@prestyj/ai 4.14.0 → 5.0.1

package/dist/index.js

@@ -1094,7 +1094,13 @@ async function prewarmAnthropicCache(options) {
       },
       {
         signal: options.signal ?? void 0,
-        ...isOAuth ? { headers: { "anthropic-beta": "claude-code-20250219,oauth-2025-04-20" } } : {}
+        ...(() => {
+          const betas = [
+            ...isOAuth ? ["claude-code-20250219", "oauth-2025-04-20"] : [],
+            ...cacheControl?.ttl === "1h" ? ["extended-cache-ttl-2025-04-11"] : []
+          ];
+          return betas.length ? { headers: { "anthropic-beta": betas.join(",") } } : {};
+        })()
       }
     );
   } catch {
@@ -1178,7 +1184,12 @@ async function* runStream(options) {
     ...options.compaction ? ["compact-2026-01-12"] : [],
     ...options.clearToolUses ? ["context-management-2025-06-27"] : [],
     "fine-grained-tool-streaming-2025-05-14",
-    ...!hasAdaptiveThinking ? ["interleaved-thinking-2025-05-14"] : []
+    ...!hasAdaptiveThinking ? ["interleaved-thinking-2025-05-14"] : [],
+    // The 1-h cache TTL (cacheRetention "long") is gated behind this beta. Without
+    // it Anthropic silently ignores ttl:"1h" and falls back to the 5-min default,
+    // so a pre-warmed cache expires before the user's first turn. cacheControl.ttl
+    // is only "1h" on the first-party endpoint (see toAnthropicCacheControl).
+    ...cacheControl?.ttl === "1h" ? ["extended-cache-ttl-2025-04-11"] : []
   ];
   const requestOptions = {
     signal: options.signal ?? void 0,
@@ -3092,11 +3103,17 @@ providerRegistry.register("anthropic", {
   stream: (options) => streamAnthropic(options)
 });
 providerRegistry.register("xiaomi", {
-  stream: (options) => streamOpenAI({
-    ...options,
-    baseUrl: options.baseUrl ?? "https://token-plan-sgp.xiaomimimo.com/v1",
-    webSearch: false
-  })
+  stream: (options) => {
+    const TOKEN_PLAN_URL = "https://token-plan-sgp.xiaomimimo.com/v1";
+    const PLATFORM_URL = "https://api.xiaomimimo.com/v1";
+    let baseUrl;
+    if (options.model === "mimo-v2.5-pro-ultraspeed") {
+      baseUrl = options.baseUrl && options.baseUrl !== TOKEN_PLAN_URL ? options.baseUrl : PLATFORM_URL;
+    } else {
+      baseUrl = options.baseUrl ?? TOKEN_PLAN_URL;
+    }
+    return streamOpenAI({ ...options, baseUrl, webSearch: false });
+  }
 });
 providerRegistry.register("openai", {
   stream: (options) => {