@usagetap/sdk 0.8.1 → 0.10.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +96 -26
- package/dist/adapters/openai.cjs +2 -1
- package/dist/adapters/openai.cjs.map +1 -1
- package/dist/adapters/openai.d.cts +1 -1
- package/dist/adapters/openai.d.ts +1 -1
- package/dist/adapters/openai.mjs +2 -1
- package/dist/adapters/openai.mjs.map +1 -1
- package/dist/adapters/openrouter.cjs +2 -1
- package/dist/adapters/openrouter.cjs.map +1 -1
- package/dist/adapters/openrouter.d.cts +1 -1
- package/dist/adapters/openrouter.d.ts +1 -1
- package/dist/adapters/openrouter.mjs +2 -1
- package/dist/adapters/openrouter.mjs.map +1 -1
- package/dist/{client-DokYK2Gv.d.cts → client-DEbk0Q2l.d.cts} +80 -5
- package/dist/{client-DokYK2Gv.d.ts → client-DEbk0Q2l.d.ts} +80 -5
- package/dist/express/index.cjs +2 -1
- package/dist/express/index.cjs.map +1 -1
- package/dist/express/index.d.cts +1 -1
- package/dist/express/index.d.ts +1 -1
- package/dist/express/index.mjs +2 -1
- package/dist/express/index.mjs.map +1 -1
- package/dist/index.cjs +20 -12
- package/dist/index.cjs.map +1 -1
- package/dist/index.d.cts +2 -2
- package/dist/index.d.ts +2 -2
- package/dist/index.mjs +20 -12
- package/dist/index.mjs.map +1 -1
- package/dist/openai/index.cjs +2 -1
- package/dist/openai/index.cjs.map +1 -1
- package/dist/openai/index.d.cts +1 -1
- package/dist/openai/index.d.ts +1 -1
- package/dist/openai/index.mjs +2 -1
- package/dist/openai/index.mjs.map +1 -1
- package/package.json +1 -1
|
@@ -76,11 +76,12 @@ function inferUsageFromResponse(response, hints) {
|
|
|
76
76
|
if (!candidate.usage) {
|
|
77
77
|
return void 0;
|
|
78
78
|
}
|
|
79
|
+
const cachedInputTokens = candidate.usage.prompt_tokens_details?.cached_tokens ?? candidate.usage.cached_tokens;
|
|
79
80
|
return {
|
|
80
81
|
modelUsed: candidate.model ?? hints?.preferredModel,
|
|
81
82
|
inputTokens: candidate.usage.prompt_tokens,
|
|
82
83
|
responseTokens: candidate.usage.completion_tokens,
|
|
83
|
-
|
|
84
|
+
cachedInputTokens
|
|
84
85
|
};
|
|
85
86
|
}
|
|
86
87
|
function wrapStreamForUsageTap(source, finalize, ctx) {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["../../src/adapters/openai.ts","../../src/adapters/openrouter.ts"],"names":[],"mappings":";AA2IO,SAAS,oBAAoB,IAAA,EAAwC;AAC1E,EAAA,MAAM,EAAE,MAAA,EAAQ,QAAA,EAAS,GAAI,IAAA;AAE7B,EAAA,OAAO;AAAA,IACL,MAAM,OAAkB,MAAA,EAA+E;AACrG,MAAA,MAAM,MAAA,GAAS,MAAM,QAAA,CAAS,SAAA;AAAA,QAC5B,MAAA,CAAO,KAAA;AAAA,QACP,OAAO,GAAA,KAAQ;AACb,UAAA,MAAM,QAAA,GAAW,MAAM,MAAA,CAAO,IAAA,CAAK,MAAA,EAAQ;AAAA,YACzC,KAAA,EAAO,GAAA,CAAI,KAAA,CAAM,IAAA,CAAK,WAAA;AAAA,YACtB,OAAO,GAAA,CAAI;AAAA,WACZ,CAAA;AAED,UAAA,aAAA,CAAc,UAAU,GAAA,CAAI,KAAA,CAAM,KAAK,WAAA,EAAa,MAAA,CAAO,cAAc,GAAG,CAAA;AAE5E,UAAA,OAAO;AAAA,YACL,IAAA,EAAM,QAAA;AAAA,YACN,OAAO,GAAA,CAAI;AAAA,WACb;AAAA,QACF,CAAA;AAAA,QACA,MAAA,CAAO;AAAA,OACT;AAEA,MAAA,OAAO,MAAA;AAAA,IACT,CAAA;AAAA,IAEA,MAAM,aAAsB,MAAA,EAA2E;AACrG,MAAA,MAAM,MAAA,GAAS,MAAM,QAAA,CAAS,SAAA;AAAA,QAC5B,MAAA,CAAO,KAAA;AAAA,QACP,OAAO,GAAA,KAAQ;AACb,UAAA,MAAM,EAAE,MAAA,EAAQ,UAAA,KAAe,MAAM,MAAA,CAAO,KAAK,MAAA,EAAQ;AAAA,YACvD,KAAA,EAAO,GAAA,CAAI,KAAA,CAAM,IAAA,CAAK,WAAA;AAAA,YACtB,OAAO,GAAA,CAAI;AAAA,WACZ,CAAA;AAED,UAAA,MAAM,OAAA,GAAU,qBAAA,CAAsB,MAAA,EAAQ,YAAY;AACxD,YAAA,IAAI,CAAC,UAAA,EAAY;AACjB,YAAA,IAAI;AACF,cAAA,MAAM,UAAA,GAAa,MAAM,UAAA,EAAW;AACpC,cAAA,IAAI,UAAA,EAAY;AACd,gBAAA,GAAA,CAAI,SAAS,UAAU,CAAA;AAAA,cACzB;AAAA,YACF,SAAS,KAAA,EAAO;AACd,cAAA,GAAA,CAAI,QAAA,CAAS;AAAA,gBACX,IAAA,EAAM,sBAAA;AAAA,gBACN,SAAS,KAAA,YAAiB,KAAA,GAAQ,KAAA,CAAM,OAAA,GAAU,OAAO,KAAK;AAAA,eAC/D,CAAA;AACD,cAAA,MAAM,KAAA;AAAA,YACR;AAAA,UACF,GAAG,GAAG,CAAA;AAEN,UAAA,MAAM,WAAW,YAA2B;AAC1C,YAAA,MAAM,QAAQ,kBAAA,IAAqB;AAAA,UACrC,CAAA;AAEA,UAAA,OAAO;AAAA,YACL,MAAA,EAAQ,OAAA;AAAA,YACR,OAAO,GAAA,CAAI,KAAA;AAAA,YACX;AAAA,WACF;AAAA,QACF,CAAA;AAAA,QACA,MAAA,CAAO;AAAA,OACT;AAEA,MAAA,OAAO,MAAA;AAAA,IACT;AAAA,GACF;AACF;AAuhBA,SAAS,eAAe,KAAA,EAAuD;AAC7E,EAAA,OAAO,OAAO,KAAA,KAAU,QAAA,IAAY,KAAA,KAAU,IAAA;AAChD;AAgRA,SAAS,aAAA,CACP,QAAA,EACA,KAAA,EACA,SAAA,EACA,GAAA,EACM;AACN,EAAA,MAAM,QAAA,GAAW,YAAY,QAAQ,CAAA;AACrC,EAAA,MAAM,QAAA,GAAW,QAAA,IAAY,sBAAA,CAAuB,QAAA,EAAU,KAAK,CAAA;AAEnE,EAAA,IAAI,QAAA,EAAU;AACZ,IAAA,GAAA,CAAI,SAAS,QAAQ,CAAA;AAAA,EACvB;AACF;AAEA,SAAS,sBAAA,CACP,UACA,KAAA,EAC+D;AAC/D,EAAA,IAAI,CAAC,QAAA,IAAY,OAAO,QAAA,KAAa,QAAA,EAAU;AAC7C,IAAA,OAAO,MAAA;AAAA,EACT;AAEA,EAAA,MAAM,SAAA,GAAY,QAAA;AAUlB,EAAA,IAAI,CAAC,UAAU,KAAA,EAAO;AACpB,IAAA,OAAO,MAAA;AAAA,EACT;AAEA,EAAA,OAAO;AAAA,IACL,SAAA,EAAW,SAAA,CAAU,KAAA,IAAS,KAAA,EAAO,cAAA;AAAA,IACrC,WAAA,EAAa,UAAU,KAAA,CAAM,aAAA;AAAA,IAC7B,cAAA,EAAgB,UAAU,KAAA,CAAM,iBAAA;AAAA,IAChC,YAAA,EAAc,UAAU,KAAA,CAAM;AAAA,GAChC;AACF;AAEA,SAAS,qBAAA,CACP,MAAA,EACA,QAAA,EACA,GAAA,EACsE;AACtE,EAAA,MAAM,WAAA,GAAc,MAAA,CAAO,MAAA,CAAO,aAAa,CAAA;AAC/C,EAAA,IAAI,OAAO,gBAAgB,UAAA,EAAY;AACrC,IAAA,MAAM,IAAI,UAAU,8BAA8B,CAAA;AAAA,EACpD;AAEA,EAAA,MAAM,QAAA,GAAW,WAAA,CAAY,IAAA,CAAK,MAAM,CAAA;AACxC,EAAA,IAAI,SAAA,GAAY,KAAA;AAEhB,EAAA,MAAM,iBAAiB,YAA2B;AAChD,IAAA,IAAI,SAAA,EAAW;AACf,IAAA,SAAA,GAAY,IAAA;AACZ,IAAA,IAAI;AACF,MAAA,MAAM,QAAA,EAAS;AAAA,IACjB,SAAS,KAAA,EAAO;AACd,MAAA,GAAA,CAAI,QAAA,CAAS;AAAA,QACX,IAAA,EAAM,sBAAA;AAAA,QACN,SAAS,KAAA,YAAiB,KAAA,GAAQ,KAAA,CAAM,OAAA,GAAU,OAAO,KAAK;AAAA,OAC/D,CAAA;AACD,MAAA,MAAM,KAAA;AAAA,IACR;AAAA,EACF,CAAA;AAEA,EAAA,MAAM,SAAA,GAAa,MAAA,CAAO,cAAA,CAAe,MAAgB,KAAuB,MAAA,CAAO,SAAA;AACvF,EAAA,MAAM,OAAA,GAAU,MAAA,CAAO,MAAA,CAAO,SAAS,CAAA;AAEvC,EAAA,KAAA,MAAW,GAAA,IAAO,OAAA,CAAQ,OAAA,CAAQ,MAAgB,CAAA,EAAG;AACnD,IAAA,IAAI;AACF,MAAA,MAAM,UAAA,GAAa,MAAA,CAAO,wBAAA,CAAyB,MAAA,EAAkB,GAAG,CAAA;AACxE,MAAA,IAAI,UAAA,EAAY;AACd,QAAA,MAAA,CAAO,cAAA,CAAe,OAAA,EAAS,GAAA,EAAK,UAAU,CAAA;AAAA,MAChD;AAAA,IACF,CAAA,CAAA,MAAQ;AAAA,IAER;AAAA,EACF;AAEA,EAAA,MAAA,CAAO,cAAA,CAAe,OAAA,EAAS,MAAA,CAAO,aAAA,EAAe;AAAA,IACnD,KAAA,GAA2C;AACzC,MAAA,OAAO,IAAA;AAAA,IACT,CAAA;AAAA,IACA,YAAA,EAAc;AAAA,GACf,CAAA;AAED,EAAA,MAAA,CAAO,cAAA,CAAe,SAAS,MAAA,EAAQ;AAAA,IACrC,KAAA,EAAO,UAAU,IAAA,KAAuF;AACtG,MAAA,IAAI;AACF,QAAA,MAAM,MAAA,GAAS,MAAM,QAAA,CAAS,IAAA,CAAK,GAAG,IAAI,CAAA;AAC1C,QAAA,IAAI,OAAO,IAAA,EAAM;AACf,UAAA,MAAM,cAAA,EAAe;AAAA,QACvB;AACA,QAAA,OAAO,MAAA;AAAA,MACT,SAAS,KAAA,EAAO;AACd,QAAA,MAAM,cAAA,EAAe,CAAE,KAAA,CAAM,MAAM,MAAS,CAAA;AAC5C,QAAA,MAAM,KAAA;AAAA,MACR;AAAA,IACF,CAAA;AAAA,IACA,YAAA,EAAc,IAAA;AAAA,IACd,QAAA,EAAU;AAAA,GACX,CAAA;AAED,EAAA,MAAA,CAAO,cAAA,CAAe,SAAS,QAAA,EAAU;AAAA,IACvC,KAAA,EAAO,OAAO,KAAA,KAAsD;AAClE,MAAA,IAAI,OAAO,QAAA,CAAS,MAAA,KAAW,UAAA,EAAY;AACzC,QAAA,MAAM,SAAA,GAAqB,MAAM,QAAA,CAAS,MAAA,CAAO,KAAK,CAAA;AACtD,QAAA,IAAI,CAAC,gBAAA,CAA0B,SAAS,CAAA,EAAG;AACzC,UAAA,MAAM,IAAI,UAAU,8CAA8C,CAAA;AAAA,QACpE;AACA,QAAA,MAAM,cAAA,EAAe;AACrB,QAAA,OAAO,SAAA;AAAA,MACT;AACA,MAAA,MAAM,cAAA,EAAe;AACrB,MAAA,OAAO,EAAE,IAAA,EAAM,IAAA,EAAM,KAAA,EAAM;AAAA,IAC7B,CAAA;AAAA,IACA,YAAA,EAAc,IAAA;AAAA,IACd,QAAA,EAAU;AAAA,GACX,CAAA;AAED,EAAA,MAAA,CAAO,cAAA,CAAe,SAAS,OAAA,EAAS;AAAA,IACtC,KAAA,EAAO,OAAO,KAAA,KAAsD;AAClE,MAAA,IAAI,OAAO,QAAA,CAAS,KAAA,KAAU,UAAA,EAAY;AACxC,QAAA,MAAM,SAAA,GAAqB,MAAM,QAAA,CAAS,KAAA,CAAM,KAAK,CAAA;AACrD,QAAA,IAAI,CAAC,gBAAA,CAA0B,SAAS,CAAA,EAAG;AACzC,UAAA,MAAM,IAAI,UAAU,6CAA6C,CAAA;AAAA,QACnE;AACA,QAAA,MAAM,cAAA,EAAe;AACrB,QAAA,OAAO,SAAA;AAAA,MACT;AACA,MAAA,MAAM,cAAA,EAAe;AACrB,MAAA,MAAM,KAAA;AAAA,IACR,CAAA;AAAA,IACA,YAAA,EAAc,IAAA;AAAA,IACd,QAAA,EAAU;AAAA,GACX,CAAA;AAED,EAAA,MAAA,CAAO,cAAA,CAAe,SAAS,oBAAA,EAAsB;AAAA,IACnD,OAAO,YAA2B;AAChC,MAAA,MAAM,cAAA,EAAe;AAAA,IACvB,CAAA;AAAA,IACA,YAAA,EAAc;AAAA,GACf,CAAA;AAED,EAAA,OAAO,OAAA;AACT;AAEA,SAAS,iBAAoB,KAAA,EAA4C;AACvE,EAAA,OAAO,cAAA,CAAe,KAAK,CAAA,IAAK,MAAA,IAAU,KAAA;AAC5C;;;ACxoCO,SAAS,wBAAwB,IAAA,EAA4C;AAClF,EAAA,OAAO,oBAAoB,IAAI,CAAA;AACjC","file":"openrouter.mjs","sourcesContent":["import type OpenAI from \"openai\";\r\nimport { UsageTapClient } from \"../client\";\r\nimport { UsageTapError } from \"../errors\";\r\nimport type {\r\n BeginCallRequest,\r\n BeginCallResponseBody,\r\n EndCallRequest,\r\n UsageTapSuccessResponse,\r\n VendorHints,\r\n WithUsageContext,\r\n WithUsageOptions,\r\n} from \"../types\";\r\n\r\nexport interface OpenAIAdapterInit {\r\n client: OpenAI;\r\n usageTap: UsageTapClient;\r\n}\r\n\r\nexport interface OpenAIRequestContext {\r\n hints?: VendorHints;\r\n begin: UsageTapSuccessResponse<BeginCallResponseBody>;\r\n}\r\n\r\nexport interface OpenAIInvokeParams<TResponse> {\r\n begin: BeginCallRequest;\r\n call: (client: OpenAI, ctx: OpenAIRequestContext) => Promise<TResponse>;\r\n extractUsage?: (response: TResponse) => Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | void;\r\n withUsageOptions?: WithUsageOptions;\r\n}\r\n\r\nexport interface OpenAIInvokeResult<TResponse> {\r\n data: TResponse;\r\n begin: UsageTapSuccessResponse<BeginCallResponseBody>;\r\n}\r\n\r\nexport interface OpenAIStreamCallResult<TStream> {\r\n stream: AsyncIterable<TStream>;\r\n onComplete?: () => Promise<Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | void> | Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | void;\r\n}\r\n\r\nexport interface OpenAIStreamParams<TStream> {\r\n begin: BeginCallRequest;\r\n call: (client: OpenAI, ctx: OpenAIRequestContext) => Promise<OpenAIStreamCallResult<TStream>>;\r\n withUsageOptions?: WithUsageOptions;\r\n}\r\n\r\nexport interface OpenAIStreamResult<TStream> {\r\n stream: AsyncIterable<TStream> & { __usageTapFinalize?: () => Promise<void> };\r\n begin: UsageTapSuccessResponse<BeginCallResponseBody>;\r\n finalize: () => Promise<void>;\r\n}\r\n\r\nexport interface OpenAIAdapter {\r\n invoke<TResponse>(params: OpenAIInvokeParams<TResponse>): Promise<OpenAIInvokeResult<TResponse>>;\r\n invokeStream<TStream>(params: OpenAIStreamParams<TStream>): Promise<OpenAIStreamResult<TStream>>;\r\n}\r\n\r\ntype ReplaceProperty<T, K extends keyof T, V> = Omit<T, K> & Record<K, V>;\r\n\r\nexport type WrapOpenAIContext = BeginCallRequest;\r\n\r\nexport interface WrapOpenAIOptions {\r\n defaultContext?: Partial<WrapOpenAIContext>;\r\n applyVendorHints?: boolean;\r\n}\r\n\r\ntype ChatCompletionsResource = OpenAI[\"chat\"][\"completions\"];\r\ntype ChatCompletionCreate = ChatCompletionsResource[\"create\"];\r\ntype ChatCompletionCreateParams = Parameters<ChatCompletionCreate>[0];\r\ntype ChatCompletionCreateOptions = Parameters<ChatCompletionCreate>[1];\r\ntype ChatCompletionCreateReturn = ReturnType<ChatCompletionCreate>;\r\n\r\nexport type WrapOpenAICallOptions = (ChatCompletionCreateOptions extends undefined\r\n ? { usageTap?: Partial<WrapOpenAIContext>; withUsage?: WithUsageOptions }\r\n : ChatCompletionCreateOptions & { usageTap?: Partial<WrapOpenAIContext>; withUsage?: WithUsageOptions });\r\n\r\ninterface WrappedChatCompletions extends Omit<ChatCompletionsResource, \"create\"> {\r\n create: (\r\n params: ChatCompletionCreateParams,\r\n options?: WrapOpenAICallOptions,\r\n ) => ChatCompletionCreateReturn;\r\n}\r\n\r\ntype ResponsesResource = OpenAI extends { responses: infer R } ? R : never;\r\ntype ResponsesCreate = ResponsesResource extends { create: infer T } ? T : never;\r\ntype ResponsesCreateParams = ResponsesCreate extends (...args: infer P) => unknown ? P[0] : never;\r\ntype ResponsesCreateOptions = ResponsesCreate extends (...args: infer P) => unknown ? P[1] : never;\r\ntype ResponsesCreateReturn = ResponsesCreate extends (...args: unknown[]) => infer R ? R : never;\r\n\r\nexport type WrapOpenAIResponseCallOptions = (ResponsesCreateOptions extends undefined\r\n ? { usageTap?: Partial<WrapOpenAIContext>; withUsage?: WithUsageOptions }\r\n : ResponsesCreateOptions & { usageTap?: Partial<WrapOpenAIContext>; withUsage?: WithUsageOptions });\r\n\r\ntype WrappedResponses = ResponsesResource extends undefined\r\n ? undefined\r\n : Omit<NonNullable<ResponsesResource>, \"create\"> & {\r\n create: (\r\n params: ResponsesCreateParams,\r\n options?: WrapOpenAIResponseCallOptions,\r\n ) => ResponsesCreateReturn;\r\n };\r\n\r\nexport type WrappedOpenAI = OpenAI & {\r\n chat: ReplaceProperty<OpenAI[\"chat\"], \"completions\", WrappedChatCompletions>;\r\n} & (ResponsesResource extends undefined\r\n ? { responses?: undefined }\r\n : { responses: WrappedResponses }) & {\r\n toNextResponse: typeof toNextResponse;\r\n pipeToResponse: typeof pipeToResponse;\r\n unwrap: () => OpenAI;\r\n };\r\n\r\nexport interface StreamOpenAIRouteOptions {\r\n getRequest: (req: Request) => Promise<{\r\n params: ChatCompletionCreateParams;\r\n usageTap?: Partial<WrapOpenAIContext>;\r\n withUsage?: WithUsageOptions;\r\n }>;\r\n wrapOptions?: WrapOpenAIOptions;\r\n defaultContext?: Partial<WrapOpenAIContext>;\r\n stream?: {\r\n mode?: StreamMode;\r\n headers?: Record<string, string>;\r\n responseInit?: ResponseInit;\r\n };\r\n}\r\n\r\nexport type StreamMode = \"text\" | \"sse\";\r\n\r\nexport interface StreamToResponseOptions {\r\n mode?: StreamMode;\r\n headers?: Record<string, string>;\r\n contentType?: string;\r\n sse?: {\r\n event?: string;\r\n retry?: number;\r\n };\r\n}\r\n\r\nexport function createOpenAIAdapter(init: OpenAIAdapterInit): OpenAIAdapter {\r\n const { client, usageTap } = init;\r\n\r\n return {\r\n async invoke<TResponse>(params: OpenAIInvokeParams<TResponse>): Promise<OpenAIInvokeResult<TResponse>> {\r\n const result = await usageTap.withUsage<OpenAIInvokeResult<TResponse>>(\r\n params.begin,\r\n async (ctx) => {\r\n const response = await params.call(client, {\r\n hints: ctx.begin.data.vendorHints,\r\n begin: ctx.begin,\r\n });\r\n\r\n tryInferUsage(response, ctx.begin.data.vendorHints, params.extractUsage, ctx);\r\n\r\n return {\r\n data: response,\r\n begin: ctx.begin,\r\n } satisfies OpenAIInvokeResult<TResponse>;\r\n },\r\n params.withUsageOptions,\r\n );\r\n\r\n return result;\r\n },\r\n\r\n async invokeStream<TStream>(params: OpenAIStreamParams<TStream>): Promise<OpenAIStreamResult<TStream>> {\r\n const result = await usageTap.withUsage<OpenAIStreamResult<TStream>>(\r\n params.begin,\r\n async (ctx) => {\r\n const { stream, onComplete } = await params.call(client, {\r\n hints: ctx.begin.data.vendorHints,\r\n begin: ctx.begin,\r\n });\r\n\r\n const wrapped = wrapStreamForUsageTap(stream, async () => {\r\n if (!onComplete) return;\r\n try {\r\n const maybeUsage = await onComplete();\r\n if (maybeUsage) {\r\n ctx.setUsage(maybeUsage);\r\n }\r\n } catch (error) {\r\n ctx.setError({\r\n code: \"USAGE_FINALIZE_ERROR\",\r\n message: error instanceof Error ? error.message : String(error),\r\n });\r\n throw error;\r\n }\r\n }, ctx);\r\n\r\n const finalize = async (): Promise<void> => {\r\n await wrapped.__usageTapFinalize?.();\r\n };\r\n\r\n return {\r\n stream: wrapped,\r\n begin: ctx.begin,\r\n finalize,\r\n } satisfies OpenAIStreamResult<TStream>;\r\n },\r\n params.withUsageOptions,\r\n );\r\n\r\n return result;\r\n },\r\n };\r\n}\r\n\r\nexport type UsageTapStream<T> = AsyncIterable<T> & { __usageTapFinalize?: () => Promise<void> };\r\ntype UsageTapIterableIterator<T> = AsyncIterator<T> & UsageTapStream<T> & {\r\n __usageTapFinalize: () => Promise<void>;\r\n};\r\n\r\nexport function toNextResponse<T>(\r\n stream: UsageTapStream<T>,\r\n options: StreamToResponseOptions = {},\r\n): Response {\r\n const mode = options.mode ?? \"text\";\r\n const headers = new Headers(options.headers ?? {});\r\n\r\n if (mode === \"sse\") {\r\n headers.set(\"content-type\", \"text/event-stream; charset=utf-8\");\r\n headers.set(\"cache-control\", \"no-cache, no-transform\");\r\n headers.set(\"connection\", \"keep-alive\");\r\n headers.set(\"x-accel-buffering\", \"no\");\r\n } else {\r\n headers.set(\"content-type\", options.contentType ?? \"text/plain; charset=utf-8\");\r\n }\r\n\r\n const encoder = new TextEncoder();\r\n let iterator: AsyncIterator<T> | undefined;\r\n\r\n const body = new ReadableStream<Uint8Array>({\r\n async start(controller: ReadableStreamDefaultController<Uint8Array>): Promise<void> {\r\n try {\r\n const getIterator = stream[Symbol.asyncIterator];\r\n if (typeof getIterator !== \"function\") {\r\n controller.close();\r\n return;\r\n }\r\n\r\n iterator = getIterator.call(stream);\r\n\r\n while (true) {\r\n const result = await iterator.next();\r\n if (result.done) {\r\n break;\r\n }\r\n\r\n const text = chunkToText(result.value);\r\n if (!text) {\r\n continue;\r\n }\r\n\r\n if (mode === \"sse\") {\r\n controller.enqueue(encoder.encode(formatSsePayload(text, options.sse)));\r\n } else {\r\n controller.enqueue(encoder.encode(text));\r\n }\r\n }\r\n controller.close();\r\n } catch (error) {\r\n controller.error(error);\r\n } finally {\r\n await stream.__usageTapFinalize?.();\r\n }\r\n },\r\n async cancel(): Promise<void> {\r\n if (!iterator) {\r\n const getIterator = stream[Symbol.asyncIterator];\r\n if (typeof getIterator === \"function\") {\r\n iterator = getIterator.call(stream);\r\n }\r\n }\r\n\r\n if (iterator && typeof iterator.return === \"function\") {\r\n await iterator.return();\r\n }\r\n await stream.__usageTapFinalize?.();\r\n },\r\n });\r\n\r\n return new Response(body, { headers });\r\n}\r\n\r\nexport async function pipeToResponse<T>(\r\n stream: UsageTapStream<T>,\r\n res: NodeResponseLike,\r\n options: StreamToResponseOptions = {},\r\n): Promise<void> {\r\n const mode = options.mode ?? \"text\";\r\n\r\n if (mode === \"sse\") {\r\n setHeaderIfPossible(res, \"Content-Type\", \"text/event-stream; charset=utf-8\");\r\n setHeaderIfPossible(res, \"Cache-Control\", \"no-cache, no-transform\");\r\n setHeaderIfPossible(res, \"Connection\", \"keep-alive\");\r\n setHeaderIfPossible(res, \"X-Accel-Buffering\", \"no\");\r\n } else {\r\n setHeaderIfPossible(res, \"Content-Type\", options.contentType ?? \"text/plain; charset=utf-8\");\r\n }\r\n\r\n const encoder = new TextEncoder();\r\n const iterator = stream[Symbol.asyncIterator]();\r\n\r\n try {\r\n while (true) {\r\n const result = await iterator.next();\r\n if (result.done) {\r\n break;\r\n }\r\n const text = chunkToText(result.value);\r\n if (!text) {\r\n continue;\r\n }\r\n\r\n const payload = mode === \"sse\" ? formatSsePayload(text, options.sse) : text;\r\n res.write(Buffer.from(encoder.encode(payload)));\r\n res.flush?.();\r\n }\r\n } finally {\r\n res.end();\r\n await stream.__usageTapFinalize?.();\r\n }\r\n}\r\n\r\nconst USAGETAP_CORRELATION_HEADER = \"x-usage-correlation-id\";\r\n\r\nexport function wrapOpenAI(\r\n client: OpenAI,\r\n usageTap: UsageTapClient,\r\n options: WrapOpenAIOptions = {},\r\n): WrappedOpenAI {\r\n if (!client) {\r\n throw new UsageTapError(\"USAGETAP_BAD_REQUEST\", \"wrapOpenAI requires an OpenAI client instance\");\r\n }\r\n\r\n const defaultContext = options.defaultContext;\r\n const applyVendorHints = options.applyVendorHints !== false;\r\n\r\n const proxiedChat = client.chat\r\n ? createChatProxy(client.chat, usageTap, defaultContext, applyVendorHints)\r\n : undefined;\r\n\r\n const proxiedResponses = typeof client.responses !== \"undefined\"\r\n ? createResponsesProxy(client.responses, usageTap, defaultContext, applyVendorHints)\r\n : undefined;\r\n\r\n const handler: ProxyHandler<OpenAI> = {\r\n get(target, prop, receiver) {\r\n if (prop === \"chat\" && proxiedChat) {\r\n return proxiedChat;\r\n }\r\n\r\n if (prop === \"responses\" && typeof target.responses !== \"undefined\") {\r\n return proxiedResponses ?? (Reflect.get(target as object, prop, receiver) as unknown);\r\n }\r\n\r\n if (prop === \"toNextResponse\") {\r\n return toNextResponse;\r\n }\r\n\r\n if (prop === \"pipeToResponse\") {\r\n return pipeToResponse;\r\n }\r\n\r\n if (prop === \"unwrap\") {\r\n return () => target;\r\n }\r\n\r\n return Reflect.get(target as object, prop, receiver) as unknown;\r\n },\r\n };\r\n\r\n return new Proxy(client, handler) as WrappedOpenAI;\r\n}\r\n\r\nexport function streamOpenAIRoute(\r\n usageTap: UsageTapClient,\r\n openai: OpenAI,\r\n options: StreamOpenAIRouteOptions,\r\n): (req: Request) => Promise<Response> {\r\n if (!options?.getRequest) {\r\n throw new UsageTapError(\"USAGETAP_BAD_REQUEST\", \"streamOpenAIRoute requires a getRequest function\");\r\n }\r\n\r\n const wrapConfig: WrapOpenAIOptions | undefined = options.wrapOptions || options.defaultContext\r\n ? {\r\n ...