@draht/ai 2026.6.11 → 2026.7.7

package/dist/api/openai-completions.lazy.d.ts

@@ -0,0 +1,3 @@
+import type { ProviderStreams } from "../types.ts";
+export declare const openAICompletionsApi: () => ProviderStreams;
+//# sourceMappingURL=openai-completions.lazy.d.ts.map

package/dist/api/openai-completions.lazy.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"openai-completions.lazy.d.ts","sourceRoot":"","sources":["../../src/api/openai-completions.lazy.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAGnD,eAAO,MAAM,oBAAoB,uBAA0E,CAAC","sourcesContent":["import type { ProviderStreams } from \"../types.ts\";\nimport { lazyApi } from \"./lazy.ts\";\n\nexport const openAICompletionsApi = (): ProviderStreams => lazyApi(() => import(\"./openai-completions.ts\"));\n"]}

package/dist/api/openai-completions.lazy.js

@@ -0,0 +1,3 @@
+import { lazyApi } from "./lazy.js";
+export const openAICompletionsApi = () => lazyApi(() => import("./openai-completions.js"));
+//# sourceMappingURL=openai-completions.lazy.js.map

package/dist/api/openai-completions.lazy.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"openai-completions.lazy.js","sourceRoot":"","sources":["../../src/api/openai-completions.lazy.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAEpC,MAAM,CAAC,MAAM,oBAAoB,GAAG,GAAoB,EAAE,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC,yBAAyB,CAAC,CAAC,CAAC","sourcesContent":["import type { ProviderStreams } from \"../types.ts\";\nimport { lazyApi } from \"./lazy.ts\";\n\nexport const openAICompletionsApi = (): ProviderStreams => lazyApi(() => import(\"./openai-completions.ts\"));\n"]}

package/dist/api/openai-prompt-cache.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"openai-prompt-cache.d.ts","sourceRoot":"","sources":["../../src/api/openai-prompt-cache.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,kCAAkC,KAAK,CAAC;AAErD,wBAAgB,yBAAyB,CAAC,GAAG,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,GAAG,SAAS,CAKrF","sourcesContent":["export const OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH = 64;\n\nexport function clampOpenAIPromptCacheKey(key: string | undefined): string | undefined {\n\tif (key === undefined) return undefined;\n\tconst chars = Array.from(key);\n\tif (chars.length <= OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH) return key;\n\treturn chars.slice(0, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH).join(\"\");\n}\n"]}

package/dist/api/openai-prompt-cache.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"openai-prompt-cache.js","sourceRoot":"","sources":["../../src/api/openai-prompt-cache.ts"],"names":[],"mappings":"AAAA,MAAM,CAAC,MAAM,kCAAkC,GAAG,EAAE,CAAC;AAErD,MAAM,UAAU,yBAAyB,CAAC,GAAuB,EAAsB;IACtF,IAAI,GAAG,KAAK,SAAS;QAAE,OAAO,SAAS,CAAC;IACxC,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC9B,IAAI,KAAK,CAAC,MAAM,IAAI,kCAAkC;QAAE,OAAO,GAAG,CAAC;IACnE,OAAO,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,kCAAkC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAAA,CACnE","sourcesContent":["export const OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH = 64;\n\nexport function clampOpenAIPromptCacheKey(key: string | undefined): string | undefined {\n\tif (key === undefined) return undefined;\n\tconst chars = Array.from(key);\n\tif (chars.length <= OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH) return key;\n\treturn chars.slice(0, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH).join(\"\");\n}\n"]}

package/dist/api/openai-responses-shared.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"openai-responses-shared.d.ts","sourceRoot":"","sources":["../../src/api/openai-responses-shared.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EACX,IAAI,IAAI,UAAU,EAClB,6BAA6B,EAE7B,aAAa,EAOb,mBAAmB,EACnB,MAAM,yCAAyC,CAAC;AAEjD,OAAO,KAAK,EACX,GAAG,EACH,gBAAgB,EAChB,OAAO,EAEP,KAAK,EAKL,IAAI,EAEJ,KAAK,EACL,MAAM,aAAa,CAAC;AACrB,OAAO,KAAK,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AAoC5E,MAAM,WAAW,4BAA4B;IAC5C,WAAW,CAAC,EAAE,6BAA6B,CAAC,cAAc,CAAC,CAAC;IAC5D,kBAAkB,CAAC,EAAE,CACpB,mBAAmB,EAAE,6BAA6B,CAAC,cAAc,CAAC,GAAG,SAAS,EAC9E,kBAAkB,EAAE,6BAA6B,CAAC,cAAc,CAAC,GAAG,SAAS,KACzE,6BAA6B,CAAC,cAAc,CAAC,GAAG,SAAS,CAAC;IAC/D,uBAAuB,CAAC,EAAE,CACzB,KAAK,EAAE,KAAK,EACZ,WAAW,EAAE,6BAA6B,CAAC,cAAc,CAAC,GAAG,SAAS,KAClE,IAAI,CAAC;CACV;AAED,MAAM,WAAW,+BAA+B;IAC/C,mBAAmB,CAAC,EAAE,OAAO,CAAC;CAC9B;AAED,MAAM,WAAW,4BAA4B;IAC5C,MAAM,CAAC,EAAE,OAAO,GAAG,IAAI,CAAC;CACxB;AAMD,wBAAgB,wBAAwB,CAAC,IAAI,SAAS,GAAG,EACxD,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,wBAAwB,EAAE,WAAW,CAAC,MAAM,CAAC,EAC7C,OAAO,CAAC,EAAE,+BAA+B,GACvC,aAAa,CA6Kf;AAMD,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,IAAI,EAAE,EAAE,OAAO,CAAC,EAAE,4BAA4B,GAAG,UAAU,EAAE,CASzG;AAaD,wBAAsB,sBAAsB,CAAC,IAAI,SAAS,GAAG,EAC5D,YAAY,EAAE,aAAa,CAAC,mBAAmB,CAAC,EAChD,MAAM,EAAE,gBAAgB,EACxB,MAAM,EAAE,2BAA2B,EACnC,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,CAAC,EAAE,4BAA4B,GACpC,OAAO,CAAC,IAAI,CAAC,CAsOf","sourcesContent":["import type OpenAI from \"openai\";\nimport type {\n\tTool as OpenAITool,\n\tResponseCreateParamsStreaming,\n\tResponseFunctionCallOutputItemList,\n\tResponseInput,\n\tResponseInputContent,\n\tResponseInputImage,\n\tResponseInputText,\n\tResponseOutputItem,\n\tResponseOutputMessage,\n\tResponseReasoningItem,\n\tResponseStreamEvent,\n} from \"openai/resources/responses/responses.js\";\nimport { calculateCost } from \"../models.ts\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tImageContent,\n\tModel,\n\tStopReason,\n\tTextContent,\n\tTextSignatureV1,\n\tThinkingContent,\n\tTool,\n\tToolCall,\n\tUsage,\n} from \"../types.ts\";\nimport type { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { shortHash } from \"../utils/hash.ts\";\nimport { parseStreamingJson } from \"../utils/json-parse.ts\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.ts\";\nimport { transformMessages } from \"./transform-messages.ts\";\n\n// =============================================================================\n// Utilities\n// =============================================================================\n\nfunction