fix(story-summary): polish vector api UX and lint

modules/story-summary/data/config.js

@@ -254,6 +254,50 @@ export const DEFAULT_SUMMARY_USER_CONFIRM_PROMPT = `怎么截断了！重新完
 </Chat_History>`;
 
 export const DEFAULT_SUMMARY_ASSISTANT_PREFILL_PROMPT = '下面重新生成完整JSON。';
+const DEFAULT_VECTOR_PROVIDER = "siliconflow";
+const DEFAULT_L0_URL = "https://api.siliconflow.cn/v1";
+const DEFAULT_OPENROUTER_URL = "https://openrouter.ai/api/v1";
+const DEFAULT_L0_MODEL = "Qwen/Qwen3-8B";
+const DEFAULT_EMBEDDING_MODEL = "BAAI/bge-m3";
+const DEFAULT_RERANK_MODEL = "BAAI/bge-reranker-v2-m3";
+
+function getVectorProviderDefaultUrl(provider) {
+    return provider === "openrouter" ? DEFAULT_OPENROUTER_URL : DEFAULT_L0_URL;
+}
+
+function createDefaultProviderProfile(provider, model = "") {
+    return {
+        url: provider === "custom" ? "" : getVectorProviderDefaultUrl(provider),
+        key: "",
+        model: model || "",
+        modelCache: [],
+    };
+}
+
+function normalizeProviderProfiles(supportedProviders, srcProfiles, currentProvider, currentValues, defaultModel) {
+    const out = {};
+    supportedProviders.forEach((provider) => {
+        const raw = srcProfiles?.[provider] || {};
+        const defaults = createDefaultProviderProfile(provider, defaultModel);
+        out[provider] = {
+            url: String(raw.url || defaults.url || "").trim(),
+            key: String(raw.key || "").trim(),
+            model: String(raw.model || defaults.model || "").trim(),
+            modelCache: Array.isArray(raw.modelCache) ? raw.modelCache.filter(Boolean) : [],
+        };
+    });
+
+    if (currentProvider && out[currentProvider]) {
+        if (currentValues?.url && !out[currentProvider].url) out[currentProvider].url = String(currentValues.url).trim();
+        if (currentValues?.key && !out[currentProvider].key) out[currentProvider].key = String(currentValues.key).trim();
+        if (currentValues?.model && !out[currentProvider].model) out[currentProvider].model = String(currentValues.model).trim();
+        if (Array.isArray(currentValues?.modelCache) && !out[currentProvider].modelCache.length) {
+            out[currentProvider].modelCache = currentValues.modelCache.filter(Boolean);
+        }
+    }
+
+    return out;
+}
 
 export function getSettings() {
     const ext = (extension_settings[EXT_ID] ||= {});
@@ -261,6 +305,63 @@ export function getSettings() {
     return ext;
 }
 
+function normalizeOpenAiCompatApiConfig(src, defaults = {}) {
+    const provider = String(src?.provider || defaults.provider || DEFAULT_VECTOR_PROVIDER).toLowerCase();
+    const supportedProviders = Array.isArray(defaults.supportedProviders) && defaults.supportedProviders.length
+        ? defaults.supportedProviders
+        : [provider, "custom"];
+    const providers = normalizeProviderProfiles(
+        supportedProviders,
+        src?.providers,
+        provider,
+        src,
+        defaults.model || ""
+    );
+    const current = providers[provider] || createDefaultProviderProfile(provider, defaults.model || "");
+    return {
+        provider,
+        url: String(current.url || "").trim(),
+        key: String(current.key || defaults.key || "").trim(),
+        model: String(current.model || defaults.model || "").trim(),
+        modelCache: Array.isArray(current.modelCache) ? current.modelCache.filter(Boolean) : [],
+        providers,
+    };
+}
+
+function normalizeVectorConfig(rawVector = null) {
+    const legacyOnline = rawVector?.online || {};
+    const sharedProvider = String(legacyOnline.provider || DEFAULT_VECTOR_PROVIDER).toLowerCase();
+    const sharedUrl = String(legacyOnline.url || (sharedProvider === "openrouter" ? DEFAULT_OPENROUTER_URL : DEFAULT_L0_URL)).trim();
+    const sharedKey = String(legacyOnline.key || "").trim();
+
+    return {
+        enabled: !!rawVector?.enabled,
+        engine: "online",
+        l0Concurrency: Math.max(1, Math.min(50, Number(rawVector?.l0Concurrency) || 10)),
+        l0Api: normalizeOpenAiCompatApiConfig(rawVector?.l0Api, {
+            provider: sharedProvider,
+            url: sharedUrl,
+            key: sharedKey,
+            model: DEFAULT_L0_MODEL,
+            supportedProviders: ["siliconflow", "openrouter", "custom"],
+        }),
+        embeddingApi: normalizeOpenAiCompatApiConfig(rawVector?.embeddingApi, {
+            provider: DEFAULT_VECTOR_PROVIDER,
+            url: DEFAULT_L0_URL,
+            key: sharedKey,
+            model: DEFAULT_EMBEDDING_MODEL,
+            supportedProviders: ["siliconflow", "custom"],
+        }),
+        rerankApi: normalizeOpenAiCompatApiConfig(rawVector?.rerankApi, {
+            provider: DEFAULT_VECTOR_PROVIDER,
+            url: DEFAULT_L0_URL,
+            key: sharedKey,
+            model: DEFAULT_RERANK_MODEL,
+            supportedProviders: ["siliconflow", "custom"],
+        }),
+    };
+}
+
 export function getSummaryPanelConfig() {
     const clampKeepVisibleCount = (value) => {
         const n = Number.parseInt(value, 10);
@@ -299,7 +400,7 @@ export function getSummaryPanelConfig() {
             summaryAssistantPrefillPrompt: DEFAULT_SUMMARY_ASSISTANT_PREFILL_PROMPT,
             memoryTemplate: DEFAULT_MEMORY_PROMPT_TEMPLATE,
         },
-        vector: null,
+        vector: normalizeVectorConfig(),
     };
 
     try {
@@ -320,7 +421,7 @@ export function getSummaryPanelConfig() {
             ui: { ...defaults.ui, ...(parsed.ui || {}) },
             textFilterRules,
             prompts: { ...defaults.prompts, ...(parsed.prompts || {}) },
-            vector: parsed.vector || null,
+            vector: normalizeVectorConfig(parsed.vector || null),
         };
 
         if (result.trigger.timing === "manual") result.trigger.enabled = false;
@@ -349,16 +450,7 @@ export function getVectorConfig() {
         if (!raw) return null;
 
         const parsed = JSON.parse(raw);
-        const cfg = parsed.vector || null;
+        return parsed.vector ? normalizeVectorConfig(parsed.vector) : normalizeVectorConfig();
-        if (!cfg) return null;
-
-        // Keep vector side normalized to online + siliconflow.
-        cfg.engine = "online";
-        cfg.online = cfg.online || {};
-        cfg.online.provider = "siliconflow";
-        cfg.online.model = "BAAI/bge-m3";
-
-        return cfg;
     } catch {
         return null;
     }
@@ -376,15 +468,7 @@ export function saveVectorConfig(vectorCfg) {
         const raw = localStorage.getItem("summary_panel_config") || "{}";
         const parsed = JSON.parse(raw);
 
-        parsed.vector = {
+        parsed.vector = normalizeVectorConfig(vectorCfg || null);
-            enabled: !!vectorCfg?.enabled,
-            engine: "online",
-            online: {
-                provider: "siliconflow",
-                key: vectorCfg?.online?.key || "",
-                model: "BAAI/bge-m3",
-            },
-        };
 
         localStorage.setItem("summary_panel_config", JSON.stringify(parsed));
         CommonSettingStorage.set(SUMMARY_CONFIG_KEY, parsed);

modules/story-summary/generate/llm.js

@@ -26,18 +26,6 @@ const PROVIDER_MAP = {
 
 const JSON_PREFILL = DEFAULT_SUMMARY_ASSISTANT_PREFILL_PROMPT;
 
-const LLM_PROMPT_CONFIG = {
-    topSystem: DEFAULT_SUMMARY_SYSTEM_PROMPT,
-    assistantDoc: DEFAULT_SUMMARY_ASSISTANT_DOC_PROMPT,
-    assistantAskSummary: DEFAULT_SUMMARY_ASSISTANT_ASK_SUMMARY_PROMPT,
-    assistantAskContent: DEFAULT_SUMMARY_ASSISTANT_ASK_CONTENT_PROMPT,
-    metaProtocolStart: DEFAULT_SUMMARY_META_PROTOCOL_START_PROMPT,
-    userJsonFormat: DEFAULT_SUMMARY_USER_JSON_FORMAT_PROMPT,
-    assistantCheck: DEFAULT_SUMMARY_ASSISTANT_CHECK_PROMPT,
-    userConfirm: DEFAULT_SUMMARY_USER_CONFIRM_PROMPT,
-    assistantPrefill: DEFAULT_SUMMARY_ASSISTANT_PREFILL_PROMPT,
-};
-
 // ═══════════════════════════════════════════════════════════════════════════
 // 工具函数
 // ═══════════════════════════════════════════════════════════════════════════

modules/story-summary/generate/prompt.js

@@ -806,11 +806,7 @@ function buildNonVectorPrompt(store) {
 
     if (!sections.length) return "";
 