(options.wrapOptions ?? {}),\r\n defaultContext: options.defaultContext ?? options.wrapOptions?.defaultContext,\r\n }\r\n : undefined;\r\n\r\n const wrappedClient = wrapConfig\r\n ? wrapOpenAI(openai, usageTap, wrapConfig)\r\n : wrapOpenAI(openai, usageTap);\r\n\r\n return async (req: Request): Promise<Response> => {\r\n const requestConfig = await options.getRequest(req);\r\n const mergedParams: ChatCompletionCreateParams = {\r\n ...requestConfig.params,\r\n stream: true,\r\n };\r\n\r\n const callOptions: Partial<WrapOpenAICallOptions> = {};\r\n if (requestConfig.usageTap) {\r\n callOptions.usageTap = requestConfig.usageTap;\r\n }\r\n if (requestConfig.withUsage) {\r\n callOptions.withUsage = requestConfig.withUsage;\r\n }\r\n\r\n const stream = await wrappedClient.chat.completions.create(\r\n mergedParams,\r\n Object.keys(callOptions).length ? (callOptions as WrapOpenAICallOptions) : undefined,\r\n );\r\n\r\n const baseResponse = toNextResponse(stream as UsageTapStream<unknown>, {\r\n mode: options.stream?.mode ?? \"sse\",\r\n headers: options.stream?.headers,\r\n });\r\n\r\n const init = options.stream?.responseInit;\r\n if (!init) {\r\n return baseResponse;\r\n }\r\n\r\n const mergedHeaders = new Headers(baseResponse.headers);\r\n if (init.headers) {\r\n const extra = normalizeHeaders(init.headers);\r\n for (const [key, value] of Object.entries(extra)) {\r\n mergedHeaders.set(key, value);\r\n }\r\n }\r\n\r\n return new Response(baseResponse.body, {\r\n status: init.status ?? baseResponse.status,\r\n statusText: init.statusText ?? baseResponse.statusText,\r\n headers: mergedHeaders,\r\n });\r\n };\r\n}\r\n\r\nexport interface NodeResponseLike {\r\n write(chunk: string | Uint8Array | Buffer): unknown;\r\n end(chunk?: string | Uint8Array | Buffer): unknown;\r\n setHeader?(name: string, value: string): void;\r\n headersSent?: boolean;\r\n statusCode?: number;\r\n status?(code: number): void;\r\n flush?(): void;\r\n}\r\n\r\nfunction createChatProxy(\r\n resource: OpenAI[\"chat\"],\r\n usageTap: UsageTapClient,\r\n defaultContext: Partial<WrapOpenAIContext> | undefined,\r\n applyVendorHints: boolean,\r\n): ReplaceProperty<OpenAI[\"chat\"], \"completions\", WrappedChatCompletions> {\r\n const completions = createChatCompletionsProxy(\r\n resource.completions,\r\n usageTap,\r\n defaultContext,\r\n applyVendorHints,\r\n );\r\n\r\n const handler: ProxyHandler<OpenAI[\"chat\"]> = {\r\n get(target, prop, receiver) {\r\n if (prop === \"completions\") {\r\n return completions;\r\n }\r\n return Reflect.get(target as object, prop, receiver) as unknown;\r\n },\r\n };\r\n\r\n return new Proxy(resource, handler) as ReplaceProperty<OpenAI[\"chat\"], \"completions\", WrappedChatCompletions>;\r\n}\r\n\r\nfunction createResponsesProxy(\r\n resource: ResponsesResource,\r\n usageTap: UsageTapClient,\r\n defaultContext: Partial<WrapOpenAIContext> | undefined,\r\n applyVendorHints: boolean,\r\n): WrappedResponses | undefined {\r\n if (!resource || typeof resource !== \"object\") {\r\n return undefined;\r\n }\r\n\r\n if (!(\"create\" in resource) || typeof (resource as { create?: unknown }).create !== \"function\") {\r\n return resource as unknown as WrappedResponses;\r\n }\r\n\r\n const originalCreate = (resource as { create: (...args: unknown[]) => unknown }).create.bind(resource);\r\n\r\n const wrappedCreate = (\r\n params: ResponsesCreateParams,\r\n options?: WrapOpenAIResponseCallOptions,\r\n ): ResponsesCreateReturn => {\r\n const { requestOptions, usageContext, withUsage } = splitUsageOptions(options);\r\n const beginRequest = resolveBeginRequest(defaultContext, usageContext);\r\n const wantsStream = isStreamingRequest(params);\r\n\r\n return usageTap.withUsage(beginRequest, (ctx) => {\r\n const finalParams = applyVendorHints\r\n ? applyResponsesVendorHints(params, ctx.begin.data.vendorHints)\r\n : params;\r\n const request = attachCorrelationHeader(requestOptions, ctx.begin.correlationId) as ResponsesCreateOptions;\r\n\r\n if (wantsStream) {\r\n const apiPromise = originalCreate(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (rawStream) => {\r\n ensureAsyncIterable(rawStream, \"responses.create\");\r\n const wrappedStream = wrapStreamForUsageTap(rawStream, async () => {\r\n const usage = await extractUsageFromStream(rawStream, ctx.begin.data.vendorHints);\r\n if (usage) {\r\n ctx.setUsage(usage);\r\n }\r\n }, ctx);\r\n return wrappedStream;\r\n });\r\n\r\n return wrappedPromise as unknown as ResponsesCreateReturn;\r\n }\r\n\r\n const apiPromise = originalCreate(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (response) => {\r\n tryInferUsage(response, ctx.begin.data.vendorHints, undefined, ctx);\r\n return response;\r\n });\r\n return wrappedPromise as unknown as ResponsesCreateReturn;\r\n }, withUsage) as ResponsesCreateReturn;\r\n };\r\n\r\n const handler: ProxyHandler<object> = {\r\n get(target, prop, receiver) {\r\n if (prop === \"create\") {\r\n return wrappedCreate;\r\n }\r\n return Reflect.get(target, prop, receiver) as unknown;\r\n },\r\n };\r\n\r\n return new Proxy(resource as object, handler) as WrappedResponses;\r\n}\r\n\r\nfunction createChatCompletionsProxy(\r\n resource: ChatCompletionsResource,\r\n usageTap: UsageTapClient,\r\n defaultContext: Partial<WrapOpenAIContext> | undefined,\r\n applyVendorHints: boolean,\r\n): WrappedChatCompletions {\r\n const originalCreate = resource.create.bind(resource);\r\n const streamCandidate = (resource as { stream?: unknown }).stream;\r\n const originalStream = typeof streamCandidate === \"function\"\r\n ? (streamCandidate as (...args: unknown[]) => unknown).bind(resource)\r\n : undefined;\r\n\r\n const wrappedCreate = (\r\n params: ChatCompletionCreateParams,\r\n options?: WrapOpenAICallOptions,\r\n ): ChatCompletionCreateReturn => {\r\n const { requestOptions, usageContext, withUsage } = splitUsageOptions(options);\r\n const beginRequest = resolveBeginRequest(defaultContext, usageContext);\r\n const wantsStream = isStreamingRequest(params);\r\n\r\n return usageTap.withUsage(beginRequest, (ctx) => {\r\n const finalParams = applyVendorHints\r\n ? applyChatVendorHints(params, ctx.begin.data.vendorHints)\r\n : params;\r\n const request = attachCorrelationHeader(requestOptions, ctx.begin.correlationId) as ChatCompletionCreateOptions;\r\n\r\n if (wantsStream) {\r\n const apiPromise = originalCreate(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (rawStream) => {\r\n ensureAsyncIterable(rawStream, \"chat.completions.create\");\r\n const wrappedStream = wrapStreamForUsageTap(rawStream, async () => {\r\n const usage = await extractUsageFromStream(rawStream, ctx.begin.data.vendorHints);\r\n if (usage) {\r\n ctx.setUsage(usage);\r\n }\r\n }, ctx);\r\n return wrappedStream;\r\n });\r\n\r\n return wrappedPromise as unknown as ChatCompletionCreateReturn;\r\n }\r\n\r\n const apiPromise = originalCreate(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (response) => {\r\n tryInferUsage(response, ctx.begin.data.vendorHints, undefined, ctx);\r\n return response;\r\n });\r\n return wrappedPromise as unknown as ChatCompletionCreateReturn;\r\n }, withUsage) as ChatCompletionCreateReturn;\r\n };\r\n\r\n const wrappedStream = originalStream\r\n ? (\r\n params: ChatCompletionCreateParams,\r\n options?: WrapOpenAICallOptions,\r\n ): ChatCompletionCreateReturn => {\r\n const { requestOptions, usageContext, withUsage } = splitUsageOptions(options);\r\n const beginRequest = resolveBeginRequest(defaultContext, usageContext);\r\n\r\n return usageTap.withUsage(beginRequest, (ctx) => {\r\n const finalParams = applyVendorHints\r\n ? applyChatVendorHints(params, ctx.begin.data.vendorHints)\r\n : params;\r\n const request = attachCorrelationHeader(requestOptions, ctx.begin.correlationId) as ChatCompletionCreateOptions;\r\n\r\n const apiPromise = originalStream(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (rawStream) => {\r\n ensureAsyncIterable(rawStream, \"chat.completions.stream\");\r\n const wrappedStreamInner = wrapStreamForUsageTap(rawStream, async () => {\r\n const usage = await extractUsageFromStream(rawStream, ctx.begin.data.vendorHints);\r\n if (usage) {\r\n ctx.setUsage(usage);\r\n }\r\n }, ctx);\r\n return wrappedStreamInner;\r\n });\r\n\r\n return wrappedPromise as unknown as ChatCompletionCreateReturn;\r\n }, withUsage) as ChatCompletionCreateReturn;\r\n }\r\n : undefined;\r\n\r\n const handler: ProxyHandler<ChatCompletionsResource> = {\r\n get(target, prop, receiver) {\r\n if (prop === \"create\") {\r\n return wrappedCreate;\r\n }\r\n\r\n if (prop === \"stream\" && wrappedStream) {\r\n return wrappedStream;\r\n }\r\n\r\n return Reflect.get(target as object, prop, receiver) as unknown;\r\n },\r\n };\r\n\r\n return new Proxy(resource, handler) as unknown as WrappedChatCompletions;\r\n}\r\n\r\ninterface SplitUsageOptionsResult {\r\n requestOptions?: Record<string, unknown>;\r\n usageContext?: Partial<WrapOpenAIContext>;\r\n withUsage?: WithUsageOptions;\r\n}\r\n\r\nfunction splitUsageOptions(options: unknown): SplitUsageOptionsResult {\r\n if (!options || typeof options !