encodeTextSignatureV1(id: string, phase?: TextSignatureV1[\"phase\"]): string {\n\tconst payload: TextSignatureV1 = { v: 1, id };\n\tif (phase) payload.phase = phase;\n\treturn JSON.stringify(payload);\n}\n\nfunction parseTextSignature(\n\tsignature: string | undefined,\n): { id: string; phase?: TextSignatureV1[\"phase\"] } | undefined {\n\tif (!signature) return undefined;\n\tif (signature.startsWith(\"{\")) {\n\t\ttry {\n\t\t\tconst parsed = JSON.parse(signature) as Partial<TextSignatureV1>;\n\t\t\tif (parsed.v === 1 && typeof parsed.id === \"string\") {\n\t\t\t\tif (parsed.phase === \"commentary\" || parsed.phase === \"final_answer\") {\n\t\t\t\t\treturn { id: parsed.id, phase: parsed.phase };\n\t\t\t\t}\n\t\t\t\treturn { id: parsed.id };\n\t\t\t}\n\t\t} catch {\n\t\t\t// Fall through to legacy plain-string handling.\n\t\t}\n\t}\n\treturn { id: signature };\n}\n\nexport interface OpenAIResponsesStreamOptions {\n\tserviceTier?: ResponseCreateParamsStreaming[\"service_tier\"];\n\tresolveServiceTier?: (\n\t\tresponseServiceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\t\trequestServiceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\t) => ResponseCreateParamsStreaming[\"service_tier\"] | undefined;\n\tapplyServiceTierPricing?: (\n\t\tusage: Usage,\n\t\tserviceTier: ResponseCreateParamsStreaming[\"service_tier\"] | undefined,\n\t) => void;\n}\n\nexport interface ConvertResponsesMessagesOptions {\n\tincludeSystemPrompt?: boolean;\n}\n\nexport interface ConvertResponsesToolsOptions {\n\tstrict?: boolean | null;\n}\n\n// =============================================================================\n// Message conversion\n// =============================================================================\n\nexport function convertResponsesMessages<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\tallowedToolCallProviders: ReadonlySet<string>,\n\toptions?: ConvertResponsesMessagesOptions,\n): ResponseInput {\n\tconst messages: ResponseInput = [];\n\n\tconst normalizeIdPart = (part: string): string => {\n\t\tconst sanitized = part.replace(/[^a-zA-Z0-9_-]/g, \"_\");\n\t\tconst normalized = sanitized.length > 64 ? sanitized.slice(0, 64) : sanitized;\n\t\treturn normalized.replace(/_+$/, \"\");\n\t};\n\n\tconst buildForeignResponsesItemId = (itemId: string): string => {\n\t\tconst normalized = `fc_${shortHash(itemId)}`;\n\t\treturn normalized.length > 64 ? normalized.slice(0, 64) : normalized;\n\t};\n\n\tconst normalizeToolCallId = (id: string, _targetModel: Model<TApi>, source: AssistantMessage): string => {\n\t\tif (!allowedToolCallProviders.has(model.provider)) return normalizeIdPart(id);\n\t\tif (!id.includes(\"|\")) return normalizeIdPart(id);\n\t\tconst [callId, itemId] = id.split(\"|\");\n\t\tconst normalizedCallId = normalizeIdPart(callId);\n\t\tconst isForeignToolCall = source.provider !== model.provider || source.api !== model.api;\n\t\tlet normalizedItemId = isForeignToolCall ? buildForeignResponsesItemId(itemId) : normalizeIdPart(itemId);\n\t\t// OpenAI Responses API requires item id to start with \"fc\"\n\t\tif (!normalizedItemId.startsWith(\"fc_\")) {\n\t\t\tnormalizedItemId = normalizeIdPart(`fc_${normalizedItemId}`);\n\t\t}\n\t\treturn `${normalizedCallId}|${normalizedItemId}`;\n\t};\n\n\tconst transformedMessages = transformMessages(context.messages, model, normalizeToolCallId);\n\n\tconst includeSystemPrompt = options?.includeSystemPrompt ?? true;\n\tif (includeSystemPrompt && context.systemPrompt) {\n\t\tconst compat = model.compat as { supportsDeveloperRole?: boolean } | undefined;\n\t\tconst role = model.reasoning && compat?.supportsDeveloperRole !== false ? \"developer\" : \"system\";\n\t\tmessages.push({\n\t\t\trole,\n\t\t\tcontent: sanitizeSurrogates(context.systemPrompt),\n\t\t});\n\t}\n\n\tlet msgIndex = 0;\n\tfor (const msg of transformedMessages) {\n\t\tif (msg.role === \"user\") {\n\t\t\tif (typeof msg.content === \"string\") {\n\t\t\t\tmessages.push({\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent: [{ type: \"input_text\", text: sanitizeSurrogates(msg.content) }],\n\t\t\t\t});\n\t\t\t} else {\n\t\t\t\tconst content: ResponseInputContent[] = msg.content.map((item): ResponseInputContent => {\n\t\t\t\t\tif (item.type === \"text\") {\n\t\t\t\t\t\treturn {\n\t\t\t\t\t\t\ttype: \"input_text\",\n\t\t\t\t\t\t\ttext: sanitizeSurrogates(item.text),\n\t\t\t\t\t\t} satisfies ResponseInputText;\n\t\t\t\t\t}\n\t\t\t\t\treturn {\n\t\t\t\t\t\ttype: \"input_image\",\n\t\t\t\t\t\tdetail: \"auto\",\n\t\t\t\t\t\timage_url: `data:${item.mimeType};base64,${item.data}`,\n\t\t\t\t\t} satisfies ResponseInputImage;\n\t\t\t\t});\n\t\t\t\tif (content.length === 0) continue;\n\t\t\t\tmessages.push({\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent,\n\t\t\t\t});\n\t\t\t}\n\t\t} else if (msg.role === \"assistant\") {\n\t\t\tconst output: ResponseInput = [];\n\t\t\tconst assistantMsg = msg as AssistantMessage;\n\t\t\tconst isDifferentModel =\n\t\t\t\tassistantMsg.model !== model.id &&\n\t\t\t\tassistantMsg.provider === model.provider &&\n\t\t\t\tassistantMsg.api === model.api;\n\t\t\tlet textBlockIndex = 0;\n\n\t\t\tfor (const block of msg.content) {\n\t\t\t\tif (block.type === \"thinking\") {\n\t\t\t\t\tif (block.thinking.trim().length === 0) continue;\n\t\t\t\t\tif (block.thinkingSignature) {\n\t\t\t\t\t\tconst reasoningItem = JSON.parse(block.thinkingSignature) as ResponseReasoningItem;\n\t\t\t\t\t\toutput.push(reasoningItem);\n\t\t\t\t\t}\n\t\t\t\t} else if (block.type === \"text\") {\n\t\t\t\t\tconst textBlock = block as TextContent;\n\t\t\t\t\tconst parsedSignature = parseTextSignature(textBlock.textSignature);\n\t\t\t\t\tconst fallbackMessageId =\n\t\t\t\t\t\ttextBlockIndex === 0 ? `msg_pi_${msgIndex}` : `msg_pi_${msgIndex}_${textBlockIndex}`;\n\t\t\t\t\ttextBlockIndex++;\n\t\t\t\t\t// OpenAI requires id to be max 64 characters\n\t\t\t\t\tlet msgId = parsedSignature?.id;\n\t\t\t\t\tif (!msgId) {\n\t\t\t\t\t\tmsgId = fallbackMessageId;\n\t\t\t\t\t} else if (msgId.length > 64) {\n\t\t\t\t\t\tmsgId = `msg_${shortHash(msgId)}`;\n\t\t\t\t\t}\n\t\t\t\t\toutput.push({\n\t\t\t\t\t\ttype: \"message\",\n\t\t\t\t\t\trole: \"assistant\",\n\t\t\t\t\t\tcontent: [{ type: \"output_text\", text: sanitizeSurrogates(textBlock.text), annotations: [] }],\n\t\t\t\t\t\tstatus: \"completed\",\n\t\t\t\t\t\tid: msgId,\n\t\t\t\t\t\tphase: parsedSignature?.phase,\n\t\t\t\t\t} satisfies ResponseOutputMessage);\n\t\t\t\t} else if (block.type === \"toolCall\") {\n\t\t\t\t\tconst toolCall = block as ToolCall;\n\t\t\t\t\tconst [callId, itemIdRaw] = toolCall.id.split(\"|\");\n\t\t\t\t\tlet itemId: string | undefined = itemIdRaw;\n\n\t\t\t\t\t// For different-model messages, set id to undefined to avoid pairing validation.