@openrouter/ai-sdk-provider 2.2.5 → 2.3.1

package/dist/internal/index.d.mts

@@ -186,6 +186,25 @@ type OpenRouterChatSettings = {
          */
         engine?: Engine;
     };
+    /**
+     * Enable Anthropic automatic prompt caching by setting a top-level cache_control
+     * directive on the request body. When set to `{ type: 'ephemeral' }`, Anthropic
+     * will automatically cache eligible content in your prompts.
+     *
+     * Only works with Anthropic models through OpenRouter.
+     *
+     * @see https://platform.claude.com/docs/en/build-with-claude/prompt-caching#automatic-caching
+     * @see https://openrouter.ai/docs
+     */
+    cache_control?: {
+        type: 'ephemeral';
+        /**
+         * Optional time-to-live for the cache entry.
+         * - `'5m'` — 5 minutes (default when omitted)
+         * - `'1h'` — 1 hour
+         */
+        ttl?: '5m' | '1h';
+    };
     /**
      * Debug options for troubleshooting API requests.
      * Only works with streaming requests.

package/dist/internal/index.d.ts

@@ -186,6 +186,25 @@ type OpenRouterChatSettings = {
          */
         engine?: Engine;
     };
+    /**
+     * Enable Anthropic automatic prompt caching by setting a top-level cache_control
+     * directive on the request body. When set to `{ type: 'ephemeral' }`, Anthropic
+     * will automatically cache eligible content in your prompts.
+     *
+     * Only works with Anthropic models through OpenRouter.
+     *
+     * @see https://platform.claude.com/docs/en/build-with-claude/prompt-caching#automatic-caching
+     * @see https://openrouter.ai/docs
+     */
+    cache_control?: {
+        type: 'ephemeral';
+        /**
+         * Optional time-to-live for the cache entry.
+         * - `'5m'` — 5 minutes (default when omitted)
+         * - `'1h'` — 1 hour
+         */
+        ttl?: '5m' | '1h';
+    };
     /**
      * Debug options for troubleshooting API requests.
      * Only works with streaming requests.

