@ai-sdk/openai 2.0.16 → 2.0.18

package/CHANGELOG.md

@@ -1,5 +1,23 @@
 # @ai-sdk/openai
 
+## 2.0.18
+
+### Patch Changes
+
+- 5e47d00: Support Responses API input_file file_url passthrough for PDFs.
+
+  This adds:
+
+  - file_url variant to OpenAIResponses user content
+  - PDF URL mapping to input_file with file_url in Responses converter
+  - PDF URL support in supportedUrls to avoid auto-download
+
+## 2.0.17
+
+### Patch Changes
+
+- 70bb696: fix(provider/openai): correct web search tool input
+
 ## 2.0.16
 
 ### Patch Changes

package/dist/index.js

@@ -436,7 +436,23 @@ var webSearchPreviewArgsSchema = import_v44.z.object({
 var webSearchPreview = (0, import_provider_utils4.createProviderDefinedToolFactory)({
   id: "openai.web_search_preview",
   name: "web_search_preview",
-  inputSchema: import_v44.z.object({})
+  inputSchema: import_v44.z.object({
+    action: import_v44.z.discriminatedUnion("type", [
+      import_v44.z.object({
+        type: import_v44.z.literal("search"),
+        query: import_v44.z.string()
+      }),
+      import_v44.z.object({
+        type: import_v44.z.literal("open_page"),
+        url: import_v44.z.string()
+      }),
+      import_v44.z.object({
+        type: import_v44.z.literal("find"),
+        url: import_v44.z.string(),
+        pattern: import_v44.z.string()
+      })
+    ]).nullish()
+  })
 });
 
 // src/chat/openai-chat-prepare-tools.ts
@@ -1887,9 +1903,10 @@ async function convertToOpenAIResponsesMessages({
                   };
                 } else if (part.mediaType === "application/pdf") {
                   if (part.data instanceof URL) {
-                    throw new import_provider6.UnsupportedFunctionalityError({
-                      functionality: "PDF file parts with URLs"
-                    });
+                    return {
+                      type: "input_file",
+                      file_url: part.data.toString()
+                    };
                   }
                   return {
                     type: "input_file",
@@ -2127,6 +2144,26 @@ function prepareResponsesTools({
 }
 
