller treats null as "ask the user to configure one."
+     * <p>
+     * An explicit sidecar selection is treated as the user's own capability
+     * declaration: a provider-compatible model can be vision-capable in practice
+     * even when the built-in heuristics don't recognize its name and it carries no
+     * declared {@code modalities}. Rejecting such a model here made it impossible to
+     * use a perfectly good compatible-mode vision model as the sidecar. We therefore
+     * honour the explicit choice and only emit a diagnostic when the heuristics
+     * can't confirm it — a wrong pick degrades gracefully (the caption call fails and
+     * the attachment is reported as un-processed) rather than being silently ignored.
      */
     private ModelConfigEntity resolveSidecar(Modality modality) {
         SystemSettingsDTO settings = systemSettingService.getSettings();
@@ -148,9 +156,9 @@ public class MultimodalRouter {
         }
         if (model == null || !Boolean.TRUE.equals(model.getEnabled())) return null;
         if (!capabilityService.supports(model.getModelName(), model.getModalities(), modality)) {
-            log.warn("Configured sidecar model {}/{} does not actually support {} — ignoring",
-                    model.getProvider(), model.getModelName(), modality);
-            return null;
+            log.info("Configured sidecar model {}/{} is not recognized as {}-capable by the "
+                            + "built-in heuristics; honouring the explicit selection anyway",
+                    model.getProvider(), model.getModelName(), modality.name().toLowerCase());
         }
         return model;
     }

mateclaw-server/src/main/java/vip/mate/llm/service/ModelConfigService.java

@@ -11,6 +11,7 @@ import vip.mate.llm.event.ModelConfigChangedEvent;
 import vip.mate.llm.model.ModelConfigEntity;
 import vip.mate.llm.repository.ModelConfigMapper;
 
+import java.util.Comparator;
 import java.util.List;
 import org.springframework.context.ApplicationEventPublisher;
 
@@ -73,9 +74,18 @@ public class ModelConfigService {
 
     /**
      * Optional modality filter (case-insensitive: {@code "vision" / "video" / "audio"}).
-     * When non-null, only enabled rows whose resolved capability set contains the
-     * requested modality survive — used by the multimodal sidecar settings UI to
-     * populate "default vision model" / "default video model" dropdowns.
+     * Used by the multimodal sidecar settings UI to populate "default vision model" /
+     * "default video model" dropdowns.
+     * <p>
+     * The filter does <b>not</b> hide models the built-in heuristics fail to recognize:
+     * a provider-compatible model (e.g. a DashScope OpenAI-compatible vision model with
+     * a custom name) is vision-capable in practice even though its name matches no
+     * built-in prefix and it carries no declared {@code modalities}. Hard-filtering
+     * those out left them un-selectable as a sidecar. Instead every <em>enabled</em>
+     * chat model is returned; each row's transient {@link ModelConfigEntity#getModalityCapable()}
+     * flag records whether its declared / heuristic capabilities already cover the
+     * requested modality, and known-capable rows are sorted to the top so the UI can
+     * highlight them while still letting the user pick any model.
      */
     public List<ModelConfigEntity> listByType(String modelType, String modality) {
         List<ModelConfigEntity> rows;
@@ -100,7 +110,12 @@ public class ModelConfigService {
         }
         return rows.stream()
                 .filter(m -> Boolean.TRUE.equals(m.getEnabled()))
-                .filter(m -> modelCapabilityService.supports(m.getModelName(), m.getModalities(), required))
+                .peek(m -> m.setModalityCapable(
+                        modelCapabilityService.supports(m.getModelName(), m.getModalities(), required)))
+                // Known-capable models first; preserve the existing default-then-name
+                // order within each group.
+                .sorted(Comparator.comparing(
+                        (ModelConfigEntity m) -> Boolean.TRUE.equals(m.getModalityCapable())).reversed())
                 .toList();
     }
 

mateclaw-ui/src/components/common/ModelPicker.vue

@@ -68,6 +68,7 @@
               >
                 <span class="model-picker__item-name">{{ m.modelName }}</span>
                 <span v-if="m.name && m.name !== m.modelName" class="model-picker__item-alias">{{ m.name }}</span>
+                <span v-if="badgeText && m.modalityCapable" class="model-picker__item-badge">{{ badgeText }}</span>
                 <svg
                   v-if="String(m.id) === String(modelValue)"
                   class="model-picker__check"
@@ -97,6 +98,8 @@ interface ModelOption {
   name?: string
   provider: string
   modelName: string
+  /** When true and `badgeText` is set, the item renders a small capability badge. */
+  modalityCapable?: boolean
 }
 
