@ai-sdk/openai 2.0.12 → 2.0.14

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # @ai-sdk/openai
 
+## 2.0.14
+
+### Patch Changes
+
+- 7f47105: fix(provider/openai): support file_citation annotations in responses api
+
+## 2.0.13
+
+### Patch Changes
+
+- ddc9d99: Implements `logprobs` for OpenAI `providerOptions` and `providerMetaData` in `OpenAIResponsesLanguageModel`
+
+  You can now set `providerOptions.openai.logprobs` when using `generateText()` and retrieve logprobs from the response via `result.providerMetadata?.openai`
+
 ## 2.0.12
 
 ### Patch Changes

package/dist/index.d.mts

@@ -161,6 +161,7 @@ declare const openaiResponsesProviderOptionsSchema: z.ZodObject<{
     include: z.ZodOptional<z.ZodNullable<z.ZodArray<z.ZodEnum<{
         "reasoning.encrypted_content": "reasoning.encrypted_content";
         "file_search_call.results": "file_search_call.results";
+        "message.output_text.logprobs": "message.output_text.logprobs";
     }>>>>;
     textVerbosity: z.ZodOptional<z.ZodNullable<z.ZodEnum<{
         low: "low";
@@ -169,6 +170,7 @@ declare const openaiResponsesProviderOptionsSchema: z.ZodObject<{
     }>>>;
     promptCacheKey: z.ZodOptional<z.ZodNullable<z.ZodString>>;
     safetyIdentifier: z.ZodOptional<z.ZodNullable<z.ZodString>>;
+    logprobs: z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodNumber]>>;
 }, z.core.$strip>;
 type OpenAIResponsesProviderOptions = z.infer<typeof openaiResponsesProviderOptionsSchema>;
 

package/dist/index.d.ts

@@ -161,6 +161,7 @@ declare const openaiResponsesProviderOptionsSchema: z.ZodObject<{
     include: z.ZodOptional<z.ZodNullable<z.ZodArray<z.ZodEnum<{
         "reasoning.encrypted_content": "reasoning.encrypted_content";
         "file_search_call.results": "file_search_call.results";
+        "message.output_text.logprobs": "message.output_text.logprobs";
     }>>>>;
     textVerbosity: z.ZodOptional<z.ZodNullable<z.ZodEnum<{
         low: "low";
@@ -169,6 +170,7 @@ declare const openaiResponsesProviderOptionsSchema: z.ZodObject<{
     }>>>;
     promptCacheKey: z.ZodOptional<z.ZodNullable<z.ZodString>>;
     safetyIdentifier: z.ZodOptional<z.ZodNullable<z.ZodString>>;
+    logprobs: z.ZodOptional<z.ZodUnion<readonly [z.ZodBoolean, z.ZodNumber]>>;
 }, z.core.$strip>;
 type OpenAIResponsesProviderOptions = z.infer<typeof openaiResponsesProviderOptionsSchema>;
 