== \"object\") {\r\n return {};\r\n }\r\n\r\n const { usageTap, withUsage, ...rest } = options as {\r\n usageTap?: Partial<WrapOpenAIContext>;\r\n withUsage?: WithUsageOptions;\r\n } & Record<string, unknown>;\r\n\r\n const requestOptions = Object.keys(rest).length ? cloneRequestOptions(rest) : undefined;\r\n\r\n return {\r\n requestOptions,\r\n usageContext: usageTap,\r\n withUsage,\r\n } satisfies SplitUsageOptionsResult;\r\n}\r\n\r\nfunction resolveBeginRequest(\r\n defaults: Partial<WrapOpenAIContext> | undefined,\r\n override: Partial<WrapOpenAIContext> | undefined,\r\n): BeginCallRequest {\r\n const base = defaults ?? {};\r\n const current = override ?? {};\r\n const customerId = current.customerId ?? base.customerId;\r\n\r\n if (!customerId) {\r\n throw new UsageTapError(\r\n \"USAGETAP_BAD_REQUEST\",\r\n \"wrapOpenAI requires usageTap.customerId (provide defaultContext or options.usageTap)\",\r\n );\r\n }\r\n\r\n const tags = mergeTags(base.tags, current.tags);\r\n const begin: BeginCallRequest = { customerId } satisfies BeginCallRequest;\r\n\r\n const requested = current.requested ?? base.requested;\r\n if (requested) begin.requested = requested;\r\n\r\n const feature = current.feature ?? base.feature;\r\n if (feature) begin.feature = feature;\r\n\r\n const idempotency = current.idempotency ?? base.idempotency;\r\n if (idempotency) begin.idempotency = idempotency;\r\n\r\n const customerName = current.customerName ?? base.customerName;\r\n if (customerName) begin.customerName = customerName;\r\n\r\n const customerEmail = current.customerEmail ?? base.customerEmail;\r\n if (customerEmail) begin.customerEmail = customerEmail;\r\n\r\n if (tags?.length) {\r\n begin.tags = tags;\r\n }\r\n\r\n return begin;\r\n}\r\n\r\ntype PromiseLikeOrValue<T> = PromiseLike<T> | T;\r\n\r\nfunction transformApiPromise<TValue, TResult>(\r\n apiPromise: PromiseLikeOrValue<TValue>,\r\n onResolve: (value: TValue) => PromiseLike<TResult> | TResult,\r\n): Promise<TResult> {\r\n const resolvedPromise = Promise.resolve(apiPromise).then(onResolve);\r\n\r\n if (isObjectRecord(apiPromise)) {\r\n const proto = Object.getPrototypeOf(apiPromise) as object | null;\r\n if (proto) {\r\n Object.setPrototypeOf(resolvedPromise, proto);\r\n }\r\n\r\n for (const key of Reflect.ownKeys(apiPromise)) {\r\n if (key === \"then\" || key === \"catch\" || key === \"finally\") {\r\n continue;\r\n }\r\n\r\n try {\r\n const descriptor = Object.getOwnPropertyDescriptor(apiPromise, key);\r\n if (descriptor) {\r\n Reflect.defineProperty(resolvedPromise, key, descriptor);\r\n }\r\n } catch {\r\n /* ignore non-configurable properties */\r\n }\r\n }\r\n }\r\n\r\n return resolvedPromise;\r\n}\r\n\r\nfunction isObjectRecord(value: unknown): value is Record<PropertyKey, unknown> {\r\n return typeof value === \"object\" && value !== null;\r\n}\r\n\r\nfunction cloneRecord(value: unknown): Record<string, unknown> {\r\n return isObjectRecord(value) ? { ...(value as Record<string, unknown>) } : {};\r\n}\r\n\r\nfunction isStringTuple(value: unknown): value is [string, string] {\r\n return Array.isArray(value) && value.length >= 2 && typeof value[0] === \"string\" && typeof value[1] === \"string\";\r\n}\r\n\r\nfunction cloneRequestOptions(source: Record<string, unknown>): Record<string, unknown> {\r\n const clone: Record<string, unknown> = { ...source };\r\n\r\n if (\"headers\" in clone) {\r\n clone.headers = normalizeHeaders((clone as { headers?: unknown }).headers);\r\n }\r\n\r\n return clone;\r\n}\r\n\r\nfunction attachCorrelationHeader(\r\n options: Record<string, unknown> | undefined,\r\n correlationId: string,\r\n): Record<string, unknown> | undefined {\r\n const normalized = normalizeHeaders(options?.headers);\r\n\r\n if (correlationId && !normalized[USAGETAP_CORRELATION_HEADER]) {\r\n normalized[USAGETAP_CORRELATION_HEADER] = correlationId;\r\n }\r\n\r\n if (!options) {\r\n return Object.keys(normalized).length\r\n ? ({ headers: normalized } satisfies Record<string, unknown>)\r\n : undefined;\r\n }\r\n\r\n const next = { ...options } satisfies Record<string, unknown>;\r\n if (Object.keys(normalized).length) {\r\n next.headers = normalized;\r\n }\r\n return next;\r\n}\r\n\r\nfunction normalizeHeaders(headers: unknown): Record<string, string> {\r\n if (!headers) {\r\n return {};\r\n }\r\n\r\n if (headers instanceof Headers) {\r\n const result: Record<string, string> = {};\r\n headers.forEach((value, key) => {\r\n result[key.toLowerCase()] = value;\r\n });\r\n return result;\r\n }\r\n\r\n if (Array.isArray(headers)) {\r\n const result: Record<string, string> = {};\r\n for (const entry of headers) {\r\n if (!isStringTuple(entry)) {\r\n continue;\r\n }\r\n const [key, value] = entry;\r\n result[key.toLowerCase()] = value;\r\n }\r\n return result;\r\n }\r\n\r\n if (isObjectRecord(headers)) {\r\n const result: Record<string, string> = {};\r\n const record = headers as Record<string, unknown>;\r\n for (const key of Object.keys(record)) {\r\n const value = record[key];\r\n if (value !== undefined && value !== null) {\r\n result[key.toLowerCase()] = String(value);\r\n }\r\n }\r\n return result;\r\n }\r\n\r\n return {};\r\n}\r\n\r\nfunction mergeTags(a?: string[], b?: string[]): string[] | undefined {\r\n const values = [...(a ?? []), ...(b ?? [])]\r\n .map((value) => (typeof value === \"string\" ? value.trim() : \"\"))\r\n .filter(Boolean);\r\n\r\n if (!values.length) {\r\n return undefined;\r\n }\r\n\r\n return dedupeStrings(values);\r\n}\r\n\r\nfunction dedupeStrings(values: string[]): string[] {\r\n return Array.from(new Set(values));\r\n}\r\n\r\nfunction isStreamingRequest(params: unknown): boolean {\r\n if (!params || typeof params !== \"object\") {\r\n return false;\r\n }\r\n\r\n const stream = (params as { stream?: unknown }).stream;\r\n if (typeof stream === \"boolean\") {\r\n return stream;\r\n }\r\n\r\n return stream != null;\r\n}\r\n\r\nfunction applyChatVendorHints(\r\n params: ChatCompletionCreateParams,\r\n hints: VendorHints | undefined,\r\n): ChatCompletionCreateParams {\r\n if (!hints) {\r\n return params;\r\n }\r\n\r\n const next = cloneRecord(params);\r\n\r\n if (hints.preferredModel && (next.model === undefined || next.model === null)) {\r\n next.model = hints.preferredModel;\r\n }\r\n\r\n if (typeof hints.maxResponseTokens === \"number\" && next.max_tokens == null) {\r\n next.max_tokens = hints.maxResponseTokens;\r\n }\r\n\r\n if (typeof hints.maxInputTokens === \"number\" && (next as { max_input_tokens?: unknown }).max_input_tokens == null) {\r\n (next as { max_input_tokens?: number }).max_input_tokens = hints.maxInputTokens;\r\n }\r\n\r\n return next as unknown as ChatCompletionCreateParams;\r\n}\r\n\r\nfunction applyResponsesVendorHints(\r\n params: ResponsesCreateParams,\r\n hints: VendorHints | undefined,\r\n): ResponsesCreateParams {\r\n if (!hints) {\r\n return params;\r\n }\r\n\r\n const next = cloneRecord(params);\r\n\r\n if (hints.preferredModel && (next.model === undefined || next.model === null)) {\r\n next.model = hints.preferredModel;\r\n }\r\n\r\n if (typeof hints.maxResponseTokens === \"number\" && (next as { max_output_tokens?: unknown }).max_output_tokens == null) {\r\n (next as { max_output_tokens?: number }).max_output_tokens = hints.maxResponseTokens;\r\n }\r\n\r\n return next as unknown as ResponsesCreateParams;\r\n}\r\n\r\nasync function extractUsageFromStream(\r\n stream: unknown,\r\n hints: VendorHints | undefined,\r\n): Promise<Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | undefined> {\r\n const finalPayload = await resolveStreamFinalPayload(stream);\r\n if (!finalPayload) {\r\n return undefined;\r\n }\r\n\r\n return inferUsageFromResponse(finalPayload, hints);\r\n}\r\n\r\nasync function resolveStreamFinalPayload(stream: unknown): Promise<unknown> {\r\n if (!stream || typeof stream !== \"object\") {\r\n return undefined;\r\n }\r\n\r\n const candidate = stream as {\r\n finalChatCompletion?: () => Promise<unknown>;\r\n finalCompletion?: () => Promise<unknown>;\r\n finalResponse?: () => Promise<unknown>;\r\n finalContent?: () => Promise<unknown>;\r\n };\r\n\r\n if (typeof candidate.finalChatCompletion === \"function\") {\r\n return candidate.finalChatCompletion();\r\n }\r\n\r\n if (typeof candidate.finalResponse === \"function\") {\r\n return candidate.finalResponse();\r\n }\r\n\r\n if (typeof candidate.finalCompletion === \"function\") {\r\n return candidate.finalCompletion();\r\n }\r\n\r\n if (typeof candidate.finalContent === \"function\") {\r\n return candidate.finalContent();\r\n }\r\n\r\n return undefined;\r\n}\r\n\r\nfunction ensureAsyncIterable(value: unknown, label: string): asserts value is AsyncIterable<unknown> {\r\n if (!value || typeof value !== \"object\" || typeof (value as AsyncIterable<unknown>)[Symbol.asyncIterator] !== \"function\") {\r\n throw new UsageTapError(\r\n \"USAGETAP_BAD_REQUEST\",\r\n `${label} expected an async iterable stream but received ${typeof value}`,\r\n );\r\n }\r\n}\r\n\r\nfunction chunkToText(chunk: unknown): string {\r\n if (chunk === undefined || chunk === null) {\r\n return \"\";\r\n }\r\n\r\n if (typeof chunk === \"string\") {\r\n return chunk;\r\n }\r\n\r\n if (typeof chunk === \"object\") {\r\n const candidate = chunk as {\r\n choices?: Array<{\r\n delta?: {\r\n content?: string | Array<{ text?: string }>;\r\n };\r\n }>;\r\n content?: string;\r\n };\r\n\r\n const delta = candidate.choices?.[0]?.delta;\r\n const content = delta?.content ?? candidate.content;\r\n\r\n if (typeof content === \"string\") {\r\n return content;\r\n }\r\n\r\n if (Array.isArray(content)) {\r\n return content\r\n .map((entry) => {\r\n if (!entry) return \"\";\r\n if (typeof entry === \"string\") return entry;\r\n if (typeof entry.text === \"string\") return entry.text;\r\n return \"\";\r\n })\r\n .join(\"\");\r\n }\r\n }\r\n\r\n return String(chunk);\r\n}\r\n\r\nfunction formatSsePayload(\r\n text: string,\r\n options: StreamToResponseOptions[\"sse\"],\r\n): string {\r\n if (!text) {\r\n return \"\";\r\n }\r\n\r\n const lines = text.split(/\\r?\\n/);\r\n const eventLine = options?.event ? `event: ${options.event}\\n` : \"\";\r\n const retryLine = options?.retry ? `retry: ${options.retry}\\n` : \"\";\r\n const dataLines = lines.map((line) => `data: ${line}`).join(\"\\n\");\r\n return `${eventLine}${retryLine}${dataLines}\\n\\n`;\r\n}\r\n\r\nfunction setHeaderIfPossible(res: NodeResponseLike, key: string, value: string): void {\r\n if (typeof res.setHeader === \"function\" && res.headersSent !== true) {\r\n res.setHeader(key, value);\r\n }\r\n}\r\n\r\nfunction tryInferUsage<TResponse>(\r\n response: TResponse,\r\n hints: VendorHints | undefined,\r\n extractor: OpenAIInvokeParams<TResponse>[\"extractUsage\"],\r\n ctx: WithUsageContext,\r\n): void {\r\n const explicit = extractor?.