-    return (
+    return buildMemoryPromptText(sections.join("\n\n"));
-        `${buildSystemPreamble()}\n` +
-        `<剧情记忆>\n\n${sections.join("\n\n")}\n\n</剧情记忆>\n` +
-        `${buildPostscript()}`
-    );
 }
 
 /**

modules/story-summary/story-summary-ui.js

@@ -297,6 +297,73 @@ All checks passed. Beginning incremental extraction...
         claude: { url: 'https://api.anthropic.com', needKey: true, canFetch: false },
         custom: { url: '', needKey: true, canFetch: true }
     };
+    const VECTOR_PROVIDER_DEFAULTS = {
+        siliconflow: { url: 'https://api.siliconflow.cn/v1', needKey: true, canFetch: true },
+        openrouter: { url: 'https://openrouter.ai/api/v1', needKey: true, canFetch: true },
+        custom: { url: '', needKey: true, canFetch: true }
+    };
+
+    const VECTOR_API_SUPPORTED_PROVIDERS = {
+        l0: ['siliconflow', 'openrouter', 'custom'],
+        embedding: ['siliconflow', 'custom'],
+        rerank: ['siliconflow', 'custom'],
+    };
+
+    const VECTOR_API_DEFAULT_MODELS = {
+        l0: 'Qwen/Qwen3-8B',
+        embedding: 'BAAI/bge-m3',
+        rerank: 'BAAI/bge-reranker-v2-m3',
+    };
+
+    function setStatusText(el, message, kind = '') {
+        if (!el) return;
+        el.textContent = message || '';
+        el.style.color = kind === 'error'
+            ? '#ef4444'
+            : kind === 'success'
+                ? '#22c55e'
+                : kind === 'loading'
+                    ? '#f59e0b'
+                    : '';
+    }
+
+    function createDefaultProviderProfile(provider, model = '') {
+        const pv = VECTOR_PROVIDER_DEFAULTS[provider] || VECTOR_PROVIDER_DEFAULTS.custom;
+        return {
+            url: pv.url || '',
+            key: '',
+            model: model || '',
+            modelCache: [],
+        };
+    }
+
+    function normalizeProviderProfiles(prefix, apiCfg = {}) {
+        const supported = VECTOR_API_SUPPORTED_PROVIDERS[prefix] || ['custom'];
+        const model = apiCfg.model || VECTOR_API_DEFAULT_MODELS[prefix] || '';
+        const out = {};
+        supported.forEach(provider => {
+            const raw = apiCfg.providers?.[provider] || {};
+            const defaults = createDefaultProviderProfile(provider, model);
+            out[provider] = {
+                url: String(raw.url || defaults.url || '').trim(),
+                key: String(raw.key || '').trim(),
+                model: String(raw.model || defaults.model || '').trim(),
+                modelCache: Array.isArray(raw.modelCache) ? raw.modelCache.filter(Boolean) : [],
+            };
+        });
+
+        const currentProvider = String(apiCfg.provider || supported[0] || 'custom').toLowerCase();
+        if (out[currentProvider]) {
+            if (apiCfg.url && !out[currentProvider].url) out[currentProvider].url = String(apiCfg.url).trim();
+            if (apiCfg.key && !out[currentProvider].key) out[currentProvider].key = String(apiCfg.key).trim();
+            if (apiCfg.model && !out[currentProvider].model) out[currentProvider].model = String(apiCfg.model).trim();
+            if (Array.isArray(apiCfg.modelCache) && !out[currentProvider].modelCache.length) {
+                out[currentProvider].modelCache = apiCfg.modelCache.filter(Boolean);
+            }
+        }
+
+        return out;
+    }
 
     const SECTION_META = {
         keywords: { title: '编辑关键词', hint: '每行一个关键词，格式：关键词|权重（核心/重要/一般）' },
@@ -344,7 +411,33 @@ All checks passed. Beginning incremental extraction...
             memoryTemplate: '',
         },
         textFilterRules: [...DEFAULT_FILTER_RULES],
-        vector: { enabled: false, engine: 'online', local: { modelId: 'bge-small-zh' }, online: { provider: 'siliconflow', url: '', key: '', model: '' } }
+        vector: {
+            enabled: false,
+            engine: 'online',
+            l0Concurrency: 10,
+            l0Api: {
+                provider: 'siliconflow', url: 'https://api.siliconflow.cn/v1', key: '', model: 'Qwen/Qwen3-8B', modelCache: [],
+                providers: {
+                    siliconflow: createDefaultProviderProfile('siliconflow', 'Qwen/Qwen3-8B'),
+                    openrouter: createDefaultProviderProfile('openrouter', 'Qwen/Qwen3-8B'),
+                    custom: createDefaultProviderProfile('custom', 'Qwen/Qwen3-8B'),
+                }
+            },
+            embeddingApi: {
+                provider: 'siliconflow', url: 'https://api.siliconflow.cn/v1', key: '', model: 'BAAI/bge-m3', modelCache: [],
+                providers: {
+                    siliconflow: createDefaultProviderProfile('siliconflow', 'BAAI/bge-m3'),
+                    custom: createDefaultProviderProfile('custom', 'BAAI/bge-m3'),
+                }
+            },
+            rerankApi: {
+                provider: 'siliconflow', url: 'https://api.siliconflow.cn/v1', key: '', model: 'BAAI/bge-reranker-v2-m3', modelCache: [],
+                providers: {
+                    siliconflow: createDefaultProviderProfile('siliconflow', 'BAAI/bge-reranker-v2-m3'),
+                    custom: createDefaultProviderProfile('custom', 'BAAI/bge-reranker-v2-m3'),
+                }
+            }
+        }
     };
 
     let summaryData = { keywords: [], events: [], characters: { main: [], relationships: [] }, arcs: [], facts: [] };
@@ -369,6 +462,45 @@ All checks passed. Beginning incremental extraction...
         window.parent.postMessage({ source: 'LittleWhiteBox-StoryFrame', type, ...data }, PARENT_ORIGIN);
     }
 
+    function normalizeVectorConfigUI(raw = null) {
+        const base = JSON.parse(JSON.stringify(config.vector));
+        const legacyOnline = raw?.online || {};
+        const sharedKey = String(legacyOnline.key || '').trim();
+        const sharedUrl = String(legacyOnline.url || '').trim();
+
+        if (raw) {
+            base.enabled = !!raw.enabled;
+            base.engine = 'online';
+            base.l0Concurrency = Math.max(1, Math.min(50, Number(raw.l0Concurrency) || 10));
+            Object.assign(base.l0Api, {
+                provider: raw.l0Api?.provider || legacyOnline.provider || base.l0Api.provider,
+                url: raw.l0Api?.url || sharedUrl || base.l0Api.url,
+                key: raw.l0Api?.key || sharedKey || base.l0Api.key,
+                model: raw.l0Api?.model || base.l0Api.model,
+                modelCache: Array.isArray(raw.l0Api?.modelCache) ? raw.l0Api.modelCache : [],
+                providers: normalizeProviderProfiles('l0', raw.l0Api || {}),
+            });
+            Object.assign(base.embeddingApi, {
+                provider: raw.embeddingApi?.provider || base.embeddingApi.provider,
+                url: raw.embeddingApi?.url || sharedUrl || base.embeddingApi.url,
+                key: raw.embeddingApi?.key || sharedKey || base.embeddingApi.key,
+                model: raw.embeddingApi?.model || legacyOnline.model || base.embeddingApi.model,
+                modelCache: Array.isArray(raw.embeddingApi?.modelCache) ? raw.embeddingApi.modelCache : [],
+                providers: normalizeProviderProfiles('embedding', raw.embeddingApi || {}),
+            });
+            Object.assign(base.rerankApi, {
+                provider: raw.rerankApi?.provider || base.rerankApi.provider,
+                url: raw.rerankApi?.url || sharedUrl || base.rerankApi.url,
+                key: raw.rerankApi?.key || sharedKey || base.rerankApi.key,
+                model: raw.rerankApi?.model || base.rerankApi.model,
+                modelCache: Array.isArray(raw.rerankApi?.modelCache) ? raw.rerankApi.modelCache : [],
+                providers: normalizeProviderProfiles('rerank', raw.rerankApi || {}),
+            });
+        }
+
+        return base;
+    }
+
     // ═══════════════════════════════════════════════════════════════════════════
     // Config Management
     // ═══════════════════════════════════════════════════════════════════════════
@@ -387,7 +519,7 @@ All checks passed. Beginning incremental extraction...
                 config.textFilterRules = Array.isArray(p.textFilterRules)
                     ? p.textFilterRules
                     : (Array.isArray(p.vector?.textFilterRules) ? p.vector.textFilterRules : [...DEFAULT_FILTER_RULES]);
-                if (p.vector) config.vector = p.vector;
+                if (p.vector) config.vector = normalizeVectorConfigUI(p.vector);
                 if (config.trigger.timing === 'manual' && config.trigger.enabled) {
                     config.trigger.enabled = false;
                     saveConfig();
@@ -409,7 +541,7 @@ All checks passed. Beginning incremental extraction...
             : (Array.isArray(cfg.vector?.textFilterRules)
                 ? cfg.vector.textFilterRules
                 : (Array.isArray(config.textFilterRules) ? config.textFilterRules : [...DEFAULT_FILTER_RULES]));
-        if (cfg.vector) config.vector = cfg.vector;
+        if (cfg.vector) config.vector = normalizeVectorConfigUI(cfg.vector);
         if (config.trigger.timing === 'manual') config.trigger.enabled = false;
         localStorage.setItem('summary_panel_config', JSON.stringify(config));
     }
@@ -422,7 +554,14 @@ All checks passed. Beginning incremental extraction...
                 config.textFilterRules = collectFilterRules();
             }
             if (!config.vector) {
-                config.vector = { enabled: false, engine: 'online', online: { provider: 'siliconflow', key: '', model: 'BAAI/bge-m3' } };
+                config.vector = {
+                    enabled: false,
+                    engine: 'online',
+                    l0Concurrency: 10,
+                    l0Api: { provider: 'siliconflow', url: 'https://api.siliconflow.cn/v1', key: '', model: 'Qwen/Qwen3-8B', modelCache: [] },
+                    embeddingApi: { provider: 'siliconflow', url: 'https://api.siliconflow.cn/v1', key: '', model: 'BAAI/bge-m3', modelCache: [] },
+                    rerankApi: { provider: 'siliconflow', url: 'https://api.siliconflow.cn/v1', key: '', model: 'BAAI/bge-reranker-v2-m3', modelCache: [] }
+                };
             }
             localStorage.setItem('summary_panel_config', JSON.stringify(config));
             postMsg('SAVE_PANEL_CONFIG', { config });
@@ -435,15 +574,160 @@ All checks passed. Beginning incremental extraction...
     // Vector Config UI
     // ═══════════════════════════════════════════════════════════════════════════
 