package/dist/index.js

@@ -2127,6 +2127,19 @@ function prepareResponsesTools({
 }
 
 // src/responses/openai-responses-language-model.ts
+var TOP_LOGPROBS_MAX = 20;
+var LOGPROBS_SCHEMA = import_v413.z.array(
+  import_v413.z.object({
+    token: import_v413.z.string(),
+    logprob: import_v413.z.number(),
+    top_logprobs: import_v413.z.array(
+      import_v413.z.object({
+        token: import_v413.z.string(),
+        logprob: import_v413.z.number()
+      })
+    )
+  })
+);
 var OpenAIResponsesLanguageModel = class {
   constructor(modelId, config) {
     this.specificationVersion = "v2";
@@ -2190,6 +2203,8 @@ var OpenAIResponsesLanguageModel = class {
       schema: openaiResponsesProviderOptionsSchema
     });
     const strictJsonSchema = (_a = openaiOptions == null ? void 0 : openaiOptions.strictJsonSchema) != null ? _a : false;
+    const topLogprobs = typeof (openaiOptions == null ? void 0 : openaiOptions.logprobs) === "number" ? openaiOptions == null ? void 0 : openaiOptions.logprobs : (openaiOptions == null ? void 0 : openaiOptions.logprobs) === true ? TOP_LOGPROBS_MAX : void 0;
+    const openaiOptionsInclude = topLogprobs ? Array.isArray(openaiOptions == null ? void 0 : openaiOptions.include) ? [...openaiOptions == null ? void 0 : openaiOptions.include, "message.output_text.logprobs"] : ["message.output_text.logprobs"] : openaiOptions == null ? void 0 : openaiOptions.include;
     const baseArgs = {
       model: this.modelId,
       input: messages,
@@ -2220,9 +2235,10 @@ var OpenAIResponsesLanguageModel = class {
       user: openaiOptions == null ? void 0 : openaiOptions.user,
       instructions: openaiOptions == null ? void 0 : openaiOptions.instructions,
       service_tier: openaiOptions == null ? void 0 : openaiOptions.serviceTier,
-      include: openaiOptions == null ? void 0 : openaiOptions.include,
+      include: openaiOptionsInclude,
       prompt_cache_key: openaiOptions == null ? void 0 : openaiOptions.promptCacheKey,
       safety_identifier: openaiOptions == null ? void 0 : openaiOptions.safetyIdentifier,
+      top_logprobs: topLogprobs,
       // model-specific settings:
       ...modelConfig.isReasoningModel && ((openaiOptions == null ? void 0 : openaiOptions.reasoningEffort) != null || (openaiOptions == null ? void 0 : openaiOptions.reasoningSummary) != null) && {
         reasoning: {
@@ -2306,7 +2322,7 @@ var OpenAIResponsesLanguageModel = class {
     };
   }
   async doGenerate(options) {
-    var _a, _b, _c, _d, _e, _f, _g, _h, _i;
+    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",
@@ -2340,14 +2356,24 @@ var OpenAIResponsesLanguageModel = class {
                   import_v413.z.object({
                     type: import_v413.z.literal("output_text"),
                     text: import_v413.z.string(),
+                    logprobs: LOGPROBS_SCHEMA.nullish(),
                     annotations: import_v413.z.array(
-                      import_v413.z.object({
-                        type: import_v413.z.literal("url_citation"),
-                        start_index: import_v413.z.number(),
-                        end_index: import_v413.z.number(),
-                        url: import_v413.z.string(),
-                        title: import_v413.z.string()
-                      })
+                      import_v413.z.discriminatedUnion("type", [
+                        import_v413.z.object({
+                          type: import_v413.z.literal("url_citation"),
+                          start_index: import_v413.z.number(),
+                          end_index: import_v413.z.number(),
+                          url: import_v413.z.string(),
+                          title: import_v413.z.string()
+                        }),
+                        import_v413.z.object({
+                          type: import_v413.z.literal("file_citation"),
+                          start_index: import_v413.z.number(),
+                          end_index: import_v413.z.number(),
+                          file_id: import_v413.z.string(),
+                          quote: import_v413.z.string()
+                        })
+                      ])
                     )
                   })
                 )
@@ -2417,6 +2443,7 @@ var OpenAIResponsesLanguageModel = class {
       });
     }
     const content = [];
+    const logprobs = [];
     for (const part of response.output) {
       switch (part.type) {
         case "reasoning": {
@@ -2439,6 +2466,9 @@ var OpenAIResponsesLanguageModel = class {
         }
         case "message": {
           for (const contentPart of part.content) {
+            if (((_c = (_b = options.providerOptions) == null ? void 0 : _b.openai) == null ? void 0 : _c.logprobs) && contentPart.logprobs) {
+              logprobs.push(contentPart.logprobs);
+            }
             content.push({
               type: "text",
               text: contentPart.text,
@@ -2449,13 +2479,24 @@ var OpenAIResponsesLanguageModel = class {
               }
             });
             for (const annotation of contentPart.annotations) {
-              content.push({
-                type: "source",
-                sourceType: "url",
-                id: (_d = (_c = (_b = this.config).generateId) == null ? void 0 : _c.call(_b)) != null ? _d : (0, import_provider_utils12.generateId)(),
-                url: annotation.url,
-                title: annotation.title
-              });
+              if (annotation.type === "url_citation") {
+                content.push({
+                  type: "source",
+                  sourceType: "url",
+                  id: (_f = (_e = (_d = this.config).generateId) == null ? void 0 : _e.call(_d)) != null ? _f : (0, import_provider_utils12.generateId)(),
+                  url: annotation.url,
+                  title: annotation.title
+                });
+              } else if (annotation.type === "file_citation") {
+                content.push({
+                  type: "source",
+                  sourceType: "document",
+                  id: (_i = (_h = (_g = this.config).generateId) == null ? void 0 : _h.call(_g)) != null ? _i : (0, import_provider_utils12.generateId)(),
+                  mediaType: "text/plain",
+                  title: annotation.quote,
+                  filename: annotation.file_id
+                });
+              }
             }