package/dist/internal/index.js

@@ -2759,11 +2759,12 @@ function convertToOpenRouterChatMessages(prompt) {
           }
           finalReasoningDetails = uniqueDetails.length > 0 ? uniqueDetails : void 0;
         }
+        const effectiveReasoning = reasoning && finalReasoningDetails ? reasoning : void 0;
         messages.push({
           role: "assistant",
           content: text,
           tool_calls: toolCalls.length > 0 ? toolCalls : void 0,
-          reasoning: reasoning || void 0,
+          reasoning: effectiveReasoning,
           reasoning_details: finalReasoningDetails,
           annotations: messageAnnotations,
           cache_control: getCacheControl(providerOptions)
@@ -3145,7 +3146,9 @@ var OpenRouterChatLanguageModel = class {
       // Provider routing settings:
       provider: this.settings.provider,
       // Debug settings:
-      debug: this.settings.debug
+      debug: this.settings.debug,
+      // Anthropic automatic caching:
+      cache_control: this.settings.cache_control
     }, this.config.extraBody), this.settings.extraBody);
     if (tools && tools.length > 0) {
       const mappedTools = tools.filter(
@@ -3166,10 +3169,11 @@ var OpenRouterChatLanguageModel = class {
     return baseArgs;
   }
   async doGenerate(options) {
-    var _a16, _b16, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v;
+    var _b16, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w;
     const providerOptions = options.providerOptions || {};
     const openrouterOptions = providerOptions.openrouter || {};
-    const args = __spreadValues(__spreadValues({}, this.getArgs(options)), openrouterOptions);
+    const _a16 = openrouterOptions, { cacheControl } = _a16, restOpenrouterOptions = __objRest(_a16, ["cacheControl"]);
+    const args = __spreadValues(__spreadValues(__spreadValues({}, this.getArgs(options)), restOpenrouterOptions), cacheControl != null && !("cache_control" in restOpenrouterOptions) ? { cache_control: cacheControl } : {});
     const { value: responseValue, responseHeaders } = await postJsonToApi({
       url: this.config.url({
         path: "/chat/completions",
@@ -3206,7 +3210,7 @@ var OpenRouterChatLanguageModel = class {
       });
     }
     const usageInfo = response.usage ? computeTokenUsage(response.usage) : emptyUsage();
-    const reasoningDetails = (_a16 = choice.message.reasoning_details) != null ? _a16 : [];
+    const reasoningDetails = (_b16 = choice.message.reasoning_details) != null ? _b16 : [];
     const reasoning = reasoningDetails.length > 0 ? reasoningDetails.map((detail) => {
       switch (detail.type) {
         case "reasoning.text" /* Text */: {
@@ -3275,9 +3279,9 @@ var OpenRouterChatLanguageModel = class {
       for (const toolCall of choice.message.tool_calls) {
         content.push({
           type: "tool-call",
-          toolCallId: (_b16 = toolCall.id) != null ? _b16 : generateId(),
+          toolCallId: (_c = toolCall.id) != null ? _c : generateId(),
           toolName: toolCall.function.name,
-          input: (_c = toolCall.function.arguments) != null ? _c : "{}",
+          input: (_d = toolCall.function.arguments) != null ? _d : "{}",
           providerMetadata: !reasoningDetailsAttachedToToolCall ? {
             openrouter: {
               reasoning_details: reasoningDetails
@@ -3304,19 +3308,19 @@ var OpenRouterChatLanguageModel = class {
             sourceType: "url",
             id: annotation.url_citation.url,
             url: annotation.url_citation.url,
-            title: (_d = annotation.url_citation.title) != null ? _d : "",
+            title: (_e = annotation.url_citation.title) != null ? _e : "",
             providerMetadata: {
               openrouter: {
-                content: (_e = annotation.url_citation.content) != null ? _e : "",
-                startIndex: (_f = annotation.url_citation.start_index) != null ? _f : 0,
-                endIndex: (_g = annotation.url_citation.end_index) != null ? _g : 0
+                content: (_f = annotation.url_citation.content) != null ? _f : "",
+                startIndex: (_g = annotation.url_citation.start_index) != null ? _g : 0,
+                endIndex: (_h = annotation.url_citation.end_index) != null ? _h : 0
               }
             }
           });
         }
       }
     }
-    const fileAnnotations = (_h = choice.message.annotations) == null ? void 0 : _h.filter(
+    const fileAnnotations = (_i = choice.message.annotations) == null ? void 0 : _i.filter(
       (a) => a.type === "file"
     );
     const hasToolCalls = choice.message.tool_calls && choice.message.tool_calls.length > 0;
@@ -3324,7 +3328,7 @@ var OpenRouterChatLanguageModel = class {
       (d) => d.type === "reasoning.encrypted" /* Encrypted */ && d.data
     );
     const shouldOverrideFinishReason = hasToolCalls && hasEncryptedReasoning && choice.finish_reason === "stop";
-    const effectiveFinishReason = shouldOverrideFinishReason ? createFinishReason("tool-calls", (_i = choice.finish_reason) != null ? _i : void 0) : mapOpenRouterFinishReason(choice.finish_reason);
+    const effectiveFinishReason = shouldOverrideFinishReason ? createFinishReason("tool-calls", (_j = choice.finish_reason) != null ? _j : void 0) : mapOpenRouterFinishReason(choice.finish_reason);