(response);\r\n const inferred = explicit ?? inferUsageFromResponse(response, hints);\r\n\r\n if (inferred) {\r\n ctx.setUsage(inferred);\r\n }\r\n}\r\n\r\nfunction inferUsageFromResponse(\r\n response: unknown,\r\n hints: VendorHints | undefined,\r\n): Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | undefined {\r\n if (!response || typeof response !== \"object\") {\r\n return undefined;\r\n }\r\n\r\n const candidate = response as {\r\n usage?: {\r\n prompt_tokens?: number;\r\n completion_tokens?: number;\r\n total_tokens?: number;\r\n cached_tokens?: number;\r\n };\r\n model?: string;\r\n };\r\n\r\n if (!candidate.usage) {\r\n return undefined;\r\n }\r\n\r\n return {\r\n modelUsed: candidate.model ?? hints?.preferredModel,\r\n inputTokens: candidate.usage.prompt_tokens,\r\n responseTokens: candidate.usage.completion_tokens,\r\n cachedTokens: candidate.usage.cached_tokens,\r\n } satisfies Partial<Omit<EndCallRequest, \"callId\" | \"error\">>;\r\n}\r\n\r\nfunction wrapStreamForUsageTap<TStream>(\r\n source: UsageTapStream<TStream> | AsyncIterable<TStream>,\r\n finalize: () => Promise<void> | void,\r\n ctx: WithUsageContext,\r\n): AsyncIterable<TStream> & { __usageTapFinalize: () => Promise<void> } {\r\n const getIterator = source[Symbol.asyncIterator];\r\n if (typeof getIterator !== \"function\") {\r\n throw new TypeError(\"Stream is not async iterable\");\r\n }\r\n\r\n const iterator = getIterator.call(source) as AsyncIterator<TStream>;\r\n let completed = false;\r\n\r\n const invokeFinalize = async (): Promise<void> => {\r\n if (completed) return;\r\n completed = true;\r\n try {\r\n await finalize();\r\n } catch (error) {\r\n ctx.setError({\r\n code: \"USAGE_FINALIZE_ERROR\",\r\n message: error instanceof Error ? error.message : String(error),\r\n });\r\n throw error;\r\n }\r\n };\r\n\r\n const prototype = (Object.getPrototypeOf(source as object) as object | null) ?? Object.prototype;\r\n const wrapped = Object.create(prototype) as unknown as UsageTapIterableIterator<TStream>;\r\n\r\n for (const key of Reflect.ownKeys(source as object)) {\r\n try {\r\n const descriptor = Object.getOwnPropertyDescriptor(source as object, key);\r\n if (descriptor) {\r\n Object.defineProperty(wrapped, key, descriptor);\r\n }\r\n } catch {\r\n /* ignore non-configurable properties */\r\n }\r\n }\r\n\r\n Object.defineProperty(wrapped, Symbol.asyncIterator, {\r\n value(): UsageTapIterableIterator<TStream> {\r\n return this as UsageTapIterableIterator<TStream>;\r\n },\r\n configurable: true,\r\n });\r\n\r\n Object.defineProperty(wrapped, \"next\", {\r\n value: async (...args: Parameters<AsyncIterator<TStream>[\"next\"]>): Promise<IteratorResult<TStream>> => {\r\n try {\r\n const result = await iterator.next(...args);\r\n if (result.done) {\r\n await invokeFinalize();\r\n }\r\n return result;\r\n } catch (error) {\r\n await invokeFinalize().catch(() => undefined);\r\n throw error;\r\n }\r\n },\r\n configurable: true,\r\n writable: true,\r\n });\r\n\r\n Object.defineProperty(wrapped, \"return\", {\r\n value: async (value?: TStream): Promise<IteratorResult<TStream>> => {\r\n if (typeof iterator.return === \"function\") {\r\n const rawResult: unknown = await iterator.return(value);\r\n if (!isIteratorResult<TStream>(rawResult)) {\r\n throw new TypeError(\"Iterator.return() returned an invalid result\");\r\n }\r\n await invokeFinalize();\r\n return rawResult;\r\n }\r\n await invokeFinalize();\r\n return { done: true, value } as IteratorResult<TStream>;\r\n },\r\n configurable: true,\r\n writable: true,\r\n });\r\n\r\n Object.defineProperty(wrapped, \"throw\", {\r\n value: async (error?: unknown): Promise<IteratorResult<TStream>> => {\r\n if (typeof iterator.throw === \"function\") {\r\n const rawResult: unknown = await iterator.throw(error);\r\n if (!isIteratorResult<TStream>(rawResult)) {\r\n throw new TypeError(\"Iterator.throw() returned an invalid result\");\r\n }\r\n await invokeFinalize();\r\n return rawResult;\r\n }\r\n await invokeFinalize();\r\n throw error;\r\n },\r\n configurable: true,\r\n writable: true,\r\n });\r\n\r\n Object.defineProperty(wrapped, \"__usageTapFinalize\", {\r\n value: async (): Promise<void> => {\r\n await invokeFinalize();\r\n },\r\n configurable: true,\r\n });\r\n\r\n return wrapped;\r\n}\r\n\r\nfunction isIteratorResult<T>(value: unknown): value is IteratorResult<T> {\r\n return isObjectRecord(value) && \"done\" in value;\r\n}\r\n","import type OpenAI from \"openai\";\r\nimport { UsageTapClient } from \"../client\";\r\nimport type { OpenAIAdapter } from \"./openai\";\r\nimport { createOpenAIAdapter } from \"./openai\";\r\n\r\nexport interface OpenRouterAdapterInit {\r\n client: OpenAI;\r\n usageTap: UsageTapClient;\r\n}\r\n\r\nexport function createOpenRouterAdapter(init: OpenRouterAdapterInit): OpenAIAdapter {\r\n return createOpenAIAdapter(init);\r\n}\r\n"]}
|
|
1
|
+
{"version":3,"sources":["../../src/adapters/openai.ts","../../src/adapters/openrouter.ts"],"names":[],"mappings":";AA2IO,SAAS,oBAAoB,IAAA,EAAwC;AAC1E,EAAA,MAAM,EAAE,MAAA,EAAQ,QAAA,EAAS,GAAI,IAAA;AAE7B,EAAA,OAAO;AAAA,IACL,MAAM,OAAkB,MAAA,EAA+E;AACrG,MAAA,MAAM,MAAA,GAAS,MAAM,QAAA,CAAS,SAAA;AAAA,QAC5B,MAAA,CAAO,KAAA;AAAA,QACP,OAAO,GAAA,KAAQ;AACb,UAAA,MAAM,QAAA,GAAW,MAAM,MAAA,CAAO,IAAA,CAAK,MAAA,EAAQ;AAAA,YACzC,KAAA,EAAO,GAAA,CAAI,KAAA,CAAM,IAAA,CAAK,WAAA;AAAA,YACtB,OAAO,GAAA,CAAI;AAAA,WACZ,CAAA;AAED,UAAA,aAAA,CAAc,UAAU,GAAA,CAAI,KAAA,CAAM,KAAK,WAAA,EAAa,MAAA,CAAO,cAAc,GAAG,CAAA;AAE5E,UAAA,OAAO;AAAA,YACL,IAAA,EAAM,QAAA;AAAA,YACN,OAAO,GAAA,CAAI;AAAA,WACb;AAAA,QACF,CAAA;AAAA,QACA,MAAA,CAAO;AAAA,OACT;AAEA,MAAA,OAAO,MAAA;AAAA,IACT,CAAA;AAAA,IAEA,MAAM,aAAsB,MAAA,EAA2E;AACrG,MAAA,MAAM,MAAA,GAAS,MAAM,QAAA,CAAS,SAAA;AAAA,QAC5B,MAAA,CAAO,KAAA;AAAA,QACP,OAAO,GAAA,KAAQ;AACb,UAAA,MAAM,EAAE,MAAA,EAAQ,UAAA,KAAe,MAAM,MAAA,CAAO,KAAK,MAAA,EAAQ;AAAA,YACvD,KAAA,EAAO,GAAA,CAAI,KAAA,CAAM,IAAA,CAAK,WAAA;AAAA,YACtB,OAAO,GAAA,CAAI;AAAA,WACZ,CAAA;AAED,UAAA,MAAM,OAAA,GAAU,qBAAA,CAAsB,MAAA,EAAQ,YAAY;AACxD,YAAA,IAAI,CAAC,UAAA,EAAY;AACjB,YAAA,IAAI;AACF,cAAA,MAAM,UAAA,GAAa,MAAM,UAAA,EAAW;AACpC,cAAA,IAAI,UAAA,EAAY;AACd,gBAAA,GAAA,CAAI,SAAS,UAAU,CAAA;AAAA,cACzB;AAAA,YACF,SAAS,KAAA,EAAO;AACd,cAAA,GAAA,CAAI,QAAA,CAAS;AAAA,gBACX,IAAA,EAAM,sBAAA;AAAA,gBACN,SAAS,KAAA,YAAiB,KAAA,GAAQ,KAAA,CAAM,OAAA,GAAU,OAAO,KAAK;AAAA,eAC/D,CAAA;AACD,cAAA,MAAM,KAAA;AAAA,YACR;AAAA,UACF,GAAG,GAAG,CAAA;AAEN,UAAA,MAAM,WAAW,YAA2B;AAC1C,YAAA,MAAM,QAAQ,kBAAA,IAAqB;AAAA,UACrC,CAAA;AAEA,UAAA,OAAO;AAAA,YACL,MAAA,EAAQ,OAAA;AAAA,YACR,OAAO,GAAA,CAAI,KAAA;AAAA,YACX;AAAA,WACF;AAAA,QACF,CAAA;AAAA,QACA,MAAA,CAAO;AAAA,OACT;AAEA,MAAA,OAAO,MAAA;AAAA,IACT;AAAA,GACF;AACF;AAuhBA,SAAS,eAAe,KAAA,EAAuD;AAC7E,EAAA,OAAO,OAAO,KAAA,KAAU,QAAA,IAAY,KAAA,KAAU,IAAA;AAChD;AAgRA,SAAS,aAAA,CACP,QAAA,EACA,KAAA,EACA,SAAA,EACA,GAAA,EACM;AACN,EAAA,MAAM,QAAA,GAAW,YAAY,QAAQ,CAAA;AACrC,EAAA,MAAM,QAAA,GAAW,QAAA,IAAY,sBAAA,CAAuB,QAAA,EAAU,KAAK,CAAA;AAEnE,EAAA,IAAI,QAAA,EAAU;AACZ,IAAA,GAAA,CAAI,SAAS,QAAQ,CAAA;AAAA,EACvB;AACF;AAEA,SAAS,sBAAA,CACP,UACA,KAAA,EAC+D;AAC/D,EAAA,IAAI,CAAC,QAAA,IAAY,OAAO,QAAA,KAAa,QAAA,EAAU;AAC7C,IAAA,OAAO,MAAA;AAAA,EACT;AAEA,EAAA,MAAM,SAAA,GAAY,QAAA;AAalB,EAAA,IAAI,CAAC,UAAU,KAAA,EAAO;AACpB,IAAA,OAAO,MAAA;AAAA,EACT;AAEA,EAAA,MAAM,oBACJ,SAAA,CAAU,KAAA,CAAM,qBAAA,EAAuB,aAAA,IACvC,UAAU,KAAA,CAAM,aAAA;AAElB,EAAA,OAAO;AAAA,IACL,SAAA,EAAW,SAAA,CAAU,KAAA,IAAS,KAAA,EAAO,cAAA;AAAA,IACrC,WAAA,EAAa,UAAU,KAAA,CAAM,aAAA;AAAA,IAC7B,cAAA,EAAgB,UAAU,KAAA,CAAM,iBAAA;AAAA,IAChC;AAAA,GACF;AACF;AAEA,SAAS,qBAAA,CACP,MAAA,EACA,QAAA,EACA,GAAA,EACsE;AACtE,EAAA,MAAM,WAAA,GAAc,MAAA,CAAO,MAAA,CAAO,aAAa,CAAA;AAC/C,EAAA,IAAI,OAAO,gBAAgB,UAAA,EAAY;AACrC,IAAA,MAAM,IAAI,UAAU,8BAA8B,CAAA;AAAA,EACpD;AAEA,EAAA,MAAM,QAAA,GAAW,WAAA,CAAY,IAAA,CAAK,MAAM,CAAA;AACxC,EAAA,IAAI,SAAA,GAAY,KAAA;AAEhB,EAAA,MAAM,iBAAiB,YAA2B;AAChD,IAAA,IAAI,SAAA,EAAW;AACf,IAAA,SAAA,GAAY,IAAA;AACZ,IAAA,IAAI;AACF,MAAA,MAAM,QAAA,EAAS;AAAA,IACjB,SAAS,KAAA,EAAO;AACd,MAAA,GAAA,CAAI,QAAA,CAAS;AAAA,QACX,IAAA,EAAM,sBAAA;AAAA,QACN,SAAS,KAAA,YAAiB,KAAA,GAAQ,KAAA,CAAM,OAAA,GAAU,OAAO,KAAK;AAAA,OAC/D,CAAA;AACD,MAAA,MAAM,KAAA;AAAA,IACR;AAAA,EACF,CAAA;AAEA,EAAA,MAAM,SAAA,GAAa,MAAA,CAAO,cAAA,CAAe,MAAgB,KAAuB,MAAA,CAAO,SAAA;AACvF,EAAA,MAAM,OAAA,GAAU,MAAA,CAAO,MAAA,CAAO,SAAS,CAAA;AAEvC,EAAA,KAAA,MAAW,GAAA,IAAO,OAAA,CAAQ,OAAA,CAAQ,MAAgB,CAAA,EAAG;AACnD,IAAA,IAAI;AACF,MAAA,MAAM,UAAA,GAAa,MAAA,CAAO,wBAAA,CAAyB,MAAA,EAAkB,GAAG,CAAA;AACxE,MAAA,IAAI,UAAA,EAAY;AACd,QAAA,MAAA,CAAO,cAAA,CAAe,OAAA,EAAS,GAAA,EAAK,UAAU,CAAA;AAAA,MAChD;AAAA,IACF,CAAA,CAAA,MAAQ;AAAA,IAER;AAAA,EACF;AAEA,EAAA,MAAA,CAAO,cAAA,CAAe,OAAA,EAAS,MAAA,CAAO,aAAA,EAAe;AAAA,IACnD,KAAA,GAA2C;AACzC,MAAA,OAAO,IAAA;AAAA,IACT,CAAA;AAAA,IACA,YAAA,EAAc;AAAA,GACf,CAAA;AAED,EAAA,MAAA,CAAO,cAAA,CAAe,SAAS,MAAA,EAAQ;AAAA,IACrC,KAAA,EAAO,UAAU,IAAA,KAAuF;AACtG,MAAA,IAAI;AACF,QAAA,MAAM,MAAA,GAAS,MAAM,QAAA,CAAS,IAAA,CAAK,GAAG,IAAI,CAAA;AAC1C,QAAA,IAAI,OAAO,IAAA,EAAM;AACf,UAAA,MAAM,cAAA,EAAe;AAAA,QACvB;AACA,QAAA,OAAO,MAAA;AAAA,MACT,SAAS,KAAA,EAAO;AACd,QAAA,MAAM,cAAA,EAAe,CAAE,KAAA,CAAM,MAAM,MAAS,CAAA;AAC5C,QAAA,MAAM,KAAA;AAAA,MACR;AAAA,IACF,CAAA;AAAA,IACA,YAAA,EAAc,IAAA;AAAA,IACd,QAAA,EAAU;AAAA,GACX,CAAA;AAED,EAAA,MAAA,CAAO,cAAA,CAAe,SAAS,QAAA,EAAU;AAAA,IACvC,KAAA,EAAO,OAAO,KAAA,KAAsD;AAClE,MAAA,IAAI,OAAO,QAAA,CAAS,MAAA,KAAW,UAAA,EAAY;AACzC,QAAA,MAAM,SAAA,GAAqB,MAAM,QAAA,CAAS,MAAA,CAAO,KAAK,CAAA;AACtD,QAAA,IAAI,CAAC,gBAAA,CAA0B,SAAS,CAAA,EAAG;AACzC,UAAA,MAAM,IAAI,UAAU,8CAA8C,CAAA;AAAA,QACpE;AACA,QAAA,MAAM,cAAA,EAAe;AACrB,QAAA,OAAO,SAAA;AAAA,MACT;AACA,MAAA,MAAM,cAAA,EAAe;AACrB,MAAA,OAAO,EAAE,IAAA,EAAM,IAAA,EAAM,KAAA,EAAM;AAAA,IAC7B,CAAA;AAAA,IACA,YAAA,EAAc,IAAA;AAAA,IACd,QAAA,EAAU;AAAA,GACX,CAAA;AAED,EAAA,MAAA,CAAO,cAAA,CAAe,SAAS,OAAA,EAAS;AAAA,IACtC,KAAA,EAAO,OAAO,KAAA,KAAsD;AAClE,MAAA,IAAI,OAAO,QAAA,CAAS,KAAA,KAAU,UAAA,EAAY;AACxC,QAAA,MAAM,SAAA,GAAqB,MAAM,QAAA,CAAS,KAAA,CAAM,KAAK,CAAA;AACrD,QAAA,IAAI,CAAC,gBAAA,CAA0B,SAAS,CAAA,EAAG;AACzC,UAAA,MAAM,IAAI,UAAU,6CAA6C,CAAA;AAAA,QACnE;AACA,QAAA,MAAM,cAAA,EAAe;AACrB,QAAA,OAAO,SAAA;AAAA,MACT;AACA,MAAA,MAAM,cAAA,EAAe;AACrB,MAAA,MAAM,KAAA;AAAA,IACR,CAAA;AAAA,IACA,YAAA,EAAc,IAAA;AAAA,IACd,QAAA,EAAU;AAAA,GACX,CAAA;AAED,EAAA,MAAA,CAAO,cAAA,CAAe,SAAS,oBAAA,EAAsB;AAAA,IACnD,OAAO,YAA2B;AAChC,MAAA,MAAM,cAAA,EAAe;AAAA,IACvB,CAAA;AAAA,IACA,YAAA,EAAc;AAAA,GACf,CAAA;AAED,EAAA,OAAO,OAAA;AACT;AAEA,SAAS,iBAAoB,KAAA,EAA4C;AACvE,EAAA,OAAO,cAAA,CAAe,KAAK,CAAA,IAAK,MAAA,IAAU,KAAA;AAC5C;;;AC/oCO,SAAS,wBAAwB,IAAA,EAA4C;AAClF,EAAA,OAAO,oBAAoB,IAAI,CAAA;AACjC","file":"openrouter.mjs","sourcesContent":["import