 const props = withDefaults(defineProps<{
@@ -116,6 +119,8 @@ const props = withDefaults(defineProps<{
   clearable?: boolean
   /** Render trigger in disabled state (no popover, no clear). */
   disabled?: boolean
+  /** When set, items whose `modalityCapable` is true show this short badge label. */
+  badgeText?: string
 }>(), {
   searchable: true,
   clearable: true,
@@ -387,6 +392,17 @@ onBeforeUnmount(() => {
   color: var(--mc-text-tertiary);
   font-family: inherit;
 }
+.model-picker__item-badge {
+  flex-shrink: 0;
+  font-size: 10px;
+  font-weight: 600;
+  padding: 1px 6px;
+  border-radius: 999px;
+  font-family: inherit;
+  letter-spacing: 0.02em;
+  background: color-mix(in srgb, var(--mc-primary) 12%, transparent);
+  color: var(--mc-primary);
+}
 .model-picker__check {
   flex-shrink: 0;
   color: var(--mc-primary);

mateclaw-ui/src/i18n/locales/en-US.ts

@@ -622,15 +622,16 @@ export default {
         notConfigured: 'Not configured (skip attachment)',
         idle: 'Off',
         reserved: 'Reserved',
+        capable: 'Vision',
         vision: {
           label: 'Vision sidecar model',
-          desc: 'Invoked once per uploaded image to produce a structured description. The primary chat model is unchanged.',
-          empty: 'No vision-capable model is enabled yet. Add one under "Cloud Models" above.',
+          desc: 'Invoked once per uploaded image to produce a structured description. The primary chat model is unchanged. Any enabled model can be selected; the "Vision" tag marks models already recognized as supporting image input.',
+          empty: 'No chat model is enabled yet. Add and enable one under "Cloud Models" above.',
         },
         video: {
           label: 'Video sidecar model (reserved)',
           desc: 'Reserved for a future iteration: when the user uploads a video, this model will sample frames and describe them. Not yet wired in v1.',
-          empty: 'No video-capable model is enabled yet.',
+          empty: 'No chat model is enabled yet.',
         },
       },
     },

mateclaw-ui/src/i18n/locales/zh-CN.ts

@@ -502,15 +502,16 @@ export default {
         notConfigured: '未配置（不路由附件）',
         idle: '未启用',
         reserved: '预留',
+        capable: '视觉',
         vision: {
           label: '视觉旁路模型',
-          desc: '用户上传图片时调用一次，把图片转成结构化描述。主对话模型不变。',
-          empty: '尚未发现支持图片输入的模型。请到上方"云端模型"中先添加一个视觉模型。',
+          desc: '用户上传图片时调用一次，把图片转成结构化描述。主对话模型不变。可选择任意已启用模型；带"视觉"标记的为已识别支持图片输入的模型。',
+          empty: '尚未发现可用的对话模型。请到上方"云端模型"中先添加并启用一个模型。',
         },
         video: {
           label: '视频旁路模型（预留）',
           desc: '用于未来版本：当用户上传视频时，由该模型负责拆帧和描述。当前版本暂不接入路由。',
-          empty: '尚未发现支持视频输入的模型。',
+          empty: '尚未发现可用的对话模型。',
         },
       },
     },

mateclaw-ui/src/views/Settings/Models/MultimodalSidecarSection.vue

@@ -39,6 +39,7 @@
             :models="visionModels"
             :placeholder="t('settings.models.sidecar.notConfigured')"
             :empty-text="t('settings.models.sidecar.vision.empty')"
+            :badge-text="t('settings.models.sidecar.capable')"
             :disabled="visionModels.length === 0"
           />
         </div>
@@ -75,6 +76,7 @@
             :models="videoModels"
             :placeholder="t('settings.models.sidecar.notConfigured')"
             :empty-text="t('settings.models.sidecar.video.empty')"
+            :badge-text="t('settings.models.sidecar.capable')"
             :disabled="videoModels.length === 0"
           />
         </div>
@@ -102,6 +104,8 @@ interface ModelOption {
   name: string
   provider: string
   modelName: string
+  /** Backend flag: declared/heuristic capabilities already cover this modality. */
+  modalityCapable?: boolean
 }
 
 const { t } = useI18n()