 // src/responses/openai-responses-language-model.ts
+var webSearchCallItem = import_v413.z.object({
+  type: import_v413.z.literal("web_search_call"),
+  id: import_v413.z.string(),
+  status: import_v413.z.string(),
+  action: import_v413.z.discriminatedUnion("type", [
+    import_v413.z.object({
+      type: import_v413.z.literal("search"),
+      query: import_v413.z.string()
+    }),
+    import_v413.z.object({
+      type: import_v413.z.literal("open_page"),
+      url: import_v413.z.string()
+    }),
+    import_v413.z.object({
+      type: import_v413.z.literal("find"),
+      url: import_v413.z.string(),
+      pattern: import_v413.z.string()
+    })
+  ]).nullish()
+});
 var TOP_LOGPROBS_MAX = 20;
 var LOGPROBS_SCHEMA = import_v413.z.array(
   import_v413.z.object({
@@ -2144,7 +2181,8 @@ var OpenAIResponsesLanguageModel = class {
   constructor(modelId, config) {
     this.specificationVersion = "v2";
     this.supportedUrls = {
-      "image/*": [/^https?:\/\/.*$/]
+      "image/*": [/^https?:\/\/.*$/],
+      "application/pdf": [/^https?:\/\/.*$/]
     };
     this.modelId = modelId;
     this.config = config;
@@ -2322,7 +2360,7 @@ var OpenAIResponsesLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q;
+    var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n;
     const { args: body, warnings } = await this.getArgs(options);
     const url = this.config.url({
       path: "/responses",
@@ -2385,15 +2423,7 @@ var OpenAIResponsesLanguageModel = class {
                 arguments: import_v413.z.string(),
                 id: import_v413.z.string()
               }),
-              import_v413.z.object({
-                type: import_v413.z.literal("web_search_call"),
-                id: import_v413.z.string(),
-                status: import_v413.z.string().optional(),
-                action: import_v413.z.object({
-                  type: import_v413.z.literal("search"),
-                  query: import_v413.z.string().optional()
-                }).nullish()
-              }),
+              webSearchCallItem,
               import_v413.z.object({
                 type: import_v413.z.literal("computer_call"),
                 id: import_v413.z.string(),
@@ -2524,17 +2554,14 @@ var OpenAIResponsesLanguageModel = class {
             type: "tool-call",
             toolCallId: part.id,
             toolName: "web_search_preview",
-            input: (_k = (_j = part.action) == null ? void 0 : _j.query) != null ? _k : "",
+            input: JSON.stringify({ action: part.action }),
             providerExecuted: true
           });
           content.push({
             type: "tool-result",
             toolCallId: part.id,
             toolName: "web_search_preview",
-            result: {
-              status: part.status || "completed",
-              ...((_l = part.action) == null ? void 0 : _l.query) && { query: part.action.query }
-            },
+            result: { status: part.status },
             providerExecuted: true
           });
           break;
@@ -2592,15 +2619,15 @@ var OpenAIResponsesLanguageModel = class {
     return {
       content,
       finishReason: mapOpenAIResponseFinishReason({
-        finishReason: (_m = response.incomplete_details) == null ? void 0 : _m.reason,
+        finishReason: (_j = response.incomplete_details) == null ? void 0 : _j.reason,
         hasToolCalls: content.some((part) => part.type === "tool-call")
       }),
       usage: {
         inputTokens: response.usage.input_tokens,
         outputTokens: response.usage.output_tokens,
         totalTokens: response.usage.input_tokens + response.usage.output_tokens,
-        reasoningTokens: (_o = (_n = response.usage.output_tokens_details) == null ? void 0 : _n.reasoning_tokens) != null ? _o : void 0,
-        cachedInputTokens: (_q = (_p = response.usage.input_tokens_details) == null ? void 0 : _p.cached_tokens) != null ? _q : void 0
+        reasoningTokens: (_l = (_k = response.usage.output_tokens_details) == null ? void 0 : _k.reasoning_tokens) != null ? _l : void 0,
+        cachedInputTokens: (_n = (_m = response.usage.input_tokens_details) == null ? void 0 : _m.cached_tokens) != null ? _n : void 0
       },
       request: { body },
       response: {
@@ -2652,7 +2679,7 @@ var OpenAIResponsesLanguageModel = class {
             controller.enqueue({ type: "stream-start", warnings });
           },
           transform(chunk, controller) {
-            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s;
+            var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k, _l, _m, _n, _o, _p;
             if (options.includeRawChunks) {
               controller.enqueue({ type: "raw", rawValue: chunk.rawValue });
             }
@@ -2759,20 +2786,14 @@ var OpenAIResponsesLanguageModel = class {
                   type: "tool-call",
                   toolCallId: value.item.id,
                   toolName: "web_search_preview",
-                  input: (_c = (_b = value.item.action) == null ? void 0 : _b.query) != null ? _c : "",
+                  input: JSON.stringify({ action: value.item.action }),
                   providerExecuted: true
                 });
                 controller.enqueue({
                   type: "tool-result",
                   toolCallId: value.item.id,
                   toolName: "web_search_preview",