+    function getVectorApiConfig(prefix) {
+        const provider = $(`${prefix}-api-provider`)?.value || 'siliconflow';
+        const providers = normalizeProviderProfiles(prefix, config.vector?.[`${prefix}Api`] || {});
+        providers[provider] = {
+            url: $(`${prefix}-api-url`)?.value?.trim() || '',
+            key: $(`${prefix}-api-key`)?.value?.trim() || '',
+            model: $(`${prefix}-api-model-text`)?.value?.trim() || '',
+            modelCache: Array.isArray(config.vector?.[`${prefix}Api`]?.providers?.[provider]?.modelCache)
+                ? [...config.vector[`${prefix}Api`].providers[provider].modelCache]
+                : [],
+        };
+        return {
+            provider,
+            url: providers[provider]?.url || '',
+            key: providers[provider]?.key || '',
+            model: providers[provider]?.model || '',
+            modelCache: Array.isArray(providers[provider]?.modelCache) ? [...providers[provider].modelCache] : [],
+            providers,
+        };
+    }
+
+    function loadVectorApiConfig(prefix, cfg) {
+        const next = cfg || {};
+        const provider = next.provider || 'siliconflow';
+        const profiles = normalizeProviderProfiles(prefix, next);
+        const profile = profiles[provider] || createDefaultProviderProfile(provider, VECTOR_API_DEFAULT_MODELS[prefix]);
+        $(`${prefix}-api-provider`).value = provider;
+        $(`${prefix}-api-url`).value = profile.url || '';
+        $(`${prefix}-api-key`).value = profile.key || '';
+        $(`${prefix}-api-model-text`).value = profile.model || '';
+
+        const cache = Array.isArray(profile.modelCache) ? profile.modelCache : [];
+        setSelectOptions($(`${prefix}-api-model-select`), cache, '请选择');
+        $(`${prefix}-api-model-select`).value = cache.includes(profile.model) ? profile.model : '';
+        updateVectorProviderUI(prefix, provider);
+    }
+
+    function saveCurrentVectorApiProfile(prefix, providerOverride = null) {
+        const apiCfg = config.vector[`${prefix}Api`] ||= {};
+        const provider = providerOverride || $(`${prefix}-api-provider`)?.value || apiCfg.provider || 'siliconflow';
+        apiCfg.providers = normalizeProviderProfiles(prefix, apiCfg);
+        apiCfg.providers[provider] = {
+            url: $(`${prefix}-api-url`)?.value?.trim() || '',
+            key: $(`${prefix}-api-key`)?.value?.trim() || '',
+            model: $(`${prefix}-api-model-text`)?.value?.trim() || '',
+            modelCache: Array.isArray(apiCfg.providers?.[provider]?.modelCache) ? [...apiCfg.providers[provider].modelCache] : [],
+        };
+        apiCfg.provider = provider;
+        apiCfg.url = apiCfg.providers[provider].url;
+        apiCfg.key = apiCfg.providers[provider].key;
+        apiCfg.model = apiCfg.providers[provider].model;
+        apiCfg.modelCache = [...apiCfg.providers[provider].modelCache];
+    }
+
+    function updateVectorProviderUI(prefix, provider) {
+        const pv = VECTOR_PROVIDER_DEFAULTS[provider] || VECTOR_PROVIDER_DEFAULTS.custom;
+        const apiCfg = config.vector?.[`${prefix}Api`] || {};
+        apiCfg.providers = normalizeProviderProfiles(prefix, apiCfg);
+        const profile = apiCfg.providers[provider] || createDefaultProviderProfile(provider, VECTOR_API_DEFAULT_MODELS[prefix]);
+        const cache = Array.isArray(profile.modelCache) ? profile.modelCache : [];
+        const hasModelCache = cache.length > 0;
+
+        $(`${prefix}-api-url-row`).classList.toggle('hidden', false);
+        $(`${prefix}-api-key-row`).classList.toggle('hidden', !pv.needKey);
+        $(`${prefix}-api-model-manual-row`).classList.toggle('hidden', false);
+        $(`${prefix}-api-model-select-row`).classList.toggle('hidden', !hasModelCache);
+        $(`${prefix}-api-connect-row`).classList.toggle('hidden', !pv.canFetch);
+        $(`${prefix}-api-connect-status`).classList.toggle('hidden', !pv.canFetch);
+
+        const urlInput = $(`${prefix}-api-url`);
+        if (urlInput) {
+            if (provider === 'custom') {
+                urlInput.readOnly = false;
+                urlInput.placeholder = 'https://your-openai-compatible-api/v1';
+                urlInput.value = profile.url || '';
+            } else {
+                urlInput.value = pv.url || '';
+                urlInput.readOnly = true;
+                urlInput.placeholder = pv.url || '';
+            }
+        }
+        $(`${prefix}-api-key`).value = profile.key || '';
+        $(`${prefix}-api-model-text`).value = profile.model || '';
+        setSelectOptions($(`${prefix}-api-model-select`), cache, '请选择');
+        $(`${prefix}-api-model-select`).value = cache.includes(profile.model) ? profile.model : '';
+    }
+
+    function saveVectorApiSection(prefix) {
+        saveCurrentVectorApiProfile(prefix);
+        saveConfig();
+        setStatusText($(`${prefix}-api-connect-status`), '此组配置已保存', 'success');
+    }
+
+    async function fetchVectorModels(prefix) {
+        const provider = $(`${prefix}-api-provider`).value;
+        const pv = VECTOR_PROVIDER_DEFAULTS[provider] || VECTOR_PROVIDER_DEFAULTS.custom;
+        const statusEl = $(`${prefix}-api-connect-status`);
+        const btn = $(`${prefix}-btn-connect`);
+        if (!pv.canFetch) {
+            statusEl.textContent = '当前渠道不支持自动拉取模型';
+            return;
+        }
+
+        let baseUrl = $(`${prefix}-api-url`).value.trim().replace(/\/+$/, '');
+        const apiKey = $(`${prefix}-api-key`).value.trim();
+        if (!apiKey) {
+            statusEl.textContent = '请先填写 API KEY';
+            return;
+        }
+
+        btn.disabled = true;
+        btn.textContent = '连接中...';
+        statusEl.textContent = '连接中...';
+
+        try {
+            const tryFetch = async url => {
+                const res = await fetch(url, {
+                    headers: { Authorization: `Bearer ${apiKey}`, Accept: 'application/json' }
+                });
+                return res.ok ? (await res.json())?.data?.map(m => m?.id).filter(Boolean) || null : null;
+            };
+
+            if (baseUrl.endsWith('/v1')) baseUrl = baseUrl.slice(0, -3);
+            let models = await tryFetch(`${baseUrl}/v1/models`);
+            if (!models) models = await tryFetch(`${baseUrl}/models`);
+            if (!models?.length) throw new Error('未获取到模型列表');
+
+            const apiCfg = config.vector[`${prefix}Api`] ||= {};
+            apiCfg.providers = normalizeProviderProfiles(prefix, apiCfg);
+            apiCfg.providers[provider] ||= createDefaultProviderProfile(provider, VECTOR_API_DEFAULT_MODELS[prefix]);
+            apiCfg.providers[provider].modelCache = [...new Set(models)];
+            setSelectOptions($(`${prefix}-api-model-select`), apiCfg.providers[provider].modelCache, '请选择');
+            $(`${prefix}-api-model-select-row`).classList.remove('hidden');
+            if (!$(`${prefix}-api-model-text`).value.trim()) {
+                $(`${prefix}-api-model-text`).value = models[0];
+                $(`${prefix}-api-model-select`).value = models[0];
+            }
+            statusEl.textContent = `拉取成功：${models.length} 个模型`;
+        } catch (e) {
+            statusEl.textContent = '拉取失败：' + (e.message || '请检查 URL 和 KEY');
+        } finally {
+            btn.disabled = false;
+            btn.textContent = '连接 / 拉取模型列表';
+        }
+    }
+
     function getVectorConfig() {
         return {
             enabled: $('vector-enabled')?.checked || false,
             engine: 'online',
-            online: {
+            l0Concurrency: Math.max(1, Math.min(50, Number($('vector-l0-concurrency')?.value) || 10)),
-                provider: 'siliconflow',
+            l0Api: getVectorApiConfig('l0'),
-                key: $('vector-api-key')?.value?.trim() || '',
+            embeddingApi: getVectorApiConfig('embedding'),
-                model: 'BAAI/bge-m3',
+            rerankApi: getVectorApiConfig('rerank'),
-            },
         };
     }
 
