@ai-sdk/openai 4.0.0-beta.7 → 4.0.0-beta.75

package/src/responses/openai-responses-language-model.ts

@@ -1,18 +1,18 @@
 import {
   APICallError,
-  JSONValue,
-  LanguageModelV4,
-  LanguageModelV4Prompt,
-  LanguageModelV4CallOptions,
-  LanguageModelV4Content,
-  LanguageModelV4FinishReason,
-  LanguageModelV4GenerateResult,
-  LanguageModelV4ProviderTool,
-  LanguageModelV4StreamPart,
-  LanguageModelV4StreamResult,
-  LanguageModelV4ToolApprovalRequest,
-  SharedV4ProviderMetadata,
-  SharedV4Warning,
+  type JSONValue,
+  type LanguageModelV4,
+  type LanguageModelV4Prompt,
+  type LanguageModelV4CallOptions,
+  type LanguageModelV4Content,
+  type LanguageModelV4FinishReason,
+  type LanguageModelV4GenerateResult,
+  type LanguageModelV4ProviderTool,
+  type LanguageModelV4StreamPart,
+  type LanguageModelV4StreamResult,
+  type LanguageModelV4ToolApprovalRequest,
+  type SharedV4ProviderMetadata,
+  type SharedV4Warning,
 } from '@ai-sdk/provider';
 import {
   combineHeaders,
@@ -20,49 +20,59 @@ import {
   createJsonResponseHandler,
   createToolNameMapping,
   generateId,
-  InferSchema,
+  isCustomReasoning,
   parseProviderOptions,
-  ParseResult,
   postJsonToApi,
+  serializeModelOptions,
+  WORKFLOW_DESERIALIZE,
+  WORKFLOW_SERIALIZE,
+  type InferSchema,
+  type ParseResult,
 } from '@ai-sdk/provider-utils';
-import { OpenAIConfig } from '../openai-config';
+import type { OpenAIConfig } from '../openai-config';
 import { openaiFailedResponseHandler } from '../openai-error';
 import { getOpenAILanguageModelCapabilities } from '../openai-language-model-capabilities';
-import { applyPatchInputSchema } from '../tool/apply-patch';
-import {
+import { throwIfOpenAIStreamErrorBeforeOutput } from '../openai-stream-error';
+import type { applyPatchInputSchema } from '../tool/apply-patch';
+import type {
   codeInterpreterInputSchema,
   codeInterpreterOutputSchema,
 } from '../tool/code-interpreter';
-import { fileSearchOutputSchema } from '../tool/file-search';
-import { imageGenerationOutputSchema } from '../tool/image-generation';
-import { localShellInputSchema } from '../tool/local-shell';
-import { mcpOutputSchema } from '../tool/mcp';
-import { shellInputSchema, shellOutputSchema } from '../tool/shell';
-import { webSearchOutputSchema } from '../tool/web-search';
+import type { fileSearchOutputSchema } from '../tool/file-search';
+import type { imageGenerationOutputSchema } from '../tool/image-generation';
+import type { localShellInputSchema } from '../tool/local-shell';
+import type { mcpOutputSchema } from '../tool/mcp';
+import type { shellInputSchema, shellOutputSchema } from '../tool/shell';
+import type {
+  toolSearchInputSchema,
+  toolSearchOutputSchema,
+} from '../tool/tool-search';
+import type { webSearchOutputSchema } from '../tool/web-search';
 import {
   convertOpenAIResponsesUsage,
-  OpenAIResponsesUsage,
+  type OpenAIResponsesUsage,
 } from './convert-openai-responses-usage';
 import { convertToOpenAIResponsesInput } from './convert-to-openai-responses-input';
 import { mapOpenAIResponseFinishReason } from './map-openai-responses-finish-reason';
 import {
-  OpenAIResponsesChunk,
   openaiResponsesChunkSchema,
-  OpenAIResponsesIncludeOptions,
-  OpenAIResponsesIncludeValue,
-  OpenAIResponsesLogprobs,
   openaiResponsesResponseSchema,
-  OpenAIResponsesWebSearchAction,
-  OpenAIResponsesApplyPatchOperationDiffDeltaChunk,
-  OpenAIResponsesApplyPatchOperationDiffDoneChunk,
+  type OpenAIResponsesChunk,
+  type OpenAIResponsesIncludeOptions,
+  type OpenAIResponsesIncludeValue,
+  type OpenAIResponsesLogprobs,
+  type OpenAIResponsesWebSearchAction,
+  type OpenAIResponsesApplyPatchOperationDiffDeltaChunk,
+  type OpenAIResponsesApplyPatchOperationDiffDoneChunk,
 } from './openai-responses-api';
 import {
-  OpenAIResponsesModelId,
   openaiLanguageModelResponsesOptionsSchema,
   TOP_LOGPROBS_MAX,
-} from './openai-responses-options';
+  type OpenAIResponsesModelId,
+} from './openai-responses-language-model-options';
 import { prepareResponsesTools } from './openai-responses-prepare-tools';
-import {
+import type {
+  ResponsesCompactionProviderMetadata,
   ResponsesProviderMetadata,
   ResponsesReasoningProviderMetadata,
   ResponsesSourceDocumentProviderMetadata,
@@ -101,6 +111,20 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
 
