@acosmi/sdk-ts 2.8.0 → 2.9.0

package/dist/index.mjs

@@ -2679,8 +2679,8 @@ var Client = class _Client {
    * V30 二轮审计 D-P1-2: 此方法不返回 entitlement-filter-status header.
    * UI 想根据 fallback 状态显示降级提示, 请改调 listModelsWithStatus.
    */
-  async listModels(signal) {
-    const r = await this.listModelsWithStatus(signal);
+  async listModels(signal, opts) {
+    const r = await this.listModelsWithStatus(signal, opts);
     return r.models;
   }
   /**
@@ -2693,16 +2693,20 @@ var Client = class _Client {
    *   - status === 'disabled-by-flag' → 不显示余量 (运维灰度中)
    *   - status === '' (Unknown) → 老 nexus, 按老 v0.17 行为 (不显示 BucketInfo)
    */
-  async listModelsWithStatus(signal) {
+  async listModelsWithStatus(signal, opts) {
+    const includeLocked = opts?.includeLocked === true;
+    const path = includeLocked ? "/managed-models?picker=1" : "/managed-models";
     const { result, headers } = await this.doJSONFull(
       "GET",
-      "/managed-models",
+      path,
       null,
       signal
     );
     const normalized = normalizeInputModalities(result.data);
-    this.modelCache = normalized;
-    this.modelCacheTimeMs = Date.now();
+    if (!includeLocked) {
+      this.modelCache = normalized;
+      this.modelCacheTimeMs = Date.now();
+    }
     let status = "";
     if (headers) {
       const h = headers.get("X-Entitlement-Filter-Status");
@@ -2897,6 +2901,34 @@ var Client = class _Client {
     const { result } = await this.doJSONFullRaw("GET", endpoint, null, signal);
     return this.unwrapAPIResponse(result);
   }
+  /**
+   * 向量 (同步)。POST /managed-models/:id/embeddings
+   *
+   * 仅 capabilities.supports_embedding=true 的托管模型可用 (上游接 DashScope)。
+   * 响应为 OpenAI /v1/embeddings 标准格式 (网关直通, 无 response.Success 包装)。
+   *
+   * @param modelID 向量托管模型 ID
+   */
+  async embeddings(modelID, req, signal) {
+    const endpoint = `/managed-models/${encodeURIComponent(modelID)}/embeddings`;
+    const { result } = await this.doJSONFullRaw("POST", endpoint, req, signal, CHAT_REQUEST_TIMEOUT_MS);
+    return JSON.parse(new TextDecoder().decode(result));
+  }
+  /**
+   * 重排序 (同步)。POST /managed-models/:id/rerank
+   *
+   * 仅 capabilities.supports_rerank=true 的托管模型可用 (上游接 DashScope)。
+   * 统一扁平契约: { query, documents[], top_n?, return_documents?, instruct? };
+   * 响应 { results: [{ index, relevance_score, document? }], usage, model }
+   * (网关已把上游原生/OpenAI 兼容两线路归一化, 无 response.Success 包装)。
+   *
+   * @param modelID 重排序托管模型 ID
+   */
+  async rerank(modelID, req, signal) {
+    const endpoint = `/managed-models/${encodeURIComponent(modelID)}/rerank`;
+    const { result } = await this.doJSONFullRaw("POST", endpoint, req, signal, CHAT_REQUEST_TIMEOUT_MS);
+    return JSON.parse(new TextDecoder().decode(result));
+  }
   /**
    * Anthropic 原生格式同步聊天
    * v0.5.0: 根据 provider 自动路由