@@ -451,11 +735,10 @@ All checks passed. Beginning incremental extraction...
         if (!cfg) return;
         $('vector-enabled').checked = !!cfg.enabled;
         $('vector-config-area').classList.toggle('hidden', !cfg.enabled);
-
+        $('vector-l0-concurrency').value = String(Math.max(1, Math.min(50, Number(cfg.l0Concurrency) || 10)));
-        if (cfg.online?.key) {
+        loadVectorApiConfig('l0', cfg.l0Api || {});
-            $('vector-api-key').value = cfg.online.key;
+        loadVectorApiConfig('embedding', cfg.embeddingApi || {});
-        }
+        loadVectorApiConfig('rerank', cfg.rerankApi || {});
-
     }
 
     // ═══════════════════════════════════════════════════════════════════════════
@@ -536,13 +819,6 @@ All checks passed. Beginning incremental extraction...
         el.textContent = count;
     }
 
-    function updateOnlineStatus(status, message) {
-        const dot = $('online-api-status').querySelector('.status-dot');
-        const text = $('online-api-status').querySelector('.status-text');
-        dot.className = 'status-dot ' + status;
-        text.textContent = message;
-    }
-
     function updateVectorStats(stats) {
         $('vector-atom-count').textContent = stats.stateVectors || 0;
         $('vector-chunk-count').textContent = stats.chunkCount || 0;
@@ -646,16 +922,45 @@ All checks passed. Beginning incremental extraction...
             $('vector-config-area').classList.toggle('hidden', !e.target.checked);
         };
 
-        $('btn-test-vector-api').onclick = () => {
+        ['l0', 'embedding', 'rerank'].forEach(prefix => {
-            saveConfig(); // 先保存新 Key 到 localStorage
+            $(`${prefix}-api-key-toggle`).onclick = () => {
-            postMsg('VECTOR_TEST_ONLINE', {
+                const input = $(`${prefix}-api-key`);
-                provider: 'siliconflow',
+                const btn = $(`${prefix}-api-key-toggle`);
-                config: {
+                if (!input || !btn) return;
-                    key: $('vector-api-key').value.trim(),
+                const show = input.type === 'password';
-                    model: 'BAAI/bge-m3',
+                input.type = show ? 'text' : 'password';
-                }
+                btn.textContent = show ? '隐藏' : '显示';
-            });
+            };
-        };
+
+            $(`${prefix}-api-provider`).onchange = e => {
+                const target = config.vector[`${prefix}Api`] ||= {};
+                const previousProvider = target.provider || 'siliconflow';
+                saveCurrentVectorApiProfile(prefix, previousProvider);
+                target.providers = normalizeProviderProfiles(prefix, target);
+                target.provider = e.target.value;
+                target.providers[e.target.value] ||= createDefaultProviderProfile(e.target.value, VECTOR_API_DEFAULT_MODELS[prefix]);
+                updateVectorProviderUI(prefix, e.target.value);
+            };
+
+            $(`${prefix}-api-model-select`).onchange = e => {
+                if (e.target.value) $(`${prefix}-api-model-text`).value = e.target.value;
+            };
+
+            $(`${prefix}-btn-connect`).onclick = () => fetchVectorModels(prefix);
+            $(`${prefix}-btn-save`).onclick = () => saveVectorApiSection(prefix);
+            $(`${prefix}-btn-test`).onclick = () => {
+                const btn = $(`${prefix}-btn-test`);
+                if (btn) btn.disabled = true;
+                setStatusText($(`${prefix}-api-connect-status`), '测试中...', 'loading');
+                saveConfig();
+                const cfg = getVectorConfig();
+                postMsg('VECTOR_TEST_ONLINE', {
+                    target: prefix,
+                    provider: cfg[`${prefix}Api`].provider,
+                    config: cfg[`${prefix}Api`],
+                });
+            };
+        });
 
         $('btn-add-filter-rule').onclick = addFilterRule;
 
@@ -701,6 +1006,17 @@ All checks passed. Beginning incremental extraction...
         postMsg('REQUEST_ANCHOR_STATS');
     }
 
+    function updateVectorOnlineStatus(target, status, message) {
+        const prefix = target || 'embedding';
+        const btn = $(`${prefix}-btn-test`);
+        if (btn) btn.disabled = false;
+        setStatusText(
+            $(`${prefix}-api-connect-status`),
+            message || '',
+            status === 'error' ? 'error' : status === 'success' ? 'success' : 'loading'
+        );
+    }
+
     function initSummaryIOUI() {
         $('btn-copy-summary').onclick = () => {
             $('btn-copy-summary').disabled = true;
@@ -1803,7 +2119,7 @@ All checks passed. Beginning incremental extraction...
                 break;
 
             case 'VECTOR_ONLINE_STATUS':
-                updateOnlineStatus(d.status, d.message);
+                updateVectorOnlineStatus(d.target, d.status, d.message);
                 break;
 
             case 'VECTOR_STATS':