   private readonly config: OpenAIConfig;
 
+  static [WORKFLOW_SERIALIZE](model: OpenAIResponsesLanguageModel) {
+    return serializeModelOptions({
+      modelId: model.modelId,
+      config: model.config,
+    });
+  }
+
+  static [WORKFLOW_DESERIALIZE](options: {
+    modelId: OpenAIResponsesModelId;
+    config: OpenAIConfig;
+  }) {
+    return new OpenAIResponsesLanguageModel(options.modelId, options.config);
+  }
+
   constructor(modelId: OpenAIResponsesModelId, config: OpenAIConfig) {
     this.modelId = modelId;
     this.config = config;
@@ -125,6 +149,7 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
     frequencyPenalty,
     seed,
     prompt,
+    reasoning,
     providerOptions,
     tools,
     toolChoice,
@@ -170,6 +195,16 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
       });
     }
 
+    const resolvedReasoningEffort =
+      openaiOptions?.reasoningEffort ??
+      (isCustomReasoning(reasoning) ? reasoning : undefined);
+    const resolvedReasoningSummary =
+      openaiOptions?.reasoningSummary !== undefined
+        ? openaiOptions.reasoningSummary
+        : resolvedReasoningEffort != null && resolvedReasoningEffort !== 'none'
+          ? 'detailed'
+          : undefined;
+
     const isReasoningModel =
       openaiOptions?.forceReasoning ?? modelCapabilities.isReasoningModel;
 
@@ -193,11 +228,8 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
         'openai.web_search_preview': 'web_search_preview',
         'openai.mcp': 'mcp',
         'openai.apply_patch': 'apply_patch',
+        'openai.tool_search': 'tool_search',
       },
-      resolveProviderToolName: tool =>
-        tool.id === 'openai.custom'
-          ? (tool.args as { name?: string }).name
-          : undefined,
     });
 
     const customProviderToolNames = new Set<string>();
@@ -208,6 +240,7 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
     } = await prepareResponsesTools({
       tools,
       toolChoice,
+      allowedTools: openaiOptions?.allowedTools ?? undefined,
       toolNameMapping,
       customProviderToolNames,
     });
@@ -223,8 +256,11 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
             : modelCapabilities.systemMessageMode),
         providerOptionsName,
         fileIdPrefixes: this.config.fileIdPrefixes,
+        passThroughUnsupportedFiles:
+          openaiOptions?.passThroughUnsupportedFiles ?? false,
         store: openaiOptions?.store ?? true,
         hasConversation: openaiOptions?.conversation != null,
+        hasPreviousResponseId: openaiOptions?.previousResponseId != null,
         hasLocalShellTool: hasOpenAITool('openai.local_shell'),
         hasShellTool: hasOpenAITool('openai.shell'),
         hasApplyPatchTool: hasOpenAITool('openai.apply_patch'),
@@ -336,16 +372,24 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
       top_logprobs: topLogprobs,
       truncation: openaiOptions?.truncation,
 