     return {
       content,
       finishReason: effectiveFinishReason,
@@ -3332,22 +3336,22 @@ var OpenRouterChatLanguageModel = class {
       warnings: [],
       providerMetadata: {
         openrouter: OpenRouterProviderMetadataSchema.parse({
-          provider: (_j = response.provider) != null ? _j : "",
-          reasoning_details: (_k = choice.message.reasoning_details) != null ? _k : [],
+          provider: (_k = response.provider) != null ? _k : "",
+          reasoning_details: (_l = choice.message.reasoning_details) != null ? _l : [],
           annotations: fileAnnotations && fileAnnotations.length > 0 ? fileAnnotations : void 0,
           usage: __spreadValues(__spreadValues(__spreadValues(__spreadValues({
-            promptTokens: (_l = usageInfo.inputTokens.total) != null ? _l : 0,
-            completionTokens: (_m = usageInfo.outputTokens.total) != null ? _m : 0,
-            totalTokens: ((_n = usageInfo.inputTokens.total) != null ? _n : 0) + ((_o = usageInfo.outputTokens.total) != null ? _o : 0)
-          }, ((_p = response.usage) == null ? void 0 : _p.cost) != null ? { cost: response.usage.cost } : {}), ((_r = (_q = response.usage) == null ? void 0 : _q.prompt_tokens_details) == null ? void 0 : _r.cached_tokens) != null ? {
+            promptTokens: (_m = usageInfo.inputTokens.total) != null ? _m : 0,
+            completionTokens: (_n = usageInfo.outputTokens.total) != null ? _n : 0,
+            totalTokens: ((_o = usageInfo.inputTokens.total) != null ? _o : 0) + ((_p = usageInfo.outputTokens.total) != null ? _p : 0)
+          }, ((_q = response.usage) == null ? void 0 : _q.cost) != null ? { cost: response.usage.cost } : {}), ((_s = (_r = response.usage) == null ? void 0 : _r.prompt_tokens_details) == null ? void 0 : _s.cached_tokens) != null ? {
             promptTokensDetails: {
               cachedTokens: response.usage.prompt_tokens_details.cached_tokens
             }
-          } : {}), ((_t = (_s = response.usage) == null ? void 0 : _s.completion_tokens_details) == null ? void 0 : _t.reasoning_tokens) != null ? {
+          } : {}), ((_u = (_t = response.usage) == null ? void 0 : _t.completion_tokens_details) == null ? void 0 : _u.reasoning_tokens) != null ? {
             completionTokensDetails: {
               reasoningTokens: response.usage.completion_tokens_details.reasoning_tokens
             }
-          } : {}), ((_v = (_u = response.usage) == null ? void 0 : _u.cost_details) == null ? void 0 : _v.upstream_inference_cost) != null ? {
+          } : {}), ((_w = (_v = response.usage) == null ? void 0 : _v.cost_details) == null ? void 0 : _w.upstream_inference_cost) != null ? {
             costDetails: {
               upstreamInferenceCost: response.usage.cost_details.upstream_inference_cost
             }
@@ -3363,10 +3367,11 @@ var OpenRouterChatLanguageModel = class {
     };
   }
   async doStream(options) {
-    var _a16;
+    var _b16;
     const providerOptions = options.providerOptions || {};
     const openrouterOptions = providerOptions.openrouter || {};
-    const args = __spreadValues(__spreadValues({}, this.getArgs(options)), openrouterOptions);
+    const _a16 = openrouterOptions, { cacheControl } = _a16, restOpenrouterOptions = __objRest(_a16, ["cacheControl"]);
+    const args = __spreadValues(__spreadValues(__spreadValues({}, this.getArgs(options)), restOpenrouterOptions), cacheControl != null && !("cache_control" in restOpenrouterOptions) ? { cache_control: cacheControl } : {});
     const { value: response, responseHeaders } = await postJsonToApi({
       url: this.config.url({
         path: "/chat/completions",
@@ -3378,7 +3383,7 @@ var OpenRouterChatLanguageModel = class {
         // only include stream_options when in strict compatibility mode:
         stream_options: this.config.compatibility === "strict" ? __spreadValues({
           include_usage: true
-        }, ((_a16 = this.settings.usage) == null ? void 0 : _a16.include) ? { include_usage: true } : {}) : void 0
+        }, ((_b16 = this.settings.usage) == null ? void 0 : _b16.include) ? { include_usage: true } : {}) : void 0
       }),
       failedResponseHandler: openrouterFailedResponseHandler,
       successfulResponseHandler: createEventSourceResponseHandler(
@@ -3418,7 +3423,7 @@ var OpenRouterChatLanguageModel = class {
       stream: response.pipeThrough(
         new TransformStream({
           transform(chunk, controller) {
-            var _a17, _b16, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u;
+            var _a17, _b17, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -3455,7 +3460,7 @@ var OpenRouterChatLanguageModel = class {
               Object.assign(usage.outputTokens, computed.outputTokens);
               rawUsage = value.usage;
               const promptTokens = (_a17 = value.usage.prompt_tokens) != null ? _a17 : 0;
-              const completionTokens = (_b16 = value.usage.completion_tokens) != null ? _b16 : 0;
+              const completionTokens = (_b17 = value.usage.completion_tokens) != null ? _b17 : 0;
               openrouterUsage.promptTokens = promptTokens;
               if (value.usage.prompt_tokens_details) {
                 openrouterUsage.promptTokensDetails = {