modules/story-summary/story-summary.html

@@ -414,31 +414,180 @@
 
                         <div id="vector-config-area" class="hidden">
 
-                            <!-- Step 0: API Key -->
+                            <!-- Step 0: API Config -->
                             <div class="neo-card" style="margin-top: 24px;">
                                 <div class="neo-card-title">
                                     <span class="neo-badge">Step.0</span>
-                                    <span>填写 API Key</span>
+                                    <span>配置模型 API</span>
                                 </div>
                                 <div class="settings-row">
                                     <div class="settings-field full">
-                                        <label>硅基流动 API Key</label>
+                                        <div class="settings-hint" style="margin-bottom:12px;">
-                                        <input type="password" id="vector-api-key" placeholder="sk-xxx">
+                                            推荐给 L0 使用便宜或免费的高并发渠道。<br>
-                                        <div class="settings-hint">
+                                            自定义渠道请填写 OpenAI 兼容基础地址（通常为 /v1）。Embedding 会自动补 <code>/embeddings</code>，Rerank 会自动补 <code>/rerank</code>。
-                                            💡 <a href="https://siliconflow.cn" target="_blank">硅基流动</a>
-                                            内置使用的模型完全免费。建议实名认证以获得更高并发。
                                         </div>
-                                    </div>
+                                        <details class="settings-collapse" id="l0-api-collapse" style="margin-bottom:12px;">
-                                </div>
+                                            <summary class="settings-collapse-header">
-                                <div class="settings-row">
+                                                <span>L0 锚点提取模型</span>
-                                    <div class="settings-field full">
+                                                <svg class="collapse-icon" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
-                                        <div class="engine-status-row" style="margin-top: 0;">
+                                                    <polyline points="6 9 12 15 18 9"></polyline>
-                                            <div class="engine-status" id="online-api-status">
+                                                </svg>
-                                                <span class="status-dot"></span>
+                                            </summary>
-                                                <span class="status-text">未测试</span>
+                                            <div class="settings-collapse-content">
+                                                <label>L0 并发数</label>
+                                                <input type="number" id="vector-l0-concurrency" min="1" max="50" step="1" value="10">
+                                                <div class="settings-hint" style="margin-bottom:16px;">请视 API 情况自行调整。</div>
+                                                <div class="settings-row">
+                                                    <div class="settings-field">
+                                                        <label>渠道</label>
+                                                        <select id="l0-api-provider">
+                                                            <option value="siliconflow">硅基流动</option>
+                                                            <option value="openrouter">OpenRouter</option>
+                                                            <option value="custom">自定义</option>
+                                                        </select>
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row" id="l0-api-url-row">
+                                                    <div class="settings-field full">
+                                                        <label>API URL</label>
+                                                        <input type="text" id="l0-api-url" placeholder="https://api.siliconflow.cn/v1">
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row" id="l0-api-key-row">
+                                                    <div class="settings-field full">
+                                                        <label>API KEY</label>
+                                                        <div style="display:flex; gap:8px; align-items:center;">
+                                                            <input type="password" id="l0-api-key" placeholder="sk-xxx" style="flex:1;">
+                                                            <button type="button" class="btn btn-sm" id="l0-api-key-toggle" style="flex:0 0 auto; min-width:64px;">显示</button>
+                                                        </div>
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row" id="l0-api-model-manual-row">
+                                                    <div class="settings-field full">
+                                                        <label>模型名</label>
+                                                        <input type="text" id="l0-api-model-text" placeholder="可手动填写，如 Qwen/Qwen3-8B">
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row hidden" id="l0-api-model-select-row">
+                                                    <div class="settings-field full">
+                                                        <label>已拉取模型</label>
+                                                        <select id="l0-api-model-select"><option value="">请选择</option></select>
+                                                    </div>
+                                                </div>
+                                                <div class="settings-btn-row" id="l0-api-connect-row" style="margin-bottom:8px;">
+                                                    <button class="btn btn-sm" id="l0-btn-connect" style="flex:1">拉取</button>
+                                                    <button class="btn btn-sm" id="l0-btn-test" style="flex:1">测试</button>
+                                                    <button class="btn btn-sm btn-p" id="l0-btn-save" style="flex:1">保存</button>
+                                                </div>
+                                                <div class="settings-hint" id="l0-api-connect-status" style="margin-bottom:16px;"></div>
                                             </div>
-                                            <button class="btn btn-sm" id="btn-test-vector-api">测试连接</button>
+                                        </details>
-                                        </div>
+
+                                        <details class="settings-collapse" id="embedding-api-collapse" style="margin-bottom:12px;">
+                                            <summary class="settings-collapse-header">
+                                                <span>Embedding 模型</span>
+                                                <svg class="collapse-icon" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                                    <polyline points="6 9 12 15 18 9"></polyline>
+                                                </svg>
+                                            </summary>
+                                            <div class="settings-collapse-content">
+                                                <div class="settings-row">
+                                                    <div class="settings-field">
+                                                        <label>渠道</label>
+                                                        <select id="embedding-api-provider">
+                                                            <option value="siliconflow">硅基流动</option>
+                                                            <option value="custom">自定义</option>
+                                                        </select>
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row" id="embedding-api-url-row">
+                                                    <div class="settings-field full">
+                                                        <label>API URL</label>
+                                                        <input type="text" id="embedding-api-url" placeholder="https://api.siliconflow.cn/v1">
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row" id="embedding-api-key-row">
+                                                    <div class="settings-field full">
+                                                        <label>API KEY</label>
+                                                        <div style="display:flex; gap:8px; align-items:center;">
+                                                            <input type="password" id="embedding-api-key" placeholder="sk-xxx" style="flex:1;">
+                                                            <button type="button" class="btn btn-sm" id="embedding-api-key-toggle" style="flex:0 0 auto; min-width:64px;">显示</button>
+                                                        </div>
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row" id="embedding-api-model-manual-row">
+                                                    <div class="settings-field full">
+                                                        <label>模型名</label>
+                                                        <input type="text" id="embedding-api-model-text" placeholder="如 BAAI/bge-m3">
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row hidden" id="embedding-api-model-select-row">
+                                                    <div class="settings-field full">
+                                                        <label>已拉取模型</label>
+                                                        <select id="embedding-api-model-select"><option value="">请选择</option></select>
+                                                    </div>
+                                                </div>
+                                                <div class="settings-btn-row" id="embedding-api-connect-row" style="margin-bottom:8px;">
+                                                    <button class="btn btn-sm" id="embedding-btn-connect" style="flex:1">拉取</button>
+                                                    <button class="btn btn-sm" id="embedding-btn-test" style="flex:1">测试</button>
+                                                    <button class="btn btn-sm btn-p" id="embedding-btn-save" style="flex:1">保存</button>
+                                                </div>
+                                                <div class="settings-hint" id="embedding-api-connect-status" style="margin-bottom:16px;"></div>
+                                            </div>
+                                        </details>
+
+                                        <details class="settings-collapse" id="rerank-api-collapse">
+                                            <summary class="settings-collapse-header">
+                                                <span>Rerank 模型</span>
+                                                <svg class="collapse-icon" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                                    <polyline points="6 9 12 15 18 9"></polyline>
+                                                </svg>
+                                            </summary>
+                                            <div class="settings-collapse-content">
+                                                <div class="settings-row">
+                                                    <div class="settings-field">
+                                                        <label>渠道</label>
+                                                        <select id="rerank-api-provider">
+                                                            <option value="siliconflow">硅基流动</option>
+                                                            <option value="custom">自定义</option>
+                                                        </select>
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row" id="rerank-api-url-row">
+                                                    <div class="settings-field full">
+                                                        <label>API URL</label>
+                                                        <input type="text" id="rerank-api-url" placeholder="https://api.siliconflow.cn/v1">
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row" id="rerank-api-key-row">
+                                                    <div class="settings-field full">
+                                                        <label>API KEY</label>
+                                                        <div style="display:flex; gap:8px; align-items:center;">
+                                                            <input type="password" id="rerank-api-key" placeholder="sk-xxx" style="flex:1;">
+                                                            <button type="button" class="btn btn-sm" id="rerank-api-key-toggle" style="flex:0 0 auto; min-width:64px;">显示</button>
+                                                        </div>
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row" id="rerank-api-model-manual-row">
+                                                    <div class="settings-field full">
+                                                        <label>模型名</label>
+                                                        <input type="text" id="rerank-api-model-text" placeholder="如 BAAI/bge-reranker-v2-m3">
+                                                    </div>
+                                                </div>
+                                                <div class="settings-row hidden" id="rerank-api-model-select-row">
+                                                    <div class="settings-field full">
+                                                        <label>已拉取模型</label>
+                                                        <select id="rerank-api-model-select"><option value="">请选择</option></select>
+                                                    </div>
+                                                </div>
+                                                <div class="settings-btn-row" id="rerank-api-connect-row" style="margin-bottom:8px;">
+                                                    <button class="btn btn-sm" id="rerank-btn-connect" style="flex:1">拉取</button>
+                                                    <button class="btn btn-sm" id="rerank-btn-test" style="flex:1">测试</button>
+                                                    <button class="btn btn-sm btn-p" id="rerank-btn-save" style="flex:1">保存</button>
+                                                </div>
+                                                <div class="settings-hint" id="rerank-api-connect-status"></div>
+                                            </div>
+                                        </details>
                                     </div>
                                 </div>
                             </div>

modules/story-summary/story-summary.js

@@ -43,6 +43,8 @@ import { runSummaryGeneration } from "./generate/generator.js";
 
 // vector service
 import { embed, getEngineFingerprint, testOnlineService } from "./vector/utils/embedder.js";
+import { testL0Service } from "./vector/llm/llm-service.js";
+import { testRerankService } from "./vector/llm/reranker.js";
 
 // tokenizer
 import { preload as preloadTokenizer, injectEntities, isReady as isTokenizerReady } from "./vector/utils/tokenizer.js";
@@ -356,8 +358,8 @@ async function handleAnchorGenerate() {
             return;
         }
 
-        if (!vectorCfg.online?.key) {
+        if (!vectorCfg.l0Api?.key) {
-            postToFrame({ type: "VECTOR_ONLINE_STATUS", status: "error", message: "请配置 API Key" });
+            postToFrame({ type: "VECTOR_ONLINE_STATUS", status: "error", message: "请配置 L0 API Key" });
             return;
         }
 
@@ -421,17 +423,23 @@ function handleAnchorCancel() {
     postToFrame({ type: "ANCHOR_GEN_PROGRESS", current: -1, total: 0 });
 }
 
-async function handleTestOnlineService(provider, config) {
+async function handleTestOnlineService(provider, config, target = "embedding") {
     try {
-        postToFrame({ type: "VECTOR_ONLINE_STATUS", status: "downloading", message: "连接中..." });
+        postToFrame({ type: "VECTOR_ONLINE_STATUS", target, status: "downloading", message: "连接中..." });
-        const result = await testOnlineService(provider, config);
+        let result;
+        if (target === "l0") result = await testL0Service(config);
+        else if (target === "rerank") result = await testRerankService(config);
+        else result = await testOnlineService(provider, config);
         postToFrame({
             type: "VECTOR_ONLINE_STATUS",
+            target,
             status: "success",
-            message: `连接成功 (${result.dims}维)`,
+            message: target === "embedding"
+                ? `连接成功 (${result.dims}维)`
+                : (result.message || "连接成功"),
         });
     } catch (e) {
-        postToFrame({ type: "VECTOR_ONLINE_STATUS", status: "error", message: e.message });
+        postToFrame({ type: "VECTOR_ONLINE_STATUS", target, status: "error", message: e.message });
     }
 }
 