\n\t\t\t\t\t// OpenAI tracks which fc_xxx IDs were paired with rs_xxx reasoning items.\n\t\t\t\t\t// By omitting the id, we avoid triggering that validation (like cross-provider does).\n\t\t\t\t\tif (isDifferentModel && itemId?.startsWith(\"fc_\")) {\n\t\t\t\t\t\titemId = undefined;\n\t\t\t\t\t}\n\n\t\t\t\t\toutput.push({\n\t\t\t\t\t\ttype: \"function_call\",\n\t\t\t\t\t\tid: itemId,\n\t\t\t\t\t\tcall_id: callId,\n\t\t\t\t\t\tname: toolCall.name,\n\t\t\t\t\t\targuments: JSON.stringify(toolCall.arguments),\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (output.length === 0) continue;\n\t\t\tmessages.push(...output);\n\t\t} else if (msg.role === \"toolResult\") {\n\t\t\tconst textResult = msg.content\n\t\t\t\t.filter((c): c is TextContent => c.type === \"text\")\n\t\t\t\t.map((c) => c.text)\n\t\t\t\t.join(\"\\n\");\n\t\t\tconst hasImages = msg.content.some((c): c is ImageContent => c.type === \"image\");\n\t\t\tconst hasText = textResult.length > 0;\n\t\t\tconst [callId] = msg.toolCallId.split(\"|\");\n\n\t\t\tlet output: string | ResponseFunctionCallOutputItemList;\n\t\t\tif (hasImages && model.input.includes(\"image\")) {\n\t\t\t\tconst contentParts: ResponseFunctionCallOutputItemList = [];\n\n\t\t\t\tif (hasText) {\n\t\t\t\t\tcontentParts.push({\n\t\t\t\t\t\ttype: \"input_text\",\n\t\t\t\t\t\ttext: sanitizeSurrogates(textResult),\n\t\t\t\t\t});\n\t\t\t\t}\n\n\t\t\t\tfor (const block of msg.content) {\n\t\t\t\t\tif (block.type === \"image\") {\n\t\t\t\t\t\tcontentParts.push({\n\t\t\t\t\t\t\ttype: \"input_image\",\n\t\t\t\t\t\t\tdetail: \"auto\",\n\t\t\t\t\t\t\timage_url: `data:${block.mimeType};base64,${block.data}`,\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\toutput = contentParts;\n\t\t\t} else {\n\t\t\t\toutput = sanitizeSurrogates(hasText ? textResult : hasImages ? \"(see attached image)\" : \"(no tool output)\");\n\t\t\t}\n\n\t\t\tmessages.push({\n\t\t\t\ttype: \"function_call_output\",\n\t\t\t\tcall_id: callId,\n\t\t\t\toutput,\n\t\t\t});\n\t\t}\n\t\tmsgIndex++;\n\t}\n\n\treturn messages;\n}\n\n// =============================================================================\n// Tool conversion\n// =============================================================================\n\nexport function convertResponsesTools(tools: Tool[], options?: ConvertResponsesToolsOptions): OpenAITool[] {\n\tconst strict = options?.strict === undefined ? false : options.strict;\n\treturn tools.map((tool) => ({\n\t\ttype: \"function\",\n\t\tname: tool.name,\n\t\tdescription: tool.description,\n\t\tparameters: tool.parameters as any, // TypeBox already generates JSON Schema\n\t\tstrict,\n\t}));\n}\n\n// =============================================================================\n// Stream processing\n// =============================================================================\n\ntype StreamingToolCall = ToolCall & { partialJson: string };\n\ntype ResponsesOutputSlot =\n\t| { type: \"thinking\"; block: ThinkingContent; contentIndex: number }\n\t| { type: \"text\"; block: TextContent; contentIndex: number }\n\t| { type: \"toolCall\"; block: StreamingToolCall; contentIndex: number };\n\nexport async function processResponsesStream<TApi extends Api>(\n\topenaiStream: AsyncIterable<ResponseStreamEvent>,\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tmodel: Model<TApi>,\n\toptions?: OpenAIResponsesStreamOptions,\n): Promise<void> {\n\tlet sawTerminalResponseEvent = false;\n\tconst outputSlots = new Map<number, ResponsesOutputSlot>();\n\tconst getSlot = <TType extends ResponsesOutputSlot[\"type\"]>(\n\t\toutputIndex: number,\n\t\ttype: TType,\n\t): Extract<ResponsesOutputSlot, { type: TType }> | undefined => {\n\t\tconst slot = outputSlots.get(outputIndex);\n\t\treturn slot?.type === type ? (slot as Extract<ResponsesOutputSlot, { type: TType }>) : undefined;\n\t};\n\tconst createSlot = (outputIndex: number, item: ResponseOutputItem): ResponsesOutputSlot | undefined => {\n\t\tif (item.type === \"reasoning\") {\n\t\t\tconst block: ThinkingContent = { type: \"thinking\", thinking: \"\" };\n\t\t\toutput.content.push(block);\n\t\t\tconst slot = {\n\t\t\t\ttype: \"thinking\",\n\t\t\t\tblock,\n\t\t\t\tcontentIndex: output.content.length - 1,\n\t\t\t} satisfies ResponsesOutputSlot;\n\t\t\toutputSlots.set(outputIndex, slot);\n\t\t\tstream.push({ type: \"thinking_start\", contentIndex: slot.contentIndex, partial: output });\n\t\t\treturn slot;\n\t\t}\n\t\tif (item.type === \"message\") {\n\t\t\tconst block: TextContent = { type: \"text\", text: \"\" };\n\t\t\toutput.content.push(block);\n\t\t\tconst slot = { type: \"text\", block, contentIndex: output.content.length - 1 } satisfies ResponsesOutputSlot;\n\t\t\toutputSlots.set(outputIndex, slot);\n\t\t\tstream.push({ type: \"text_start\", contentIndex: slot.contentIndex, partial: output });\n\t\t\treturn slot;\n\t\t}\n\t\tif (item.type === \"function_call\") {\n\t\t\tconst block: StreamingToolCall = {\n\t\t\t\ttype: \"toolCall\",\n\t\t\t\tid: `${item.call_id}|${item.id}`,\n\t\t\t\tname: item.name,\n\t\t\t\targuments: {},\n\t\t\t\tpartialJson: item.arguments || \"\",\n\t\t\t};\n\t\t\toutput.content.push(block);\n\t\t\tconst slot = {\n\t\t\t\ttype: \"toolCall\",\n\t\t\t\tblock,\n\t\t\t\tcontentIndex: output.content.length - 1,\n\t\t\t} satisfies ResponsesOutputSlot;\n\t\t\toutputSlots.set(outputIndex, slot);\n\t\t\tstream.push({ type: \"toolcall_start\", contentIndex: slot.contentIndex, partial: output });\n\t\t\treturn slot;\n\t\t}\n\t\treturn undefined;\n\t};\n\tconst getOrCreateSlot = (outputIndex: number, item: ResponseOutputItem): ResponsesOutputSlot | undefined => {\n\t\treturn outputSlots.get(outputIndex) ?? createSlot(outputIndex, item);\n\t};\n\tconst finalizeResponse = (\n\t\tresponse: Extract<ResponseStreamEvent, { type: \"response.completed\" | \"response.incomplete\" }>[\"response\"],\n\t): void => {\n\t\tsawTerminalResponseEvent = true;\n\t\tif (response?.id) {\n\t\t\toutput.responseId = response.id;\n\t\t}\n\t\tif (response?.usage) {\n\t\t\tconst cachedTokens = response.usage.input_tokens_details?.cached_tokens || 0;\n\t\t\toutput.usage = {\n\t\t\t\t// OpenAI includes cached tokens in input_tokens, so subtract to get non-cached input\n\t\t\t\tinput: (response.usage.input_tokens || 0) - cachedTokens,\n\t\t\t\toutput: response.usage.output_tokens || 0,\n\t\t\t\tcacheRead: cachedTokens,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\treasoning: response.usage.output_tokens_details?.reasoning_tokens || 0,\n\t\t\t\ttotalTokens: response.usage.total_tokens || 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t};\n\t\t}\n\t\tcalculateCost(model, output.usage);\n\t\tif (options?.applyServiceTierPricing) {\n\t\t\tconst serviceTier = options.resolveServiceTier\n\t\t\t\t? options.resolveServiceTier(response?.service_tier, options.serviceTier)\n\t\t\t\t: (response?.service_tier ?? options.serviceTier);\n\t\t\toptions.applyServiceTierPricing(output.usage, serviceTier);\n\t\t}\n\t\t// Map status to stop reason\n\t\toutput.stopReason = mapStopReason(response?.status);\n\t\tif (output.content.some((b) => b.type === \"toolCall\") && output.stopReason === \"stop\") {\n\t\t\toutput.stopReason = \"toolUse\";\n\t\t}\n\t};\n\n\tfor await (const event of openaiStream) {\n\t\tif (event.type === \"response.created\") {\n\t\t\toutput.responseId = event.response.id;\n\t\t} else if (event.type === \"response.output_item.added\") {\n\t\t\tcreateSlot(event.output_index, event.item);\n\t\t} else if (event.type === \"response.reasoning_summary_text.delta\") {\n\t\t\tconst slot = getSlot(event.output_index, \"thinking\");\n\t\t\tif (!slot) continue;\n\t\t\tslot.block.thinking += event.delta;\n\t\t\tstream.push({\n\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\tdelta: event.delta,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t} else if (event.type === \"response.reasoning_summary_part.done\") {\n\t\t\tconst slot = getSlot(event.output_index, \"thinking\");\n\t\t\tif (!slot) continue;\n\t\t\tslot.block.thinking += \"\\n\\n\";\n\t\t\tstream.push({\n\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\tdelta: \"\\n\\n\",\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t} else if (event.type === \"response.reasoning_text.delta\") {\n\t\t\tconst slot = getSlot(event.output_index, \"thinking\");\n\t\t\tif (!slot) continue;\n\t\t\tslot.block.thinking += event.delta;\n\t\t\tstream.push({\n\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\tdelta: event.delta,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t} else if (event.type === \"response.output_text.delta\") {\n\t\t\tconst slot = getSlot(event.output_index, \"text\");\n\t\t\tif (!slot) continue;\n\t\t\tslot.block.text += event.delta;\n\t\t\tstream.push({\n\t\t\t\ttype: \"text_delta\",\n\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\tdelta: event.delta,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t} else if (event.type === \"response.refusal.delta\") {\n\t\t\tconst slot = getSlot(event.output_index, \"text\");\n\t\t\tif (!slot) continue;\n\t\t\tslot.block.text += event.delta;\n\t\t\tstream.push({\n\t\t\t\ttype: \"text_delta\",\n\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\tdelta: event.delta,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t} else if (event.type === \"response.function_call_arguments.delta\") {\n\t\t\tconst slot = getSlot(event.output_index, \"toolCall\");\n\t\t\tif (!slot) continue;\n\t\t\tslot.block.partialJson += event.delta;\n\t\t\tslot.block.arguments = parseStreamingJson(slot.block.partialJson);\n\t\t\tstream.push({\n\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\tdelta: event.delta,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t} else if (event.type === \"response.function_call_arguments.done\") {\n\t\t\tconst slot = getSlot(event.output_index, \"toolCall\");\n\t\t\tif (!slot) continue;\n\t\t\tconst previousPartialJson = slot.block.partialJson;\n\t\t\tslot.block.partialJson = event.arguments;\n\t\t\tslot.block.arguments = parseStreamingJson(slot.block.partialJson);\n\n\t\t\tif (event.arguments.startsWith(previousPartialJson)) {\n\t\t\t\tconst delta = event.arguments.slice(previousPartialJson.length);\n\t\t\t\tif (delta.length > 0) {\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\t\t\tdelta,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t}\n\t\t} else if (event.type === \"response.output_item.done\") {\n\t\t\tconst item = event.item;\n\t\t\tconst slot = getOrCreateSlot(event.output_index, item);\n\n\t\t\tif (item.type === \"reasoning\" && slot?.type === \"thinking\") {\n\t\t\t\tconst summaryText = item.summary?.map((s) => s.text).join(\"\\n\\n\") || \"\";\n\t\t\t\tconst contentText = item.content?.map((c) => c.text).join(\"\\n\\n\") || \"\";\n\t\t\t\tslot.block.thinking = summaryText || contentText || slot.block.thinking;\n\t\t\t\tslot.block.thinkingSignature = JSON.stringify(item);\n\t\t\t\tstream.push({\n\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\t\tcontent: slot.block.thinking,\n\t\t\t\t\tpartial: output,\n\t\t\t\t});\n\t\t\t\toutputSlots.delete(event.output_index);\n\t\t\t} else if (item.type === \"message\" && slot?.type === \"text\") {\n\t\t\t\tslot.block.text = item.content?.map((c) => (c.type === \"output_text\" ? c.text : c.refusal)).join(\"\") || \"\";\n\t\t\t\tslot.block.textSignature = encodeTextSignatureV1(item.id, item.phase ?? undefined);\n\t\t\t\tstream.push({\n\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\t\tcontent: slot.block.text,\n\t\t\t\t\tpartial: output,\n\t\t\t\t});\n\t\t\t\toutputSlots.delete(event.output_index);\n\t\t\t} else if (item.type === \"function_call\" && slot?.type === \"toolCall\") {\n\t\t\t\tslot.block.arguments = parseStreamingJson(item.arguments || slot.block.partialJson || \"{}\");\n\t\t\t\t// Finalize in-place and strip the scratch buffer so replay only\n\t\t\t\t// carries parsed arguments.