           }
           break;
@@ -2535,18 +2576,24 @@ var OpenAIResponsesLanguageModel = class {
         }
       }
     }
+    const providerMetadata = {
+      openai: { responseId: response.id }
+    };
+    if (logprobs.length > 0) {
+      providerMetadata.openai.logprobs = logprobs;
+    }
     return {
       content,
       finishReason: mapOpenAIResponseFinishReason({
-        finishReason: (_e = response.incomplete_details) == null ? void 0 : _e.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: (_g = (_f = response.usage.output_tokens_details) == null ? void 0 : _f.reasoning_tokens) != null ? _g : void 0,
-        cachedInputTokens: (_i = (_h = response.usage.input_tokens_details) == null ? void 0 : _h.cached_tokens) != null ? _i : 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: {
@@ -2556,11 +2603,7 @@ var OpenAIResponsesLanguageModel = class {
         headers: responseHeaders,
         body: rawResponse
       },
-      providerMetadata: {
-        openai: {
-          responseId: response.id
-        }
-      },
+      providerMetadata,
       warnings
     };
   }
@@ -2590,6 +2633,7 @@ var OpenAIResponsesLanguageModel = class {
       outputTokens: void 0,
       totalTokens: void 0
     };
+    const logprobs = [];
     let responseId = null;
     const ongoingToolCalls = {};
     let hasToolCalls = false;
@@ -2601,7 +2645,7 @@ 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;
+            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 });
             }
@@ -2815,6 +2859,9 @@ var OpenAIResponsesLanguageModel = class {
                 id: value.item_id,
                 delta: value.delta
               });
+              if (value.logprobs) {
+                logprobs.push(value.logprobs);
+              }
             } else if (isResponseReasoningSummaryPartAddedChunk(value)) {
               if (value.summary_index > 0) {
                 (_c = activeReasoning[value.item_id]) == null ? void 0 : _c.summaryParts.push(
@@ -2853,27 +2900,42 @@ var OpenAIResponsesLanguageModel = class {
               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)) {
-              controller.enqueue({
-                type: "source",
-                sourceType: "url",
-                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
-              });
+              if (value.annotation.type === "url_citation") {
+                controller.enqueue({
+                  type: "source",
+                  sourceType: "url",
+                  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
+                });
+              } else if (value.annotation.type === "file_citation") {
+                controller.enqueue({
+                  type: "source",
+                  sourceType: "document",
+                  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
+                });
+              }
             } else if (isErrorChunk(value)) {
               controller.enqueue({ type: "error", error: value });
             }
           },
           flush(controller) {
+            const providerMetadata = {
+              openai: {
+                responseId
+              }
+            };
+            if (logprobs.length > 0) {
+              providerMetadata.openai.logprobs = logprobs;
+            }
             controller.enqueue({
               type: "finish",
               finishReason,
               usage,
-              providerMetadata: {
-                openai: {
-                  responseId
-                }
-              }
+              providerMetadata
             });
           }
         })
@@ -2892,7 +2954,8 @@ var usageSchema2 = import_v413.z.object({
 var textDeltaChunkSchema = import_v413.z.object({
   type: import_v413.z.literal("response.output_text.delta"),
   item_id: import_v413.z.string(),
-  delta: import_v413.z.string()
+  delta: import_v413.z.string(),
+  logprobs: LOGPROBS_SCHEMA.nullish()
 });
 var errorChunkSchema = import_v413.z.object({
   type: import_v413.z.literal("error"),
@@ -3021,11 +3084,18 @@ var responseFunctionCallArgumentsDeltaSchema = import_v413.z.object({
 });
 var responseAnnotationAddedSchema = import_v413.z.object({
   type: import_v413.z.literal("response.output_text.annotation.added"),
-  annotation: import_v413.z.object({
-    type: import_v413.z.literal("url_citation"),
-    url: import_v413.z.string(),
-    title: import_v413.z.string()
-  })
+  annotation: import_v413.z.discriminatedUnion("type", [
+    import_v413.z.object({
+      type: import_v413.z.literal("url_citation"),
+      url: import_v413.z.string(),
+      title: import_v413.z.string()
+    }),
+    import_v413.z.object({
+      type: import_v413.z.literal("file_citation"),
+      file_id: import_v413.z.string(),
+      quote: import_v413.z.string()
+    })
+  ])
 });
 var responseReasoningSummaryPartAddedSchema = import_v413.z.object({
   type: import_v413.z.literal("response.reasoning_summary_part.added"),
@@ -3133,10 +3203,29 @@ var openaiResponsesProviderOptionsSchema = import_v413.z.object({
   instructions: import_v413.z.string().nullish(),
   reasoningSummary: import_v413.z.string().nullish(),
   serviceTier: import_v413.z.enum(["auto", "flex", "priority"]).nullish(),
-  include: import_v413.z.array(import_v413.z.enum(["reasoning.encrypted_content", "file_search_call.results"])).nullish(),
+  include: import_v413.z.array(
+    import_v413.z.enum([
+      "reasoning.encrypted_content",
+      "file_search_call.results",
+      "message.output_text.logprobs"
+    ])
+  ).nullish(),
   textVerbosity: import_v413.z.enum(["low", "medium", "high"]).nullish(),
   promptCacheKey: import_v413.z.string().nullish(),
-  safetyIdentifier: import_v413.z.string().nullish()
+  safetyIdentifier: import_v413.z.string().nullish(),
+  /**
+   * Return the log probabilities of the tokens.
+   *
+   * Setting to true will return the log probabilities of the tokens that
+   * were generated.
+   *
+   * Setting to a number will return the log probabilities of the top n
+   * tokens that were generated.
+   *
+   * @see https://platform.openai.com/docs/api-reference/responses/create
+   * @see https://cookbook.openai.com/examples/using_logprobs
+   */
+  logprobs: import_v413.z.union([import_v413.z.boolean(), import_v413.z.number().min(1).max(TOP_LOGPROBS_MAX)]).optional()
 });
 
 // src/speech/openai-speech-model.ts