@@ -448,8 +456,8 @@ async function handleGenerateVectors(vectorCfg) {
         const { chatId, chat } = getContext();
         if (!chatId || !chat?.length) return;
 
-        if (!vectorCfg.online?.key) {
+        if (!vectorCfg.embeddingApi?.key) {
-            postToFrame({ type: "VECTOR_ONLINE_STATUS", status: "error", message: "请配置 API Key" });
+            postToFrame({ type: "VECTOR_ONLINE_STATUS", status: "error", message: "请配置 Embedding API Key" });
             return;
         }
 
@@ -1631,7 +1639,7 @@ async function handleFrameMessage(event) {
             break;
 
         case "VECTOR_TEST_ONLINE":
-            handleTestOnlineService(data.provider, data.config);
+            handleTestOnlineService(data.provider, data.config, data.target || "embedding");
             break;
 
         case "VECTOR_GENERATE":

modules/story-summary/vector/llm/atom-extraction.js

@@ -8,14 +8,14 @@
 // 每楼层 1-2 个场景锚点（非碎片原子），60-100 字场景摘要
 // ============================================================================
 
-import { callLLM, parseJson } from './llm-service.js';
+import { callLLM, cancelAllL0Requests, parseJson } from './llm-service.js';
 import { xbLog } from '../../../../core/debug-core.js';
 import { filterText } from '../utils/text-filter.js';
 
 const MODULE_ID = 'atom-extraction';
 
 const CONCURRENCY = 10;
-const RETRY_COUNT = 2;
+const RETRY_COUNT = 1;
 const RETRY_DELAY = 500;
 const DEFAULT_TIMEOUT = 40000;
 const STAGGER_DELAY = 80;
@@ -25,6 +25,7 @@ let batchCancelled = false;
 
 export function cancelBatchExtraction() {
     batchCancelled = true;
+    cancelAllL0Requests();
 }
 
 export function isBatchCancelled() {

modules/story-summary/vector/llm/llm-service.js

@@ -2,13 +2,15 @@
 // vector/llm/llm-service.js - 修复 prefill 传递方式
 // ═══════════════════════════════════════════════════════════════════════════
 import { xbLog } from '../../../../core/debug-core.js';
-import { getApiKey } from './siliconflow.js';
+import { getVectorConfig } from '../../data/config.js';
 
 const MODULE_ID = 'vector-llm-service';
-const SILICONFLOW_API_URL = 'https://api.siliconflow.cn/v1';
 const DEFAULT_L0_MODEL = 'Qwen/Qwen3-8B';
+const DEFAULT_L0_API_URL = 'https://api.siliconflow.cn/v1';
 
 let callCounter = 0;
+const activeL0SessionIds = new Set();
+let l0KeyIndex = 0;
 
 function getStreamingModule() {
     const mod = window.xiaobaixStreamingGeneration;
@@ -27,6 +29,39 @@ function b64UrlEncode(str) {
     return btoa(bin).replace(/\+/g, '-').replace(/\//g, '_').replace(/=+$/, '');
 }
 
+function getL0ApiConfig() {
+    const cfg = getVectorConfig() || {};
+    return cfg.l0Api || {
+        provider: 'siliconflow',
+        url: DEFAULT_L0_API_URL,
+        key: '',
+        model: DEFAULT_L0_MODEL,
+    };
+}
+
+function normalizeL0ApiConfig(apiConfig = null) {
+    const fallback = getL0ApiConfig();
+    const next = apiConfig || {};
+    return {
+        provider: String(next.provider || fallback.provider || 'siliconflow').trim(),
+        url: String(next.url || fallback.url || DEFAULT_L0_API_URL).trim(),
+        key: String(next.key || fallback.key || '').trim(),
+        model: String(next.model || fallback.model || DEFAULT_L0_MODEL).trim(),
+    };
+}
+
+function getNextKey(rawKey) {
+    const keys = String(rawKey || '')
+        .split(/[,;|\n]+/)
+        .map(k => k.trim())
+        .filter(Boolean);
+    if (!keys.length) return '';
+    if (keys.length === 1) return keys[0];
+    const idx = l0KeyIndex % keys.length;
+    l0KeyIndex = (l0KeyIndex + 1) % keys.length;
+    return keys[idx];
+}
+
 /**
  * 统一LLM调用 - 走酒馆后端（非流式）
  * 临时改为标准 messages 调用，避免 bottomassistant prefill 兼容性问题。
@@ -36,12 +71,14 @@ export async function callLLM(messages, options = {}) {
         temperature = 0.2,
         max_tokens = 500,
         timeout = 40000,
+        apiConfig = null,
     } = options;
 
     const mod = getStreamingModule();
     if (!mod) throw new Error('Streaming module not ready');
 
-    const apiKey = getApiKey() || '';
+    const apiCfg = normalizeL0ApiConfig(apiConfig);
+    const apiKey = getNextKey(apiCfg.key);
     if (!apiKey) {
         throw new Error('L0 requires siliconflow API key');
     }
@@ -59,16 +96,17 @@ export async function callLLM(messages, options = {}) {
         temperature: String(temperature),
         max_tokens: String(max_tokens),
         api: 'openai',
-        apiurl: SILICONFLOW_API_URL,
+        apiurl: String(apiCfg.url || DEFAULT_L0_API_URL).trim(),
         apipassword: apiKey,
-        model: DEFAULT_L0_MODEL,
+        model: String(apiCfg.model || DEFAULT_L0_MODEL).trim(),
     };
-    const isQwen3 = String(DEFAULT_L0_MODEL || '').includes('Qwen3');
+    const isQwen3 = String(args.model || '').includes('Qwen3');
     if (isQwen3) {
         args.enable_thinking = 'false';
     }
 
     try {
+        activeL0SessionIds.add(uniqueId);
         const timeoutPromise = new Promise((_, reject) => {
             setTimeout(() => reject(new Error(`L0 request timeout after ${timeout}ms`)), timeout);
         });
@@ -80,9 +118,38 @@ export async function callLLM(messages, options = {}) {
     } catch (e) {
         xbLog.error(MODULE_ID, 'LLM调用失败', e);
         throw e;
+    } finally {
+        activeL0SessionIds.delete(uniqueId);
     }
 }
 
+export async function testL0Service(apiConfig = {}) {
+    if (!apiConfig?.key) {
+        throw new Error('请配置 L0 API Key');
+    }
+    const result = await callLLM([
+        { role: 'system', content: '你是一个测试助手。请只输出 OK。' },
+        { role: 'user', content: '只输出 OK' },
+    ], {
+        apiConfig,
+        temperature: 0,
+        max_tokens: 16,
+        timeout: 15000,
+    });
+    const text = String(result || '').trim();
+    if (!text) throw new Error('返回为空');
+    return { success: true, message: `连接成功：${text.slice(0, 60)}` };
+}
+
+export function cancelAllL0Requests() {
+    const mod = getStreamingModule();
+    if (!mod?.cancel) return;
+    for (const sessionId of activeL0SessionIds) {
+        try { mod.cancel(sessionId); } catch {}
+    }
+    activeL0SessionIds.clear();
+}
+
 export function parseJson(text) {
     if (!text) return null;
     let s = text.trim().replace(/^```(?:json)?\s*/i, '').replace(/\s*```$/i, '').trim();

modules/story-summary/vector/llm/reranker.js

@@ -4,15 +4,38 @@
 // ═══════════════════════════════════════════════════════════════════════════
 
 import { xbLog } from '../../../../core/debug-core.js';
-import { getApiKey } from './siliconflow.js';
+import { getVectorConfig } from '../../data/config.js';
 
 const MODULE_ID = 'reranker';
-const RERANK_URL = 'https://api.siliconflow.cn/v1/rerank';
+const DEFAULT_RERANK_URL = 'https://api.siliconflow.cn/v1';
 const RERANK_MODEL = 'BAAI/bge-reranker-v2-m3';
 const DEFAULT_TIMEOUT = 15000;
 const MAX_DOCUMENTS = 100;  // API 限制
 const RERANK_BATCH_SIZE = 20;
 const RERANK_MAX_CONCURRENCY = 5;
+let rerankKeyIndex = 0;
+
+function getRerankApiConfig() {
+    const cfg = getVectorConfig() || {};
+    return cfg.rerankApi || {
+        provider: 'siliconflow',
+        url: DEFAULT_RERANK_URL,
+        key: '',
+        model: RERANK_MODEL,
+    };
+}
+
+function getNextRerankKey(rawKey) {
+    const keys = String(rawKey || '')
+        .split(/[,;|\n]+/)
+        .map(k => k.trim())
+        .filter(Boolean);
+    if (!keys.length) return '';
+    if (keys.length === 1) return keys[0];
+    const idx = rerankKeyIndex % keys.length;
+    rerankKeyIndex = (rerankKeyIndex + 1) % keys.length;
+    return keys[idx];
+}
 