\n\t\t\t\tdelete (slot.block as { partialJson?: string }).partialJson;\n\t\t\t\tstream.push({\n\t\t\t\t\ttype: \"toolcall_end\",\n\t\t\t\t\tcontentIndex: slot.contentIndex,\n\t\t\t\t\ttoolCall: slot.block,\n\t\t\t\t\tpartial: output,\n\t\t\t\t});\n\t\t\t\toutputSlots.delete(event.output_index);\n\t\t\t}\n\t\t} else if (event.type === \"response.completed\" || event.type === \"response.incomplete\") {\n\t\t\tfinalizeResponse(event.response);\n\t\t} else if (event.type === \"error\") {\n\t\t\tthrow new Error(`Error Code ${event.code}: ${event.message}` || \"Unknown error\");\n\t\t} else if (event.type === \"response.failed\") {\n\t\t\tsawTerminalResponseEvent = true;\n\t\t\tconst error = event.response?.error;\n\t\t\tconst details = event.response?.incomplete_details;\n\t\t\tconst msg = error\n\t\t\t\t? `${error.code || \"unknown\"}: ${error.message || \"no message\"}`\n\t\t\t\t: details?.reason\n\t\t\t\t\t? `incomplete: ${details.reason}`\n\t\t\t\t\t: \"Unknown error (no error details in response)\";\n\t\t\tthrow new Error(msg);\n\t\t}\n\t}\n\tif (!sawTerminalResponseEvent) {\n\t\tthrow new Error(\"OpenAI Responses stream ended before a terminal response event\");\n\t}\n}\n\nfunction mapStopReason(status: OpenAI.Responses.ResponseStatus | undefined): StopReason {\n\tif (!status) return \"stop\";\n\tswitch (status) {\n\t\tcase \"completed\":\n\t\t\treturn \"stop\";\n\t\tcase \"incomplete\":\n\t\t\treturn \"length\";\n\t\tcase \"failed\":\n\t\tcase \"cancelled\":\n\t\t\treturn \"error\";\n\t\t// These two are wonky ...\n\t\tcase \"in_progress\":\n\t\tcase \"queued\":\n\t\t\treturn \"stop\";\n\t\tdefault: {\n\t\t\tconst _exhaustive: never = status;\n\t\t\tthrow new Error(`Unhandled stop reason: ${_exhaustive}`);\n\t\t}\n\t}\n}\n"]}

package/dist/{providers → api}/openai-responses-shared.js

@@ -191,7 +191,7 @@ export function convertResponsesMessages(model, context, allowedToolCallProvider
                 output = contentParts;
             }
             else {
-                output = sanitizeSurrogates(hasText ? textResult : "(see attached image)");
+                output = sanitizeSurrogates(hasText ? textResult : hasImages ? "(see attached image)" : "(no tool output)");
             }
             messages.push({
                 type: "function_call_output",
@@ -216,253 +216,236 @@ export function convertResponsesTools(tools, options) {
         strict,
     }));
 }
-// =============================================================================
-// Stream processing
-// =============================================================================
 export async function processResponsesStream(openaiStream, output, stream, model, options) {
-    let currentItem = null;
-    let currentBlock = null;
-    const blocks = output.content;
-    const blockIndex = () => blocks.length - 1;
+    let sawTerminalResponseEvent = false;
+    const outputSlots = new Map();
+    const getSlot = (outputIndex, type) => {
+        const slot = outputSlots.get(outputIndex);
+        return slot?.type === type ? slot : undefined;
+    };
+    const createSlot = (outputIndex, item) => {
+        if (item.type === "reasoning") {
+            const block = { type: "thinking", thinking: "" };
+            output.content.push(block);
+            const slot = {
+                type: "thinking",
+                block,
+                contentIndex: output.content.length - 1,
+            };
+            outputSlots.set(outputIndex, slot);
+            stream.push({ type: "thinking_start", contentIndex: slot.contentIndex, partial: output });
+            return slot;
+        }
+        if (item.type === "message") {
+            const block = { type: "text", text: "" };
+            output.content.push(block);
+            const slot = { type: "text", block, contentIndex: output.content.length - 1 };
+            outputSlots.set(outputIndex, slot);
+            stream.push({ type: "text_start", contentIndex: slot.contentIndex, partial: output });
+            return slot;
+        }
+        if (item.type === "function_call") {
+            const block = {
+                type: "toolCall",
+                id: `${item.call_id}|${item.id}`,
+                name: item.name,
+                arguments: {},
+                partialJson: item.arguments || "",
+            };
+            output.content.push(block);
+            const slot = {
+                type: "toolCall",
+                block,
+                contentIndex: output.content.length - 1,
+            };
+            outputSlots.set(outputIndex, slot);
+            stream.push({ type: "toolcall_start", contentIndex: slot.contentIndex, partial: output });
+            return slot;
+        }
+        return undefined;
+    };
+    const getOrCreateSlot = (outputIndex, item) => {
+        return outputSlots.get(outputIndex) ?? createSlot(outputIndex, item);
+    };
+    const finalizeResponse = (response) => {
+        sawTerminalResponseEvent = true;
+        if (response?.id) {
+            output.responseId = response.id;
+        }
+        if (response?.usage) {
+            const cachedTokens = response.usage.input_tokens_details?.cached_tokens || 0;
+            output.usage = {
+                // OpenAI includes cached tokens in input_tokens, so subtract to get non-cached input
+                input: (response.usage.input_tokens || 0) - cachedTokens,