+      // context management (server-side compaction):
+      ...(openaiOptions?.contextManagement && {
+        context_management: openaiOptions.contextManagement.map(cm => ({
+          type: cm.type,
+          compact_threshold: cm.compactThreshold,
+        })),
+      }),
+
       // model-specific settings:
       ...(isReasoningModel &&
-        (openaiOptions?.reasoningEffort != null ||
-          openaiOptions?.reasoningSummary != null) && {
+        (resolvedReasoningEffort != null ||
+          resolvedReasoningSummary != null) && {
           reasoning: {
-            ...(openaiOptions?.reasoningEffort != null && {
-              effort: openaiOptions.reasoningEffort,
+            ...(resolvedReasoningEffort != null && {
+              effort: resolvedReasoningEffort,
             }),
-            ...(openaiOptions?.reasoningSummary != null && {
-              summary: openaiOptions.reasoningSummary,
+            ...(resolvedReasoningSummary != null && {
+              summary: resolvedReasoningSummary,
             }),
           },
         }),
@@ -358,7 +402,7 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
       //  https://platform.openai.com/docs/guides/latest-model#gpt-5-1-parameter-compatibility
       if (
         !(
-          openaiOptions?.reasoningEffort === 'none' &&
+          resolvedReasoningEffort === 'none' &&
           modelCapabilities.supportsNonReasoningParameters
         )
       ) {
@@ -478,7 +522,7 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
       rawValue: rawResponse,
     } = await postJsonToApi({
       url,
-      headers: combineHeaders(this.config.headers(), options.headers),
+      headers: combineHeaders(this.config.headers?.(), options.headers),
       body,
       failedResponseHandler: openaiFailedResponseHandler,
       successfulResponseHandler: createJsonResponseHandler(
@@ -505,6 +549,7 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
 
     // flag that checks if there have been client-side tool calls (not executed by openai)
     let hasFunctionCall = false;
+    const hostedToolSearchCallIds: string[] = [];
 
     // map response content to content array (defined when there is no error)
     for (const part of response.output!) {
@@ -551,6 +596,54 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
           break;
         }
 
+        case 'tool_search_call': {
+          const toolCallId = part.call_id ?? part.id;
+          const isHosted = part.execution === 'server';
+
+          if (isHosted) {
+            hostedToolSearchCallIds.push(toolCallId);
+          }
+
+          content.push({
+            type: 'tool-call',
+            toolCallId,
+            toolName: toolNameMapping.toCustomToolName('tool_search'),
+            input: JSON.stringify({
+              arguments: part.arguments,
+              call_id: part.call_id,
+            } satisfies InferSchema<typeof toolSearchInputSchema>),
+            ...(isHosted ? { providerExecuted: true } : {}),
+            providerMetadata: {
+              [providerOptionsName]: {
+                itemId: part.id,
+              },
+            },
+          });
+
+          break;
+        }
+
+        case 'tool_search_output': {
+          const toolCallId =
+            part.call_id ?? hostedToolSearchCallIds.shift() ?? part.id;
+
+          content.push({
+            type: 'tool-result',
+            toolCallId,
+            toolName: toolNameMapping.toCustomToolName('tool_search'),
+            result: {
+              tools: part.tools,
+            } satisfies InferSchema<typeof toolSearchOutputSchema>,
+            providerMetadata: {
+              [providerOptionsName]: {
+                itemId: part.id,
+              },
+            },
+          });
+
+          break;
+        }
+
         case 'local_shell_call': {
           content.push({
             type: 'tool-call',
@@ -721,6 +814,7 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
             providerMetadata: {
               [providerOptionsName]: {
                 itemId: part.id,
+                ...(part.namespace != null && { namespace: part.namespace }),
               },
             },
           });
@@ -927,6 +1021,21 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
 
           break;
         }
+
+        case 'compaction': {
+          content.push({
+            type: 'custom',
+            kind: 'openai.compaction',
+            providerMetadata: {
+              [providerOptionsName]: {
+                type: 'compaction',
+                itemId: part.id,
+                encryptedContent: part.encrypted_content,
+              } satisfies ResponsesCompactionProviderMetadata,
+            },
+          });
+          break;
+        }
       }
     }
 
@@ -978,12 +1087,14 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
       isShellProviderExecuted,
     } = await this.getArgs(options);
 
+    const url = this.config.url({
+      path: '/responses',
+      modelId: this.modelId,
+    });
+
     const { responseHeaders, value: response } = await postJsonToApi({
-      url: this.config.url({
-        path: '/responses',
-        modelId: this.modelId,
-      }),
-      headers: combineHeaders(this.config.headers(), options.headers),
+      url,
+      headers: combineHeaders(this.config.headers?.(), options.headers),
       body: {
         ...body,
         stream: true,
@@ -996,6 +1107,19 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
       fetch: this.config.fetch,
     });
 