 /**
  * 对文档列表进行 Rerank 精排
@@ -37,7 +60,8 @@ export async function rerank(query, documents, options = {}) {
         return { results: [], failed: false };
     }
 
-    const key = getApiKey();
+    const apiCfg = getRerankApiConfig();
+    const key = getNextRerankKey(apiCfg.key);
     if (!key) {
         xbLog.warn(MODULE_ID, '未配置 API Key，跳过 rerank');
         return { results: documents.map((_, i) => ({ index: i, relevance_score: 0 })), failed: true };
@@ -72,14 +96,15 @@ export async function rerank(query, documents, options = {}) {
     try {
         const T0 = performance.now();
 
-        const response = await fetch(RERANK_URL, {
+        const baseUrl = String(apiCfg.url || DEFAULT_RERANK_URL).replace(/\/+$/, '');
+        const response = await fetch(`${baseUrl}/rerank`, {
             method: 'POST',
             headers: {
                 'Authorization': `Bearer ${key}`,
                 'Content-Type': 'application/json',
             },
             body: JSON.stringify({
-                model: RERANK_MODEL,
+                model: String(apiCfg.model || RERANK_MODEL),
                 // Zero-darkbox: do not silently truncate query.
                 query,
                 documents: validDocs,
@@ -248,19 +273,51 @@ export async function rerankChunks(query, chunks, options = {}) {
 /**
  * 测试 Rerank 服务连接
  */
-export async function testRerankService() {
+export async function testRerankService(apiConfig = {}) {
-    const key = getApiKey();
+    const next = {
-    if (!key) {
+        provider: String(apiConfig.provider || 'siliconflow').trim(),
-        throw new Error('请配置硅基 API Key');
+        url: String(apiConfig.url || DEFAULT_RERANK_URL).trim(),
+        key: String(apiConfig.key || '').trim(),
+        model: String(apiConfig.model || RERANK_MODEL).trim(),
+    };
+    if (!next.key) {
+        throw new Error('请配置 Rerank API Key');
     }
 
+    const key = getNextRerankKey(next.key);
+    const controller = new AbortController();
+    const timeoutId = setTimeout(() => controller.abort(), 15000);
     try {
-        const { results } = await rerank('测试查询', ['测试文档1', '测试文档2'], { topN: 2 });
+        const baseUrl = String(next.url || DEFAULT_RERANK_URL).replace(/\/+$/, '');
-        return { 
+        const response = await fetch(`${baseUrl}/rerank`, {
-            success: true, 
+            method: 'POST',
-            message: `连接成功，返回 ${results.length} 个结果`,
+            headers: {
+                'Authorization': `Bearer ${key}`,
+                'Content-Type': 'application/json',
+            },
+            body: JSON.stringify({
+                model: next.model,
+                query: '测试查询',
+                documents: ['测试文档1', '测试文档2'],
+                top_n: 2,
+                return_documents: false,
+            }),
+            signal: controller.signal,
+        });
+        clearTimeout(timeoutId);
+        if (!response.ok) {
+            const errorText = await response.text().catch(() => '');
+            throw new Error(`Rerank API ${response.status}: ${errorText.slice(0, 200)}`);
+        }
+        const data = await response.json();
+        const results = Array.isArray(data.results) ? data.results : [];
+        return {
+            success: true,
+            message: `连接成功：返回 ${results.length} 个结果`,
         };
     } catch (e) {
         throw new Error(`连接失败: ${e.message}`);
+    } finally {
+        clearTimeout(timeoutId);
     }
 }

modules/story-summary/vector/llm/siliconflow.js

@@ -1,31 +1,39 @@
 // ═══════════════════════════════════════════════════════════════════════════
-// siliconflow.js - Embedding + 多 Key 轮询
+// siliconflow.js - OpenAI-compatible Embedding + 多 Key 轮询
 //
 // 在 API Key 输入框中用逗号、分号、竖线或换行分隔多个 Key，例如：
 //   sk-aaa,sk-bbb,sk-ccc
 // 每次调用自动轮询到下一个 Key，并发请求会均匀分布到所有 Key 上。
 // ═══════════════════════════════════════════════════════════════════════════
 
+import { getVectorConfig } from '../../data/config.js';
+
 const BASE_URL = 'https://api.siliconflow.cn';
 const EMBEDDING_MODEL = 'BAAI/bge-m3';
 
 // ★ 多 Key 轮询状态
 let _keyIndex = 0;
 
+function getEmbeddingApiConfig() {
+    const cfg = getVectorConfig() || {};
+    return cfg.embeddingApi || {
+        provider: 'siliconflow',
+        url: `${BASE_URL}/v1`,
+        key: '',
+        model: EMBEDDING_MODEL,
+    };
+}
+
 /**
  * 从 localStorage 解析所有 Key（支持逗号、分号、竖线、换行分隔）
  */
-function parseKeys() {
+function parseKeys(rawKey) {
     try {
-        const raw = localStorage.getItem('summary_panel_config');
+        const keyStr = String(rawKey || '');
-        if (raw) {
+        return keyStr
-            const parsed = JSON.parse(raw);
+            .split(/[,;|\n]+/)
-            const keyStr = parsed.vector?.online?.key || '';
+            .map(k => k.trim())
-            return keyStr
+            .filter(k => k.length > 0);
-                .split(/[,;|\n]+/)
-                .map(k => k.trim())
-                .filter(k => k.length > 0);
-        }
     } catch { }
     return [];
 }
@@ -34,8 +42,8 @@ function parseKeys() {
  * 获取下一个可用的 API Key（轮询）
  * 每次调用返回不同的 Key，自动循环
  */
-export function getApiKey() {
+export function getApiKey(rawKey = null) {
-    const keys = parseKeys();
+    const keys = parseKeys(rawKey ?? getEmbeddingApiConfig().key);
     if (!keys.length) return null;
     if (keys.length === 1) return keys[0];
 
@@ -51,7 +59,7 @@ export function getApiKey() {
  * 获取当前配置的 Key 数量（供外部模块动态调整并发用）
  */
 export function getKeyCount() {
-    return Math.max(1, parseKeys().length);
+    return Math.max(1, parseKeys(getEmbeddingApiConfig().key).length);
 }
 
 // ═══════════════════════════════════════════════════════════════════════════
@@ -61,22 +69,24 @@ export function getKeyCount() {
 export async function embed(texts, options = {}) {
     if (!texts?.length) return [];
 
-    const key = getApiKey();
+    const apiCfg = options.apiConfig || getEmbeddingApiConfig();
-    if (!key) throw new Error('未配置硅基 API Key');
+    const key = getApiKey(apiCfg.key);
+    if (!key) throw new Error('未配置 Embedding API Key');
 
     const { timeout = 30000, signal } = options;
     const controller = new AbortController();
     const timeoutId = setTimeout(() => controller.abort(), timeout);
 
     try {
-        const response = await fetch(`${BASE_URL}/v1/embeddings`, {
+        const baseUrl = String(apiCfg.url || `${BASE_URL}/v1`).replace(/\/+$/, '');
+        const response = await fetch(`${baseUrl}/embeddings`, {
             method: 'POST',
             headers: {
                 'Authorization': `Bearer ${key}`,
                 'Content-Type': 'application/json',
             },
             body: JSON.stringify({
-                model: EMBEDDING_MODEL,
+                model: String(apiCfg.model || EMBEDDING_MODEL),
                 input: texts,
             }),
             signal: signal || controller.signal,

modules/story-summary/vector/pipeline/state-integration.js

@@ -29,7 +29,7 @@ import { filterText } from '../utils/text-filter.js';
 const MODULE_ID = 'state-integration';
 
 // ★ 并发配置
-const CONCURRENCY = 50;
+const DEFAULT_CONCURRENCY = 10;
 const STAGGER_DELAY = 15;
 const DEBUG_CONCURRENCY = true;
 const R_AGG_MAX_CHARS = 256;
@@ -168,7 +168,9 @@ export async function incrementalExtractAtoms(chatId, chat, onProgress, options
         return { built: 0 };
     }
 
-    xbLog.info(MODULE_ID, `增量 L0 提取：pending=${pendingPairs.length}, concurrency=${CONCURRENCY}`);
+    const concurrency = Math.max(1, Math.min(50, Number(vectorCfg?.l0Concurrency) || DEFAULT_CONCURRENCY));
+
+    xbLog.info(MODULE_ID, `增量 L0 提取：pending=${pendingPairs.length}, concurrency=${concurrency}`);
 
     let completed = 0;
     let failed = 0;
@@ -181,14 +183,6 @@ export async function incrementalExtractAtoms(chatId, chat, onProgress, options
     // ★ Phase 1: 收集所有新提取的 atoms（不向量化）
     const allNewAtoms = [];
 