type OpenAI from \"openai\";\r\nimport { UsageTapClient } from \"../client\";\r\nimport { UsageTapError } from \"../errors\";\r\nimport type {\r\n BeginCallRequest,\r\n BeginCallResponseBody,\r\n EndCallRequest,\r\n UsageTapSuccessResponse,\r\n VendorHints,\r\n WithUsageContext,\r\n WithUsageOptions,\r\n} from \"../types\";\r\n\r\nexport interface OpenAIAdapterInit {\r\n client: OpenAI;\r\n usageTap: UsageTapClient;\r\n}\r\n\r\nexport interface OpenAIRequestContext {\r\n hints?: VendorHints;\r\n begin: UsageTapSuccessResponse<BeginCallResponseBody>;\r\n}\r\n\r\nexport interface OpenAIInvokeParams<TResponse> {\r\n begin: BeginCallRequest;\r\n call: (client: OpenAI, ctx: OpenAIRequestContext) => Promise<TResponse>;\r\n extractUsage?: (response: TResponse) => Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | void;\r\n withUsageOptions?: WithUsageOptions;\r\n}\r\n\r\nexport interface OpenAIInvokeResult<TResponse> {\r\n data: TResponse;\r\n begin: UsageTapSuccessResponse<BeginCallResponseBody>;\r\n}\r\n\r\nexport interface OpenAIStreamCallResult<TStream> {\r\n stream: AsyncIterable<TStream>;\r\n onComplete?: () => Promise<Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | void> | Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | void;\r\n}\r\n\r\nexport interface OpenAIStreamParams<TStream> {\r\n begin: BeginCallRequest;\r\n call: (client: OpenAI, ctx: OpenAIRequestContext) => Promise<OpenAIStreamCallResult<TStream>>;\r\n withUsageOptions?: WithUsageOptions;\r\n}\r\n\r\nexport interface OpenAIStreamResult<TStream> {\r\n stream: AsyncIterable<TStream> & { __usageTapFinalize?: () => Promise<void> };\r\n begin: UsageTapSuccessResponse<BeginCallResponseBody>;\r\n finalize: () => Promise<void>;\r\n}\r\n\r\nexport interface OpenAIAdapter {\r\n invoke<TResponse>(params: OpenAIInvokeParams<TResponse>): Promise<OpenAIInvokeResult<TResponse>>;\r\n invokeStream<TStream>(params: OpenAIStreamParams<TStream>): Promise<OpenAIStreamResult<TStream>>;\r\n}\r\n\r\ntype ReplaceProperty<T, K extends keyof T, V> = Omit<T, K> & Record<K, V>;\r\n\r\nexport type WrapOpenAIContext = BeginCallRequest;\r\n\r\nexport interface WrapOpenAIOptions {\r\n defaultContext?: Partial<WrapOpenAIContext>;\r\n applyVendorHints?: boolean;\r\n}\r\n\r\ntype ChatCompletionsResource = OpenAI[\"chat\"][\"completions\"];\r\ntype ChatCompletionCreate = ChatCompletionsResource[\"create\"];\r\ntype ChatCompletionCreateParams = Parameters<ChatCompletionCreate>[0];\r\ntype ChatCompletionCreateOptions = Parameters<ChatCompletionCreate>[1];\r\ntype ChatCompletionCreateReturn = ReturnType<ChatCompletionCreate>;\r\n\r\nexport type WrapOpenAICallOptions = (ChatCompletionCreateOptions extends undefined\r\n ? { usageTap?: Partial<WrapOpenAIContext>; withUsage?: WithUsageOptions }\r\n : ChatCompletionCreateOptions & { usageTap?: Partial<WrapOpenAIContext>; withUsage?: WithUsageOptions });\r\n\r\ninterface WrappedChatCompletions extends Omit<ChatCompletionsResource, \"create\"> {\r\n create: (\r\n params: ChatCompletionCreateParams,\r\n options?: WrapOpenAICallOptions,\r\n ) => ChatCompletionCreateReturn;\r\n}\r\n\r\ntype ResponsesResource = OpenAI extends { responses: infer R } ? R : never;\r\ntype ResponsesCreate = ResponsesResource extends { create: infer T } ? T : never;\r\ntype ResponsesCreateParams = ResponsesCreate extends (...args: infer P) => unknown ? P[0] : never;\r\ntype ResponsesCreateOptions = ResponsesCreate extends (...args: infer P) => unknown ? P[1] : never;\r\ntype ResponsesCreateReturn = ResponsesCreate extends (...args: unknown[]) => infer R ? R : never;\r\n\r\nexport type WrapOpenAIResponseCallOptions = (ResponsesCreateOptions extends undefined\r\n ? { usageTap?: Partial<WrapOpenAIContext>; withUsage?: WithUsageOptions }\r\n : ResponsesCreateOptions & { usageTap?: Partial<WrapOpenAIContext>; withUsage?: WithUsageOptions });\r\n\r\ntype WrappedResponses = ResponsesResource extends undefined\r\n ? undefined\r\n : Omit<NonNullable<ResponsesResource>, \"create\"> & {\r\n create: (\r\n params: ResponsesCreateParams,\r\n options?: WrapOpenAIResponseCallOptions,\r\n ) => ResponsesCreateReturn;\r\n };\r\n\r\nexport type WrappedOpenAI = OpenAI & {\r\n chat: ReplaceProperty<OpenAI[\"chat\"], \"completions\", WrappedChatCompletions>;\r\n} & (ResponsesResource extends undefined\r\n ? { responses?: undefined }\r\n : { responses: WrappedResponses }) & {\r\n toNextResponse: typeof toNextResponse;\r\n pipeToResponse: typeof pipeToResponse;\r\n unwrap: () => OpenAI;\r\n };\r\n\r\nexport interface StreamOpenAIRouteOptions {\r\n getRequest: (req: Request) => Promise<{\r\n params: ChatCompletionCreateParams;\r\n usageTap?: Partial<WrapOpenAIContext>;\r\n withUsage?: WithUsageOptions;\r\n }>;\r\n wrapOptions?: WrapOpenAIOptions;\r\n defaultContext?: Partial<WrapOpenAIContext>;\r\n stream?: {\r\n mode?: StreamMode;\r\n headers?: Record<string, string>;\r\n responseInit?: ResponseInit;\r\n };\r\n}\r\n\r\nexport type StreamMode = \"text\" | \"sse\";\r\n\r\nexport interface StreamToResponseOptions {\r\n mode?: StreamMode;\r\n headers?: Record<string, string>;\r\n contentType?: string;\r\n sse?: {\r\n event?: string;\r\n retry?: number;\r\n };\r\n}\r\n\r\nexport function createOpenAIAdapter(init: OpenAIAdapterInit): OpenAIAdapter {\r\n const { client, usageTap } = init;\r\n\r\n return {\r\n async invoke<TResponse>(params: OpenAIInvokeParams<TResponse>): Promise<OpenAIInvokeResult<TResponse>> {\r\n const result = await usageTap.withUsage<OpenAIInvokeResult<TResponse>>(\r\n params.begin,\r\n async (ctx) => {\r\n const response = await params.call(client, {\r\n hints: ctx.begin.data.vendorHints,\r\n begin: ctx.begin,\r\n });\r\n\r\n tryInferUsage(response, ctx.begin.data.vendorHints, params.extractUsage, ctx);\r\n\r\n return {\r\n data: response,\r\n begin: ctx.begin,\r\n } satisfies OpenAIInvokeResult<TResponse>;\r\n },\r\n params.withUsageOptions,\r\n );\r\n\r\n return result;\r\n },\r\n\r\n async invokeStream<TStream>(params: OpenAIStreamParams<TStream>): Promise<OpenAIStreamResult<TStream>> {\r\n const result = await usageTap.withUsage<OpenAIStreamResult<TStream>>(\r\n params.begin,\r\n async (ctx) => {\r\n const { stream, onComplete } = await params.call(client, {\r\n hints: ctx.begin.data.vendorHints,\r\n begin: ctx.begin,\r\n });\r\n\r\n const wrapped = wrapStreamForUsageTap(stream, async () => {\r\n if (!onComplete) return;\r\n try {\r\n const maybeUsage = await onComplete();\r\n if (maybeUsage) {\r\n ctx.setUsage(maybeUsage);\r\n }\r\n } catch (error) {\r\n ctx.setError({\r\n code: \"USAGE_FINALIZE_ERROR\",\r\n message: error instanceof Error ? error.message : String(error),\r\n });\r\n throw error;\r\n }\r\n }, ctx);\r\n\r\n const finalize = async (): Promise<void> => {\r\n await wrapped.__usageTapFinalize?.();\r\n };\r\n\r\n return {\r\n stream: wrapped,\r\n begin: ctx.begin,\r\n finalize,\r\n } satisfies OpenAIStreamResult<TStream>;\r\n },\r\n params.withUsageOptions,\r\n );\r\n\r\n return result;\r\n },\r\n };\r\n}\r\n\r\nexport type UsageTapStream<T> = AsyncIterable<T> & { __usageTapFinalize?: () => Promise<void> };\r\ntype UsageTapIterableIterator<T> = AsyncIterator<T> & UsageTapStream<T> & {\r\n __usageTapFinalize: () => Promise<void>;\r\n};\r\n\r\nexport function toNextResponse<T>(\r\n stream: UsageTapStream<T>,\r\n options: StreamToResponseOptions = {},\r\n): Response {\r\n const mode = options.mode ?? \"text\";\r\n const headers = new Headers(options.headers ?? {});\r\n\r\n if (mode === \"sse\") {\r\n headers.set(\"content-type\", \"text/event-stream; charset=utf-8\");\r\n headers.set(\"cache-control\", \"no-cache, no-transform\");\r\n headers.set(\"connection\", \"keep-alive\");\r\n headers.set(\"x-accel-buffering\", \"no\");\r\n } else {\r\n headers.set(\"content-type\", options.contentType ?? \"text/plain; charset=utf-8\");\r\n }\r\n\r\n const encoder = new TextEncoder();\r\n let iterator: AsyncIterator<T> | undefined;\r\n\r\n const body = new ReadableStream<Uint8Array>({\r\n async start(controller: ReadableStreamDefaultController<Uint8Array>): Promise<void> {\r\n try {\r\n const getIterator = stream[Symbol.asyncIterator];\r\n if (typeof getIterator !== \"function\") {\r\n controller.close();\r\n return;\r\n }\r\n\r\n iterator = getIterator.call(stream);\r\n\r\n while (true) {\r\n const result = await iterator.next();\r\n if (result.done) {\r\n break;\r\n }\r\n\r\n const text = chunkToText(result.value);\r\n if (!text) {\r\n continue;\r\n }\r\n\r\n if (mode === \"sse\") {\r\n controller.enqueue(encoder.encode(formatSsePayload(text, options.sse)));\r\n } else {\r\n controller.enqueue(encoder.encode(text));\r\n }\r\n }\r\n controller.close();\r\n } catch (error) {\r\n controller.error(error);\r\n } finally {\r\n await stream.__usageTapFinalize?.();\r\n }\r\n },\r\n async cancel(): Promise<void> {\r\n if (!iterator) {\r\n const getIterator = stream[Symbol.asyncIterator];\r\n if (typeof getIterator === \"function\") {\r\n iterator = getIterator.call(stream);\r\n }\r\n }\r\n\r\n if (iterator && typeof iterator.return === \"function\") {\r\n await iterator.return();\r\n }\r\n await stream.__usageTapFinalize?.