+    const checkedResponse = await throwIfOpenAIStreamErrorBeforeOutput({
+      stream: response,
+      getError: chunk =>
+        isErrorChunk(chunk) ||
+        (isResponseFailedChunk(chunk) && chunk.response.error != null)
+          ? chunk
+          : undefined,
+      isOutputChunk: isResponseOutputChunk,
+      url,
+      requestBodyValues: body,
+      responseHeaders,
+    });
+
     const self = this;
 
     const approvalRequestIdToDummyToolCallIdFromPrompt =
@@ -1026,6 +1150,7 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
             hasDiff: boolean;
             endEmitted: boolean;
           };
+          toolSearchExecution?: 'server' | 'client';
         }
       | undefined
     > = {};
@@ -1054,9 +1179,11 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
     > = {};
 
     let serviceTier: string | undefined;
+    const hostedToolSearchCallIds: string[] = [];
+    let encounteredStreamError = false;
 
-    return {
-      stream: response.pipeThrough(
+    const result = {
+      stream: checkedResponse.pipeThrough(
         new TransformStream<
           ParseResult<OpenAIResponsesChunk>,
           LanguageModelV4StreamPart
@@ -1188,6 +1315,28 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
                   input: '{}',
                   providerExecuted: true,
                 });
+              } else if (value.item.type === 'tool_search_call') {
+                const toolCallId = value.item.id;
+                const toolName =
+                  toolNameMapping.toCustomToolName('tool_search');
+                const isHosted = value.item.execution === 'server';
+
+                ongoingToolCalls[value.output_index] = {
+                  toolName,
+                  toolCallId,
+                  toolSearchExecution: value.item.execution ?? 'server',
+                };
+
+                if (isHosted) {
+                  controller.enqueue({
+                    type: 'tool-input-start',
+                    id: toolCallId,
+                    toolName,
+                    providerExecuted: true,
+                  });
+                }
+              } else if (value.item.type === 'tool_search_output') {
+                // handled on output_item.done so we can pair it with the call
               } else if (
                 value.item.type === 'mcp_call' ||
                 value.item.type === 'mcp_list_tools' ||
@@ -1305,6 +1454,13 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
                 controller.enqueue({
                   type: 'tool-input-end',
                   id: value.item.call_id,
+                  ...(value.item.namespace != null && {
+                    providerMetadata: {
+                      [providerOptionsName]: {
+                        namespace: value.item.namespace,
+                      },
+                    },
+                  }),
                 });
 
                 controller.enqueue({
@@ -1315,6 +1471,9 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
                   providerMetadata: {
                     [providerOptionsName]: {
                       itemId: value.item.id,
+                      ...(value.item.namespace != null && {
+                        namespace: value.item.namespace,
+                      }),
                     },
                   },
                 });
@@ -1418,6 +1577,67 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
                     result: value.item.result,
                   } satisfies InferSchema<typeof imageGenerationOutputSchema>,
                 });
+              } else if (value.item.type === 'tool_search_call') {
+                const toolCall = ongoingToolCalls[value.output_index];
+                const isHosted = value.item.execution === 'server';
+
+                if (toolCall != null) {
+                  const toolCallId = isHosted
+                    ? toolCall.toolCallId
+                    : (value.item.call_id ?? value.item.id);
+
+                  if (isHosted) {
+                    hostedToolSearchCallIds.push(toolCallId);
+                  } else {
+                    controller.enqueue({
+                      type: 'tool-input-start',
+                      id: toolCallId,
+                      toolName: toolCall.toolName,
+                    });
+                  }
+
+                  controller.enqueue({
+                    type: 'tool-input-end',
+                    id: toolCallId,
+                  });
+
+                  controller.enqueue({
+                    type: 'tool-call',
+                    toolCallId,
+                    toolName: toolCall.toolName,
+                    input: JSON.stringify({
+                      arguments: value.item.arguments,
+                      call_id: isHosted ? null : toolCallId,
+                    } satisfies InferSchema<typeof toolSearchInputSchema>),
+                    ...(isHosted ? { providerExecuted: true } : {}),
+                    providerMetadata: {
+                      [providerOptionsName]: {
+                        itemId: value.item.id,
+                      },
+                    },
+                  });
+                }
+
+                ongoingToolCalls[value.output_index] = undefined;
+              } else if (value.item.type === 'tool_search_output') {
+                const toolCallId =
+                  value.item.call_id ??
+                  hostedToolSearchCallIds.shift() ??
+                  value.item.id;
+
+                controller.enqueue({
+                  type: 'tool-result',
+                  toolCallId,
+                  toolName: toolNameMapping.toCustomToolName('tool_search'),
+                  result: {
+                    tools: value.item.tools,
+                  } satisfies InferSchema<typeof toolSearchOutputSchema>,
+                  providerMetadata: {
+                    [providerOptionsName]: {
+                      itemId: value.item.id,
+                    },
+                  },
+                });
               } else if (value.item.type === 'mcp_call') {
                 ongoingToolCalls[value.output_index] = undefined;
 