-    // ★ 限流检测：连续失败 N 次后暂停并降速
-    let consecutiveFailures = 0;
-    let rateLimited = false;
-    const RATE_LIMIT_THRESHOLD = 6;       // 连续失败多少次触发限流保护
-    const RATE_LIMIT_WAIT_MS = 60000;      // 限流后等待时间（60 秒）
-    const RETRY_INTERVAL_MS = 1000;        // 降速模式下每次请求间隔（1 秒）
-    const RETRY_CONCURRENCY = 1;           // ★ 降速模式下的并发数（默认1，建议不要超过5）
-
     // ★ 通用处理单个 pair 的逻辑（复用于正常模式和降速模式）
     const processPair = async (pair, idx, workerId) => {
         const floor = pair.aiFloor;
@@ -209,9 +203,6 @@ export async function incrementalExtractAtoms(chatId, chat, onProgress, options
                 throw new Error('llm_failed');
             }
 
-            // ★ 成功：重置连续失败计数
-            consecutiveFailures = 0;
-
             if (!atoms.length) {
                 setL0FloorStatus(floor, { status: 'empty', reason: 'llm_empty', atoms: 0 });
             } else {
@@ -231,13 +222,6 @@ export async function incrementalExtractAtoms(chatId, chat, onProgress, options
                 reason: String(e?.message || e).replace(/\s+/g, ' ').slice(0, 120),
             });
             failed++;
-
-            // ★ 限流检测：连续失败累加
-            consecutiveFailures++;
-            if (consecutiveFailures >= RATE_LIMIT_THRESHOLD && !rateLimited) {
-                rateLimited = true;
-                xbLog.warn(MODULE_ID, `连续失败 ${consecutiveFailures} 次，疑似触发 API 限流，将暂停所有并发`);
-            }
         } finally {
             active--;
             if (!extractionCancelled) {
@@ -252,12 +236,12 @@ export async function incrementalExtractAtoms(chatId, chat, onProgress, options
     };
 
     // ★ 并发池处理（保持固定并发度）
-    const poolSize = Math.min(CONCURRENCY, pendingPairs.length);
+    const poolSize = Math.min(concurrency, pendingPairs.length);
     let nextIndex = 0;
     let started = 0;
     const runWorker = async (workerId) => {
         while (true) {
-            if (extractionCancelled || rateLimited) return;
+            if (extractionCancelled) return;
             const idx = nextIndex++;
             if (idx >= pendingPairs.length) return;
 
@@ -267,7 +251,7 @@ export async function incrementalExtractAtoms(chatId, chat, onProgress, options
                 await new Promise(r => setTimeout(r, stagger * STAGGER_DELAY));
             }
 
-            if (extractionCancelled || rateLimited) return;
+            if (extractionCancelled) return;
 
             await processPair(pair, idx, workerId);
         }
@@ -279,61 +263,6 @@ export async function incrementalExtractAtoms(chatId, chat, onProgress, options
         xbLog.info(MODULE_ID, `L0 pool done completed=${completed}/${total} failed=${failed} peakActive=${peakActive} elapsedMs=${elapsed}`);
     }
 
-    // ═════════════════════════════════════════════════════════════════════
-    // ★ 限流恢复：重置进度，从头开始以限速模式慢慢跑
-    // ═════════════════════════════════════════════════════════════════════
-    if (rateLimited && !extractionCancelled) {
-        const waitSec = RATE_LIMIT_WAIT_MS / 1000;
-        xbLog.info(MODULE_ID, `限流保护：将重置进度并从头开始降速重来（并发=${RETRY_CONCURRENCY}, 间隔=${RETRY_INTERVAL_MS}ms）`);
-        onProgress?.(`疑似限流，${waitSec}s 后降速重头开始...`, completed, total);
-
-        await new Promise(r => setTimeout(r, RATE_LIMIT_WAIT_MS));
-
-        if (!extractionCancelled) {
-            // ★ 核心逻辑：重置计数器，让 UI 从 0 开始跑，给用户“重头开始”的反馈
-            rateLimited = false;
-            consecutiveFailures = 0;
-            completed = 0;
-            failed = 0;
-
-            let retryNextIdx = 0;
-
-            xbLog.info(MODULE_ID, `限流恢复：开始降速模式扫描 ${pendingPairs.length} 个楼层`);
-
-            const retryWorkers = Math.min(RETRY_CONCURRENCY, pendingPairs.length);
-            const runRetryWorker = async (wid) => {
-                while (true) {
-                    if (extractionCancelled) return;
-                    const idx = retryNextIdx++;
-                    if (idx >= pendingPairs.length) return;
-
-                    const pair = pendingPairs[idx];
-                    const floor = pair.aiFloor;
-
-                    // ★ 检查该楼层状态
-                    const st = getL0FloorStatus(floor);
-                    if (st?.status === 'ok' || st?.status === 'empty') {
-                        // 刚才已经成功了，直接跳过（仅增加进度计数）
-                        completed++;
-                        onProgress?.(`提取: ${completed}/${total} (跳过已完成)`, completed, total);
-                        continue;
-                    }
-
-                    // ★ 没做过的，用 slow 模式处理
-                    await processPair(pair, idx, `retry-${wid}`);
-
-                    // 每个请求后休息，避免再次触发限流
-                    if (idx < pendingPairs.length - 1 && RETRY_INTERVAL_MS > 0) {
-                        await new Promise(r => setTimeout(r, RETRY_INTERVAL_MS));
-                    }
-                }
-            };
-
-            await Promise.all(Array.from({ length: retryWorkers }, (_, i) => runRetryWorker(i)));
-            xbLog.info(MODULE_ID, `降速重头开始阶段结束`);
-        }
-    }
-
     try {
         saveMetadataDebounced?.();
     } catch { }

modules/story-summary/vector/storage/vector-io.js

@@ -83,16 +83,6 @@ function uint8ToBase64(uint8) {
     return btoa(result);
 }
 
-// base64 → Uint8Array
-function base64ToUint8(base64) {
-    const binary = atob(base64);
-    const bytes = new Uint8Array(binary.length);
-    for (let i = 0; i < binary.length; i++) {
-        bytes[i] = binary.charCodeAt(i);
-    }
-    return bytes;
-}
-
 // 服务器备份文件名
 function getBackupFilename(chatId) {
     // chatId 可能含中文/特殊字符，ST 只接受 [a-zA-Z0-9_-]

modules/story-summary/vector/utils/embedder.js

@@ -1,15 +1,13 @@
 // ═══════════════════════════════════════════════════════════════════════════
-// Story Summary - Embedder (v2 - 统一硅基)
+// Story Summary - Embedder
-// 所有 embedding 请求转发到 siliconflow.js
 // ═══════════════════════════════════════════════════════════════════════════
 
-import { embed as sfEmbed, getApiKey } from '../llm/siliconflow.js';
+import { embed as sfEmbed } from '../llm/siliconflow.js';
 // ═══════════════════════════════════════════════════════════════════════════
 // 统一 embed 接口
 // ═══════════════════════════════════════════════════════════════════════════
 
 export async function embed(texts, config, options = {}) {
-    // 忽略旧的 config 参数，统一走硅基
     return await sfEmbed(texts, options);
 }
 
@@ -18,8 +16,10 @@ export async function embed(texts, config, options = {}) {
 // ═══════════════════════════════════════════════════════════════════════════
 
 export function getEngineFingerprint(config) {
-    // 统一使用硅基 bge-m3
+    const api = config?.embeddingApi || {};
-    return 'siliconflow:bge-m3:1024';
+    const provider = String(api.provider || 'siliconflow').toLowerCase();
+    const model = String(api.model || 'BAAI/bge-m3').trim() || 'BAAI/bge-m3';
+    return `${provider}:${model}:1024`;
 }
 
 // ═══════════════════════════════════════════════════════════════════════════
@@ -47,14 +47,13 @@ export async function deleteLocalModelCache() { }
 // 在线服务测试
 // ═══════════════════════════════════════════════════════════════════════════
 
-export async function testOnlineService() {
+export async function testOnlineService(_provider, config = {}) {
-    const key = getApiKey();
+    if (!config?.key) {
-    if (!key) {
+        throw new Error('请配置 Embedding API Key');
-        throw new Error('请配置硅基 API Key');
     }
 
     try {
-        const [vec] = await sfEmbed(['测试连接']);
+        const [vec] = await sfEmbed(['测试连接'], { apiConfig: config });
         return { success: true, dims: vec?.length || 0 };
     } catch (e) {
         throw new Error(`连接失败: ${e.message}`);
@@ -62,7 +61,6 @@ export async function testOnlineService() {
 }
 
 export async function fetchOnlineModels() {
-    // 硅基模型固定
     return ['BAAI/bge-m3'];
 }
 
@@ -78,6 +76,6 @@ export const ONLINE_PROVIDERS = {
     siliconflow: {
         id: 'siliconflow',
         name: '硅基流动',
-        baseUrl: 'https://api.siliconflow.cn',
+        baseUrl: 'https://api.siliconflow.cn/v1',
     },
 };