+                output: response.usage.output_tokens || 0,
+                cacheRead: cachedTokens,
+                cacheWrite: 0,
+                reasoning: response.usage.output_tokens_details?.reasoning_tokens || 0,
+                totalTokens: response.usage.total_tokens || 0,
+                cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+            };
+        }
+        calculateCost(model, output.usage);
+        if (options?.applyServiceTierPricing) {
+            const serviceTier = options.resolveServiceTier
+                ? options.resolveServiceTier(response?.service_tier, options.serviceTier)
+                : (response?.service_tier ?? options.serviceTier);
+            options.applyServiceTierPricing(output.usage, serviceTier);
+        }
+        // Map status to stop reason
+        output.stopReason = mapStopReason(response?.status);
+        if (output.content.some((b) => b.type === "toolCall") && output.stopReason === "stop") {
+            output.stopReason = "toolUse";
+        }
+    };
     for await (const event of openaiStream) {
         if (event.type === "response.created") {
             output.responseId = event.response.id;
         }
         else if (event.type === "response.output_item.added") {
-            const item = event.item;
-            if (item.type === "reasoning") {
-                currentItem = item;
-                currentBlock = { type: "thinking", thinking: "" };
-                output.content.push(currentBlock);
-                stream.push({ type: "thinking_start", contentIndex: blockIndex(), partial: output });
-            }
-            else if (item.type === "message") {
-                currentItem = item;
-                currentBlock = { type: "text", text: "" };
-                output.content.push(currentBlock);
-                stream.push({ type: "text_start", contentIndex: blockIndex(), partial: output });
-            }
-            else if (item.type === "function_call") {
-                currentItem = item;
-                currentBlock = {
-                    type: "toolCall",
-                    id: `${item.call_id}|${item.id}`,
-                    name: item.name,
-                    arguments: {},
-                    partialJson: item.arguments || "",
-                };
-                output.content.push(currentBlock);
-                stream.push({ type: "toolcall_start", contentIndex: blockIndex(), partial: output });
-            }
-        }
-        else if (event.type === "response.reasoning_summary_part.added") {
-            if (currentItem && currentItem.type === "reasoning") {
-                currentItem.summary = currentItem.summary || [];
-                currentItem.summary.push(event.part);
-            }
+            createSlot(event.output_index, event.item);
         }
         else if (event.type === "response.reasoning_summary_text.delta") {
-            if (currentItem?.type === "reasoning" && currentBlock?.type === "thinking") {
-                currentItem.summary = currentItem.summary || [];
-                const lastPart = currentItem.summary[currentItem.summary.length - 1];
-                if (lastPart) {
-                    currentBlock.thinking += event.delta;
-                    lastPart.text += event.delta;
-                    stream.push({
-                        type: "thinking_delta",
-                        contentIndex: blockIndex(),
-                        delta: event.delta,
-                        partial: output,
-                    });
-                }
-            }
+            const slot = getSlot(event.output_index, "thinking");
+            if (!slot)
+                continue;
+            slot.block.thinking += event.delta;
+            stream.push({
+                type: "thinking_delta",
+                contentIndex: slot.contentIndex,
+                delta: event.delta,
+                partial: output,
+            });
         }
         else if (event.type === "response.reasoning_summary_part.done") {
-            if (currentItem?.type === "reasoning" && currentBlock?.type === "thinking") {
-                currentItem.summary = currentItem.summary || [];
-                const lastPart = currentItem.summary[currentItem.summary.length - 1];
-                if (lastPart) {
-                    currentBlock.thinking += "\n\n";
-                    lastPart.text += "\n\n";
-                    stream.push({
-                        type: "thinking_delta",
-                        contentIndex: blockIndex(),
-                        delta: "\n\n",
-                        partial: output,
-                    });
-                }
-            }
+            const slot = getSlot(event.output_index, "thinking");
+            if (!slot)
+                continue;
+            slot.block.thinking += "\n\n";
+            stream.push({
+                type: "thinking_delta",
+                contentIndex: slot.contentIndex,
+                delta: "\n\n",
+                partial: output,
+            });
         }
         else if (event.type === "response.reasoning_text.delta") {
-            if (currentItem?.type === "reasoning" && currentBlock?.type === "thinking") {
-                currentBlock.thinking += event.delta;
-                stream.push({
-                    type: "thinking_delta",
-                    contentIndex: blockIndex(),
-                    delta: event.delta,
-                    partial: output,
-                });
-            }
-        }
-        else if (event.type === "response.content_part.added") {
-            if (currentItem?.type === "message") {
-                currentItem.content = currentItem.content || [];
-                // Filter out ReasoningText, only accept output_text and refusal
-                if (event.part.type === "output_text" || event.part.type === "refusal") {
-                    currentItem.content.push(event.part);
-                }
-            }