@@ -1647,6 +1867,18 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
                 }
 
                 delete activeReasoning[value.item.id];
+              } else if (value.item.type === 'compaction') {
+                controller.enqueue({
+                  type: 'custom',
+                  kind: 'openai.compaction',
+                  providerMetadata: {
+                    [providerOptionsName]: {
+                      type: 'compaction',
+                      itemId: value.item.id,
+                      encryptedContent: value.item.encrypted_content,
+                    } satisfies ResponsesCompactionProviderMetadata,
+                  },
+                });
               }
             } else if (isResponseFunctionCallArgumentsDeltaChunk(value)) {
               const toolCall = ongoingToolCalls[value.output_index];
@@ -1867,6 +2099,35 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
               if (typeof value.response.service_tier === 'string') {
                 serviceTier = value.response.service_tier;
               }
+            } else if (isResponseFailedChunk(value)) {
+              const incompleteReason =
+                value.response.incomplete_details?.reason;
+              finishReason = {
+                unified: incompleteReason
+                  ? mapOpenAIResponseFinishReason({
+                      finishReason: incompleteReason,
+                      hasFunctionCall,
+                    })
+                  : 'error',
+                raw: incompleteReason ?? 'error',
+              };
+              usage = value.response.usage ?? undefined;
+
+              if (!encounteredStreamError && value.response.error != null) {
+                encounteredStreamError = true;
+                controller.enqueue({
+                  type: 'error',
+                  error: {
+                    type: 'response.failed',
+                    sequence_number: value.sequence_number,
+                    response: {
+                      error: value.response.error,
+                      incomplete_details: value.response.incomplete_details,
+                      service_tier: value.response.service_tier,
+                    },
+                  },
+                });
+              }
             } else if (isResponseAnnotationAddedChunk(value)) {
               ongoingAnnotations.push(value.annotation);
               if (value.annotation.type === 'url_citation') {
@@ -1936,6 +2197,8 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
                 });
               }
             } else if (isErrorChunk(value)) {
+              encounteredStreamError = true;
+              finishReason = { unified: 'error', raw: 'error' };
               controller.enqueue({ type: 'error', error: value });
             }
           },
@@ -1961,6 +2224,8 @@ export class OpenAIResponsesLanguageModel implements LanguageModelV4 {
       request: { body },
       response: { headers: responseHeaders },
     };
+
+    return result;
   }
 }
 
@@ -1986,6 +2251,12 @@ function isResponseFinishedChunk(
   );
 }
 
+function isResponseFailedChunk(
+  chunk: OpenAIResponsesChunk,
+): chunk is OpenAIResponsesChunk & { type: 'response.failed' } {
+  return chunk.type === 'response.failed';
+}
+
 function isResponseCreatedChunk(
   chunk: OpenAIResponsesChunk,
 ): chunk is OpenAIResponsesChunk & { type: 'response.created' } {
@@ -2064,6 +2335,15 @@ function isErrorChunk(
   return chunk.type === 'error';
 }
 
+function isResponseOutputChunk(chunk: OpenAIResponsesChunk): boolean {
+  return !(
+    chunk.type === 'response.created' ||
+    chunk.type === 'response.failed' ||
+    chunk.type === 'error' ||
+    chunk.type === 'unknown_chunk'
+  );
+}
+
 function mapWebSearchOutput(
   action: OpenAIResponsesWebSearchAction | null | undefined,
 ): InferSchema<typeof webSearchOutputSchema> {
@@ -2074,7 +2354,11 @@ function mapWebSearchOutput(
   switch (action.type) {
     case 'search':
       return {
-        action: { type: 'search', query: action.query ?? undefined },
+        action: {
+          type: 'search',
+          query: action.query ?? undefined,
+          ...(action.queries != null && { queries: action.queries }),
+        },
         // include sources when provided by the Responses API (behind include flag)
         ...(action.sources != null && { sources: action.sources }),
       };