();\r\n },\r\n });\r\n\r\n return new Response(body, { headers });\r\n}\r\n\r\nexport async function pipeToResponse<T>(\r\n stream: UsageTapStream<T>,\r\n res: NodeResponseLike,\r\n options: StreamToResponseOptions = {},\r\n): Promise<void> {\r\n const mode = options.mode ?? \"text\";\r\n\r\n if (mode === \"sse\") {\r\n setHeaderIfPossible(res, \"Content-Type\", \"text/event-stream; charset=utf-8\");\r\n setHeaderIfPossible(res, \"Cache-Control\", \"no-cache, no-transform\");\r\n setHeaderIfPossible(res, \"Connection\", \"keep-alive\");\r\n setHeaderIfPossible(res, \"X-Accel-Buffering\", \"no\");\r\n } else {\r\n setHeaderIfPossible(res, \"Content-Type\", options.contentType ?? \"text/plain; charset=utf-8\");\r\n }\r\n\r\n const encoder = new TextEncoder();\r\n const iterator = stream[Symbol.asyncIterator]();\r\n\r\n try {\r\n while (true) {\r\n const result = await iterator.next();\r\n if (result.done) {\r\n break;\r\n }\r\n const text = chunkToText(result.value);\r\n if (!text) {\r\n continue;\r\n }\r\n\r\n const payload = mode === \"sse\" ? formatSsePayload(text, options.sse) : text;\r\n res.write(Buffer.from(encoder.encode(payload)));\r\n res.flush?.();\r\n }\r\n } finally {\r\n res.end();\r\n await stream.__usageTapFinalize?.();\r\n }\r\n}\r\n\r\nconst USAGETAP_CORRELATION_HEADER = \"x-usage-correlation-id\";\r\n\r\nexport function wrapOpenAI(\r\n client: OpenAI,\r\n usageTap: UsageTapClient,\r\n options: WrapOpenAIOptions = {},\r\n): WrappedOpenAI {\r\n if (!client) {\r\n throw new UsageTapError(\"USAGETAP_BAD_REQUEST\", \"wrapOpenAI requires an OpenAI client instance\");\r\n }\r\n\r\n const defaultContext = options.defaultContext;\r\n const applyVendorHints = options.applyVendorHints !== false;\r\n\r\n const proxiedChat = client.chat\r\n ? createChatProxy(client.chat, usageTap, defaultContext, applyVendorHints)\r\n : undefined;\r\n\r\n const proxiedResponses = typeof client.responses !== \"undefined\"\r\n ? createResponsesProxy(client.responses, usageTap, defaultContext, applyVendorHints)\r\n : undefined;\r\n\r\n const handler: ProxyHandler<OpenAI> = {\r\n get(target, prop, receiver) {\r\n if (prop === \"chat\" && proxiedChat) {\r\n return proxiedChat;\r\n }\r\n\r\n if (prop === \"responses\" && typeof target.responses !== \"undefined\") {\r\n return proxiedResponses ?? (Reflect.get(target as object, prop, receiver) as unknown);\r\n }\r\n\r\n if (prop === \"toNextResponse\") {\r\n return toNextResponse;\r\n }\r\n\r\n if (prop === \"pipeToResponse\") {\r\n return pipeToResponse;\r\n }\r\n\r\n if (prop === \"unwrap\") {\r\n return () => target;\r\n }\r\n\r\n return Reflect.get(target as object, prop, receiver) as unknown;\r\n },\r\n };\r\n\r\n return new Proxy(client, handler) as WrappedOpenAI;\r\n}\r\n\r\nexport function streamOpenAIRoute(\r\n usageTap: UsageTapClient,\r\n openai: OpenAI,\r\n options: StreamOpenAIRouteOptions,\r\n): (req: Request) => Promise<Response> {\r\n if (!options?.getRequest) {\r\n throw new UsageTapError(\"USAGETAP_BAD_REQUEST\", \"streamOpenAIRoute requires a getRequest function\");\r\n }\r\n\r\n const wrapConfig: WrapOpenAIOptions | undefined = options.wrapOptions || options.defaultContext\r\n ? {\r\n ...(options.wrapOptions ?? {}),\r\n defaultContext: options.defaultContext ?? options.wrapOptions?.defaultContext,\r\n }\r\n : undefined;\r\n\r\n const wrappedClient = wrapConfig\r\n ? wrapOpenAI(openai, usageTap, wrapConfig)\r\n : wrapOpenAI(openai, usageTap);\r\n\r\n return async (req: Request): Promise<Response> => {\r\n const requestConfig = await options.getRequest(req);\r\n const mergedParams: ChatCompletionCreateParams = {\r\n ...requestConfig.params,\r\n stream: true,\r\n };\r\n\r\n const callOptions: Partial<WrapOpenAICallOptions> = {};\r\n if (requestConfig.usageTap) {\r\n callOptions.usageTap = requestConfig.usageTap;\r\n }\r\n if (requestConfig.withUsage) {\r\n callOptions.withUsage = requestConfig.withUsage;\r\n }\r\n\r\n const stream = await wrappedClient.chat.completions.create(\r\n mergedParams,\r\n Object.keys(callOptions).length ? (callOptions as WrapOpenAICallOptions) : undefined,\r\n );\r\n\r\n const baseResponse = toNextResponse(stream as UsageTapStream<unknown>, {\r\n mode: options.stream?.mode ?? \"sse\",\r\n headers: options.stream?.headers,\r\n });\r\n\r\n const init = options.stream?.responseInit;\r\n if (!init) {\r\n return baseResponse;\r\n }\r\n\r\n const mergedHeaders = new Headers(baseResponse.headers);\r\n if (init.headers) {\r\n const extra = normalizeHeaders(init.headers);\r\n for (const [key, value] of Object.entries(extra)) {\r\n mergedHeaders.set(key, value);\r\n }\r\n }\r\n\r\n return new Response(baseResponse.body, {\r\n status: init.status ?? baseResponse.status,\r\n statusText: init.statusText ?? baseResponse.statusText,\r\n headers: mergedHeaders,\r\n });\r\n };\r\n}\r\n\r\nexport interface NodeResponseLike {\r\n write(chunk: string | Uint8Array | Buffer): unknown;\r\n end(chunk?: string | Uint8Array | Buffer): unknown;\r\n setHeader?(name: string, value: string): void;\r\n headersSent?: boolean;\r\n statusCode?: number;\r\n status?(code: number): void;\r\n flush?(): void;\r\n}\r\n\r\nfunction createChatProxy(\r\n resource: OpenAI[\"chat\"],\r\n usageTap: UsageTapClient,\r\n defaultContext: Partial<WrapOpenAIContext> | undefined,\r\n applyVendorHints: boolean,\r\n): ReplaceProperty<OpenAI[\"chat\"], \"completions\", WrappedChatCompletions> {\r\n const completions = createChatCompletionsProxy(\r\n resource.completions,\r\n usageTap,\r\n defaultContext,\r\n applyVendorHints,\r\n );\r\n\r\n const handler: ProxyHandler<OpenAI[\"chat\"]> = {\r\n get(target, prop, receiver) {\r\n if (prop === \"completions\") {\r\n return completions;\r\n }\r\n return Reflect.get(target as object, prop, receiver) as unknown;\r\n },\r\n };\r\n\r\n return new Proxy(resource, handler) as ReplaceProperty<OpenAI[\"chat\"], \"completions\", WrappedChatCompletions>;\r\n}\r\n\r\nfunction createResponsesProxy(\r\n resource: ResponsesResource,\r\n usageTap: UsageTapClient,\r\n defaultContext: Partial<WrapOpenAIContext> | undefined,\r\n applyVendorHints: boolean,\r\n): WrappedResponses | undefined {\r\n if (!resource || typeof resource !== \"object\") {\r\n return undefined;\r\n }\r\n\r\n if (!(\"create\" in resource) || typeof (resource as { create?: unknown }).create !== \"function\") {\r\n return resource as unknown as WrappedResponses;\r\n }\r\n\r\n const originalCreate = (resource as { create: (...args: unknown[]) => unknown }).create.bind(resource);\r\n\r\n const wrappedCreate = (\r\n params: ResponsesCreateParams,\r\n options?: WrapOpenAIResponseCallOptions,\r\n ): ResponsesCreateReturn => {\r\n const { requestOptions, usageContext, withUsage } = splitUsageOptions(options);\r\n const beginRequest = resolveBeginRequest(defaultContext, usageContext);\r\n const wantsStream = isStreamingRequest(params);\r\n\r\n return usageTap.withUsage(beginRequest, (ctx) => {\r\n const finalParams = applyVendorHints\r\n ? applyResponsesVendorHints(params, ctx.begin.data.vendorHints)\r\n : params;\r\n const request = attachCorrelationHeader(requestOptions, ctx.begin.correlationId) as ResponsesCreateOptions;\r\n\r\n if (wantsStream) {\r\n const apiPromise = originalCreate(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (rawStream) => {\r\n ensureAsyncIterable(rawStream, \"responses.create\");\r\n const wrappedStream = wrapStreamForUsageTap(rawStream, async () => {\r\n const usage = await extractUsageFromStream(rawStream, ctx.begin.data.vendorHints);\r\n if (usage) {\r\n ctx.setUsage(usage);\r\n }\r\n }, ctx);\r\n return wrappedStream;\r\n });\r\n\r\n return wrappedPromise as unknown as ResponsesCreateReturn;\r\n }\r\n\r\n const apiPromise = originalCreate(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (response) => {\r\n tryInferUsage(response, ctx.begin.data.vendorHints, undefined, ctx);\r\n return response;\r\n });\r\n return wrappedPromise as unknown as ResponsesCreateReturn;\r\n }, withUsage) as ResponsesCreateReturn;\r\n };\r\n\r\n const handler: ProxyHandler<object> = {\r\n get(target, prop, receiver) {\r\n if (prop === \"create\") {\r\n return wrappedCreate;\r\n }\r\n return Reflect.get(target, prop, receiver) as unknown;\r\n },\r\n };\r\n\r\n return new Proxy(resource as object, handler) as WrappedResponses;\r\n}\r\n\r\nfunction createChatCompletionsProxy(\r\n resource: ChatCompletionsResource,\r\n usageTap: UsageTapClient,\r\n defaultContext: Partial<WrapOpenAIContext> | undefined,\r\n applyVendorHints: boolean,\r\n): WrappedChatCompletions {\r\n const originalCreate = resource.create.bind(resource);\r\n const streamCandidate = (resource as { stream?: unknown }).stream;\r\n const originalStream = typeof streamCandidate === \"function\"\r\n ? (streamCandidate as (...args: unknown[]) => unknown).bind(resource)\r\n : undefined;\r\n\r\n const wrappedCreate = (\r\n params: ChatCompletionCreateParams,\r\n options?: WrapOpenAICallOptions,\r\n ): ChatCompletionCreateReturn => {\r\n const { requestOptions, usageContext, withUsage } = splitUsageOptions(options);\r\n const beginRequest = resolveBeginRequest(defaultContext, usageContext);\r\n const wantsStream = isStreamingRequest(params);\r\n\r\n return usageTap.withUsage(beginRequest, (ctx) => {\r\n const finalParams = applyVendorHints\r\n ? applyChatVendorHints(params, ctx.begin.data.vendorHints)\r\n : params;\r\n const request = attachCorrelationHeader(requestOptions, ctx.begin.correlationId) as ChatCompletionCreateOptions;\r\n\r\n if (wantsStream) {\r\n const apiPromise = originalCreate(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (rawStream) => {\r\n ensureAsyncIterable(rawStream, \"chat.completions.create\");\r\n const wrappedStream = wrapStreamForUsageTap(rawStream, async () => {\r\n const usage = await extractUsageFromStream(rawStream, ctx.begin.data.vendorHints);\r\n if (usage) {\r\n ctx.setUsage(usage);\r\n }\r\n }, ctx);\r\n return wrappedStream;\r\n });\r\n\r\n return wrappedPromise as unknown as ChatCompletionCreateReturn;\r\n }\r\n\r\n const apiPromise = originalCreate(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (response) => {\r\n