+            const slot = getSlot(event.output_index, "thinking");
+            if (!slot)
+                continue;
+            slot.block.thinking += event.delta;
+            stream.push({
+                type: "thinking_delta",
+                contentIndex: slot.contentIndex,
+                delta: event.delta,
+                partial: output,
+            });
         }
         else if (event.type === "response.output_text.delta") {
-            if (currentItem?.type === "message" && currentBlock?.type === "text") {
-                if (!currentItem.content || currentItem.content.length === 0) {
-                    continue;
-                }
-                const lastPart = currentItem.content[currentItem.content.length - 1];
-                if (lastPart?.type === "output_text") {
-                    currentBlock.text += event.delta;
-                    lastPart.text += event.delta;
-                    stream.push({
-                        type: "text_delta",
-                        contentIndex: blockIndex(),
-                        delta: event.delta,
-                        partial: output,
-                    });
-                }
-            }
+            const slot = getSlot(event.output_index, "text");
+            if (!slot)
+                continue;
+            slot.block.text += event.delta;
+            stream.push({
+                type: "text_delta",
+                contentIndex: slot.contentIndex,
+                delta: event.delta,
+                partial: output,
+            });
         }
         else if (event.type === "response.refusal.delta") {
-            if (currentItem?.type === "message" && currentBlock?.type === "text") {
-                if (!currentItem.content || currentItem.content.length === 0) {
-                    continue;
-                }
-                const lastPart = currentItem.content[currentItem.content.length - 1];
-                if (lastPart?.type === "refusal") {
-                    currentBlock.text += event.delta;
-                    lastPart.refusal += event.delta;
-                    stream.push({
-                        type: "text_delta",
-                        contentIndex: blockIndex(),
-                        delta: event.delta,
-                        partial: output,
-                    });
-                }
-            }
+            const slot = getSlot(event.output_index, "text");
+            if (!slot)
+                continue;
+            slot.block.text += event.delta;
+            stream.push({
+                type: "text_delta",
+                contentIndex: slot.contentIndex,
+                delta: event.delta,
+                partial: output,
+            });
         }
         else if (event.type === "response.function_call_arguments.delta") {
-            if (currentItem?.type === "function_call" && currentBlock?.type === "toolCall") {
-                currentBlock.partialJson += event.delta;
-                currentBlock.arguments = parseStreamingJson(currentBlock.partialJson);
-                stream.push({
-                    type: "toolcall_delta",
-                    contentIndex: blockIndex(),
-                    delta: event.delta,
-                    partial: output,
-                });
-            }
+            const slot = getSlot(event.output_index, "toolCall");
+            if (!slot)
+                continue;
+            slot.block.partialJson += event.delta;
+            slot.block.arguments = parseStreamingJson(slot.block.partialJson);
+            stream.push({
+                type: "toolcall_delta",
+                contentIndex: slot.contentIndex,
+                delta: event.delta,
+                partial: output,
+            });
         }
         else if (event.type === "response.function_call_arguments.done") {
-            if (currentItem?.type === "function_call" && currentBlock?.type === "toolCall") {
-                const previousPartialJson = currentBlock.partialJson;
-                currentBlock.partialJson = event.arguments;
-                currentBlock.arguments = parseStreamingJson(currentBlock.partialJson);
-                if (event.arguments.startsWith(previousPartialJson)) {
-                    const delta = event.arguments.slice(previousPartialJson.length);
-                    if (delta.length > 0) {
-                        stream.push({
-                            type: "toolcall_delta",
-                            contentIndex: blockIndex(),
-                            delta,
-                            partial: output,
-                        });
-                    }
+            const slot = getSlot(event.output_index, "toolCall");
+            if (!slot)
+                continue;
+            const previousPartialJson = slot.block.partialJson;
+            slot.block.partialJson = event.arguments;
+            slot.block.arguments = parseStreamingJson(slot.block.partialJson);
+            if (event.arguments.startsWith(previousPartialJson)) {
+                const delta = event.arguments.slice(previousPartialJson.length);
+                if (delta.length > 0) {
+                    stream.push({
+                        type: "toolcall_delta",
+                        contentIndex: slot.contentIndex,
+                        delta,
+                        partial: output,
+                    });
                 }
             }
         }
         else if (event.type === "response.output_item.done") {
             const item = event.item;
-            if (item.type === "reasoning" && currentBlock?.type === "thinking") {
+            const slot = getOrCreateSlot(event.output_index, item);
+            if (item.type === "reasoning" && slot?.type === "thinking") {
                 const summaryText = item.summary?.map((s) => s.text).join("\n\n") || "";
                 const contentText = item.content?.map((c) => c.text).join("\n\n") || "";
-                currentBlock.thinking = summaryText || contentText || currentBlock.thinking;