-                  result: {
-                    type: "web_search_tool_result",
-                    status: value.item.status || "completed",
-                    ...((_d = value.item.action) == null ? void 0 : _d.query) && {
-                      query: value.item.action.query
-                    }
-                  },
+                  result: { status: value.item.status },
                   providerExecuted: true
                 });
               } else if (value.item.type === "computer_call") {
@@ -2839,7 +2860,7 @@ var OpenAIResponsesLanguageModel = class {
                     providerMetadata: {
                       openai: {
                         itemId: value.item.id,
-                        reasoningEncryptedContent: (_e = value.item.encrypted_content) != null ? _e : null
+                        reasoningEncryptedContent: (_b = value.item.encrypted_content) != null ? _b : null
                       }
                     }
                   });
@@ -2874,7 +2895,7 @@ var OpenAIResponsesLanguageModel = class {
               }
             } else if (isResponseReasoningSummaryPartAddedChunk(value)) {
               if (value.summary_index > 0) {
-                (_f = activeReasoning[value.item_id]) == null ? void 0 : _f.summaryParts.push(
+                (_c = activeReasoning[value.item_id]) == null ? void 0 : _c.summaryParts.push(
                   value.summary_index
                 );
                 controller.enqueue({
@@ -2883,7 +2904,7 @@ var OpenAIResponsesLanguageModel = class {
                   providerMetadata: {
                     openai: {
                       itemId: value.item_id,
-                      reasoningEncryptedContent: (_h = (_g = activeReasoning[value.item_id]) == null ? void 0 : _g.encryptedContent) != null ? _h : null
+                      reasoningEncryptedContent: (_e = (_d = activeReasoning[value.item_id]) == null ? void 0 : _d.encryptedContent) != null ? _e : null
                     }
                   }
                 });
@@ -2901,20 +2922,20 @@ var OpenAIResponsesLanguageModel = class {
               });
             } else if (isResponseFinishedChunk(value)) {
               finishReason = mapOpenAIResponseFinishReason({
-                finishReason: (_i = value.response.incomplete_details) == null ? void 0 : _i.reason,
+                finishReason: (_f = value.response.incomplete_details) == null ? void 0 : _f.reason,
                 hasToolCalls
               });
               usage.inputTokens = value.response.usage.input_tokens;
               usage.outputTokens = value.response.usage.output_tokens;
               usage.totalTokens = value.response.usage.input_tokens + value.response.usage.output_tokens;
-              usage.reasoningTokens = (_k = (_j = value.response.usage.output_tokens_details) == null ? void 0 : _j.reasoning_tokens) != null ? _k : void 0;
-              usage.cachedInputTokens = (_m = (_l = value.response.usage.input_tokens_details) == null ? void 0 : _l.cached_tokens) != null ? _m : void 0;
+              usage.reasoningTokens = (_h = (_g = value.response.usage.output_tokens_details) == null ? void 0 : _g.reasoning_tokens) != null ? _h : void 0;
+              usage.cachedInputTokens = (_j = (_i = value.response.usage.input_tokens_details) == null ? void 0 : _i.cached_tokens) != null ? _j : void 0;
             } else if (isResponseAnnotationAddedChunk(value)) {
               if (value.annotation.type === "url_citation") {
                 controller.enqueue({
                   type: "source",
                   sourceType: "url",
-                  id: (_p = (_o = (_n = self.config).generateId) == null ? void 0 : _o.call(_n)) != null ? _p : (0, import_provider_utils12.generateId)(),
+                  id: (_m = (_l = (_k = self.config).generateId) == null ? void 0 : _l.call(_k)) != null ? _m : (0, import_provider_utils12.generateId)(),
                   url: value.annotation.url,
                   title: value.annotation.title
                 });
@@ -2922,7 +2943,7 @@ var OpenAIResponsesLanguageModel = class {
                 controller.enqueue({
                   type: "source",
                   sourceType: "document",
-                  id: (_s = (_r = (_q = self.config).generateId) == null ? void 0 : _r.call(_q)) != null ? _s : (0, import_provider_utils12.generateId)(),
+                  id: (_p = (_o = (_n = self.config).generateId) == null ? void 0 : _o.call(_n)) != null ? _p : (0, import_provider_utils12.generateId)(),
                   mediaType: "text/plain",
                   title: value.annotation.quote,
                   filename: value.annotation.file_id
@@ -3062,15 +3083,7 @@ var responseOutputItemDoneSchema = import_v413.z.object({
       arguments: import_v413.z.string(),
       status: import_v413.z.literal("completed")
     }),
-    import_v413.z.object({
-      type: import_v413.z.literal("web_search_call"),
-      id: import_v413.z.string(),
-      status: import_v413.z.literal("completed"),
-      action: import_v413.z.object({
-        type: import_v413.z.literal("search"),
-        query: import_v413.z.string().optional()
-      }).nullish()
-    }),
+    webSearchCallItem,
     import_v413.z.object({
       type: import_v413.z.literal("computer_call"),
       id: import_v413.z.string(),