-                currentBlock.thinkingSignature = JSON.stringify(item);
+                slot.block.thinking = summaryText || contentText || slot.block.thinking;
+                slot.block.thinkingSignature = JSON.stringify(item);
                 stream.push({
                     type: "thinking_end",
-                    contentIndex: blockIndex(),
-                    content: currentBlock.thinking,
+                    contentIndex: slot.contentIndex,
+                    content: slot.block.thinking,
                     partial: output,
                 });
-                currentBlock = null;
+                outputSlots.delete(event.output_index);
             }
-            else if (item.type === "message" && currentBlock?.type === "text") {
-                currentBlock.text = item.content.map((c) => (c.type === "output_text" ? c.text : c.refusal)).join("");
-                currentBlock.textSignature = encodeTextSignatureV1(item.id, item.phase ?? undefined);
+            else if (item.type === "message" && slot?.type === "text") {
+                slot.block.text = item.content?.map((c) => (c.type === "output_text" ? c.text : c.refusal)).join("") || "";
+                slot.block.textSignature = encodeTextSignatureV1(item.id, item.phase ?? undefined);
                 stream.push({
                     type: "text_end",
-                    contentIndex: blockIndex(),
-                    content: currentBlock.text,
+                    contentIndex: slot.contentIndex,
+                    content: slot.block.text,
                     partial: output,
                 });
-                currentBlock = null;
+                outputSlots.delete(event.output_index);
             }
-            else if (item.type === "function_call") {
-                const args = currentBlock?.type === "toolCall" && currentBlock.partialJson
-                    ? parseStreamingJson(currentBlock.partialJson)
-                    : parseStreamingJson(item.arguments || "{}");
-                let toolCall;
-                if (currentBlock?.type === "toolCall") {
-                    // Finalize in-place and strip the scratch buffer so replay only
-                    // carries parsed arguments.
-                    currentBlock.arguments = args;
-                    delete currentBlock.partialJson;
-                    toolCall = currentBlock;
-                }
-                else {
-                    toolCall = {
-                        type: "toolCall",
-                        id: `${item.call_id}|${item.id}`,
-                        name: item.name,
-                        arguments: args,
-                    };
-                }
-                currentBlock = null;
-                stream.push({ type: "toolcall_end", contentIndex: blockIndex(), toolCall, partial: output });
+            else if (item.type === "function_call" && slot?.type === "toolCall") {
+                slot.block.arguments = parseStreamingJson(item.arguments || slot.block.partialJson || "{}");
+                // Finalize in-place and strip the scratch buffer so replay only
+                // carries parsed arguments.
+                delete slot.block.partialJson;
+                stream.push({
+                    type: "toolcall_end",
+                    contentIndex: slot.contentIndex,
+                    toolCall: slot.block,
+                    partial: output,
+                });
+                outputSlots.delete(event.output_index);
             }
         }
-        else if (event.type === "response.completed") {
-            const response = event.response;
-            if (response?.id) {
-                output.responseId = response.id;
-            }
-            if (response?.usage) {
-                const cachedTokens = response.usage.input_tokens_details?.cached_tokens || 0;
-                output.usage = {
-                    // OpenAI includes cached tokens in input_tokens, so subtract to get non-cached input
-                    input: (response.usage.input_tokens || 0) - cachedTokens,
-                    output: response.usage.output_tokens || 0,
-                    cacheRead: cachedTokens,
-                    cacheWrite: 0,
-                    totalTokens: response.usage.total_tokens || 0,
-                    cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
-                };
-            }
-            calculateCost(model, output.usage);
-            if (options?.applyServiceTierPricing) {
-                const serviceTier = options.resolveServiceTier
-                    ? options.resolveServiceTier(response?.service_tier, options.serviceTier)
-                    : (response?.service_tier ?? options.serviceTier);
-                options.applyServiceTierPricing(output.usage, serviceTier);
-            }
-            // Map status to stop reason
-            output.stopReason = mapStopReason(response?.status);
-            if (output.content.some((b) => b.type === "toolCall") && output.stopReason === "stop") {
-                output.stopReason = "toolUse";
-            }
+        else if (event.type === "response.completed" || event.type === "response.incomplete") {
+            finalizeResponse(event.response);
         }
         else if (event.type === "error") {
             throw new Error(`Error Code ${event.code}: ${event.message}` || "Unknown error");
         }
         else if (event.type === "response.failed") {
+            sawTerminalResponseEvent = true;
             const error = event.response?.error;
             const details = event.response?.incomplete_details;
             const msg = error
@@ -473,6 +456,9 @@ export async function processResponsesStream(openaiStream, output, stream, model
             throw new Error(msg);
         }
     }
+    if (!sawTerminalResponseEvent) {
+        throw new Error("OpenAI Responses stream ended before a terminal response event");
+    }
 }
 function mapStopReason(status) {
     if (!status)