tryInferUsage(response, ctx.begin.data.vendorHints, undefined, ctx);\r\n return response;\r\n });\r\n return wrappedPromise as unknown as ChatCompletionCreateReturn;\r\n }, withUsage) as ChatCompletionCreateReturn;\r\n };\r\n\r\n const wrappedStream = originalStream\r\n ? (\r\n params: ChatCompletionCreateParams,\r\n options?: WrapOpenAICallOptions,\r\n ): ChatCompletionCreateReturn => {\r\n const { requestOptions, usageContext, withUsage } = splitUsageOptions(options);\r\n const beginRequest = resolveBeginRequest(defaultContext, usageContext);\r\n\r\n return usageTap.withUsage(beginRequest, (ctx) => {\r\n const finalParams = applyVendorHints\r\n ? applyChatVendorHints(params, ctx.begin.data.vendorHints)\r\n : params;\r\n const request = attachCorrelationHeader(requestOptions, ctx.begin.correlationId) as ChatCompletionCreateOptions;\r\n\r\n const apiPromise = originalStream(finalParams, request);\r\n const wrappedPromise = transformApiPromise(apiPromise, (rawStream) => {\r\n ensureAsyncIterable(rawStream, \"chat.completions.stream\");\r\n const wrappedStreamInner = wrapStreamForUsageTap(rawStream, async () => {\r\n const usage = await extractUsageFromStream(rawStream, ctx.begin.data.vendorHints);\r\n if (usage) {\r\n ctx.setUsage(usage);\r\n }\r\n }, ctx);\r\n return wrappedStreamInner;\r\n });\r\n\r\n return wrappedPromise as unknown as ChatCompletionCreateReturn;\r\n }, withUsage) as ChatCompletionCreateReturn;\r\n }\r\n : undefined;\r\n\r\n const handler: ProxyHandler<ChatCompletionsResource> = {\r\n get(target, prop, receiver) {\r\n if (prop === \"create\") {\r\n return wrappedCreate;\r\n }\r\n\r\n if (prop === \"stream\" && wrappedStream) {\r\n return wrappedStream;\r\n }\r\n\r\n return Reflect.get(target as object, prop, receiver) as unknown;\r\n },\r\n };\r\n\r\n return new Proxy(resource, handler) as unknown as WrappedChatCompletions;\r\n}\r\n\r\ninterface SplitUsageOptionsResult {\r\n requestOptions?: Record<string, unknown>;\r\n usageContext?: Partial<WrapOpenAIContext>;\r\n withUsage?: WithUsageOptions;\r\n}\r\n\r\nfunction splitUsageOptions(options: unknown): SplitUsageOptionsResult {\r\n if (!options || typeof options !== \"object\") {\r\n return {};\r\n }\r\n\r\n const { usageTap, withUsage, ...rest } = options as {\r\n usageTap?: Partial<WrapOpenAIContext>;\r\n withUsage?: WithUsageOptions;\r\n } & Record<string, unknown>;\r\n\r\n const requestOptions = Object.keys(rest).length ? cloneRequestOptions(rest) : undefined;\r\n\r\n return {\r\n requestOptions,\r\n usageContext: usageTap,\r\n withUsage,\r\n } satisfies SplitUsageOptionsResult;\r\n}\r\n\r\nfunction resolveBeginRequest(\r\n defaults: Partial<WrapOpenAIContext> | undefined,\r\n override: Partial<WrapOpenAIContext> | undefined,\r\n): BeginCallRequest {\r\n const base = defaults ?? {};\r\n const current = override ?? {};\r\n const customerId = current.customerId ?? base.customerId;\r\n\r\n if (!customerId) {\r\n throw new UsageTapError(\r\n \"USAGETAP_BAD_REQUEST\",\r\n \"wrapOpenAI requires usageTap.customerId (provide defaultContext or options.usageTap)\",\r\n );\r\n }\r\n\r\n const tags = mergeTags(base.tags, current.tags);\r\n const begin: BeginCallRequest = { customerId } satisfies BeginCallRequest;\r\n\r\n const requested = current.requested ?? base.requested;\r\n if (requested) begin.requested = requested;\r\n\r\n const feature = current.feature ?? base.feature;\r\n if (feature) begin.feature = feature;\r\n\r\n const idempotency = current.idempotency ?? base.idempotency;\r\n if (idempotency) begin.idempotency = idempotency;\r\n\r\n const customerName = current.customerName ?? base.customerName;\r\n if (customerName) begin.customerName = customerName;\r\n\r\n const customerEmail = current.customerEmail ?? base.customerEmail;\r\n if (customerEmail) begin.customerEmail = customerEmail;\r\n\r\n if (tags?.length) {\r\n begin.tags = tags;\r\n }\r\n\r\n return begin;\r\n}\r\n\r\ntype PromiseLikeOrValue<T> = PromiseLike<T> | T;\r\n\r\nfunction transformApiPromise<TValue, TResult>(\r\n apiPromise: PromiseLikeOrValue<TValue>,\r\n onResolve: (value: TValue) => PromiseLike<TResult> | TResult,\r\n): Promise<TResult> {\r\n const resolvedPromise = Promise.resolve(apiPromise).then(onResolve);\r\n\r\n if (isObjectRecord(apiPromise)) {\r\n const proto = Object.getPrototypeOf(apiPromise) as object | null;\r\n if (proto) {\r\n Object.setPrototypeOf(resolvedPromise, proto);\r\n }\r\n\r\n for (const key of Reflect.ownKeys(apiPromise)) {\r\n if (key === \"then\" || key === \"catch\" || key === \"finally\") {\r\n continue;\r\n }\r\n\r\n try {\r\n const descriptor = Object.getOwnPropertyDescriptor(apiPromise, key);\r\n if (descriptor) {\r\n Reflect.defineProperty(resolvedPromise, key, descriptor);\r\n }\r\n } catch {\r\n /* ignore non-configurable properties */\r\n }\r\n }\r\n }\r\n\r\n return resolvedPromise;\r\n}\r\n\r\nfunction isObjectRecord(value: unknown): value is Record<PropertyKey, unknown> {\r\n return typeof value === \"object\" && value !== null;\r\n}\r\n\r\nfunction cloneRecord(value: unknown): Record<string, unknown> {\r\n return isObjectRecord(value) ? { ...(value as Record<string, unknown>) } : {};\r\n}\r\n\r\nfunction isStringTuple(value: unknown): value is [string, string] {\r\n return Array.isArray(value) && value.length >= 2 && typeof value[0] === \"string\" && typeof value[1] === \"string\";\r\n}\r\n\r\nfunction cloneRequestOptions(source: Record<string, unknown>): Record<string, unknown> {\r\n const clone: Record<string, unknown> = { ...source };\r\n\r\n if (\"headers\" in clone) {\r\n clone.headers = normalizeHeaders((clone as { headers?: unknown }).headers);\r\n }\r\n\r\n return clone;\r\n}\r\n\r\nfunction attachCorrelationHeader(\r\n options: Record<string, unknown> | undefined,\r\n correlationId: string,\r\n): Record<string, unknown> | undefined {\r\n const normalized = normalizeHeaders(options?.headers);\r\n\r\n if (correlationId && !normalized[USAGETAP_CORRELATION_HEADER]) {\r\n normalized[USAGETAP_CORRELATION_HEADER] = correlationId;\r\n }\r\n\r\n if (!options) {\r\n return Object.keys(normalized).length\r\n ? ({ headers: normalized } satisfies Record<string, unknown>)\r\n : undefined;\r\n }\r\n\r\n const next = { ...options } satisfies Record<string, unknown>;\r\n if (Object.keys(normalized).length) {\r\n next.headers = normalized;\r\n }\r\n return next;\r\n}\r\n\r\nfunction normalizeHeaders(headers: unknown): Record<string, string> {\r\n if (!headers) {\r\n return {};\r\n }\r\n\r\n if (headers instanceof Headers) {\r\n const result: Record<string, string> = {};\r\n headers.forEach((value, key) => {\r\n result[key.toLowerCase()] = value;\r\n });\r\n return result;\r\n }\r\n\r\n if (Array.isArray(headers)) {\r\n const result: Record<string, string> = {};\r\n for (const entry of headers) {\r\n if (!isStringTuple(entry)) {\r\n continue;\r\n }\r\n const [key, value] = entry;\r\n result[key.toLowerCase()] = value;\r\n }\r\n return result;\r\n }\r\n\r\n if (isObjectRecord(headers)) {\r\n const result: Record<string, string> = {};\r\n const record = headers as Record<string, unknown>;\r\n for (const key of Object.keys(record)) {\r\n const value = record[key];\r\n if (value !== undefined && value !== null) {\r\n result[key.toLowerCase()] = String(value);\r\n }\r\n }\r\n return result;\r\n }\r\n\r\n return {};\r\n}\r\n\r\nfunction mergeTags(a?: string[], b?: string[]): string[] | undefined {\r\n const values = [...(a ?? []), ...(b ?? [])]\r\n .map((value) => (typeof value === \"string\" ? value.trim() : \"\"))\r\n .filter(Boolean);\r\n\r\n if (!values.length) {\r\n return undefined;\r\n }\r\n\r\n return dedupeStrings(values);\r\n}\r\n\r\nfunction dedupeStrings(values: string[]): string[] {\r\n return Array.from(new Set(values));\r\n}\r\n\r\nfunction isStreamingRequest(params: unknown): boolean {\r\n if (!params || typeof params !== \"object\") {\r\n return false;\r\n }\r\n\r\n const stream = (params as { stream?: unknown }).stream;\r\n if (typeof stream === \"boolean\") {\r\n return stream;\r\n }\r\n\r\n return stream != null;\r\n}\r\n\r\nfunction applyChatVendorHints(\r\n params: ChatCompletionCreateParams,\r\n hints: VendorHints | undefined,\r\n): ChatCompletionCreateParams {\r\n if (!hints) {\r\n return params;\r\n }\r\n\r\n const next = cloneRecord(params);\r\n\r\n if (hints.preferredModel && (next.model === undefined || next.model === null)) {\r\n next.model = hints.preferredModel;\r\n }\r\n\r\n if (typeof hints.maxResponseTokens === \"number\" && next.max_tokens == null) {\r\n next.max_tokens = hints.maxResponseTokens;\r\n }\r\n\r\n if (typeof hints.maxInputTokens === \"number\" && (next as { max_input_tokens?: unknown }).max_input_tokens == null) {\r\n (next as { max_input_tokens?: number }).max_input_tokens = hints.maxInputTokens;\r\n }\r\n\r\n return next as unknown as ChatCompletionCreateParams;\r\n}\r\n\r\nfunction applyResponsesVendorHints(\r\n params: ResponsesCreateParams,\r\n hints: VendorHints | undefined,\r\n): ResponsesCreateParams {\r\n if (!hints) {\r\n return params;\r\n }\r\n\r\n const next = cloneRecord(params);\r\n\r\n if (hints.preferredModel && (next.model === undefined || next.model === null)) {\r\n next.model = hints.preferredModel;\r\n }\r\n\r\n if (typeof hints.maxResponseTokens === \"number\" && (next as { max_output_tokens?: unknown }).max_output_tokens == null) {\r\n (next as { max_output_tokens?: number }).max_output_tokens = hints.maxResponseTokens;\r\n }\r\n\r\n return next as unknown as ResponsesCreateParams;\r\n}\r\n\r\nasync function extractUsageFromStream(\r\n stream: unknown,\r\n hints: VendorHints | undefined,\r\n): Promise<Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | undefined> {\r\n const finalPayload = await resolveStreamFinalPayload(stream);\r\n if (!finalPayload) {\r\n return undefined;\r\n }\r\n\r\n return inferUsageFromResponse(finalPayload, hints);\r\n}\r\n\r\nasync function resolveStreamFinalPayload(stream: unknown): Promise<unknown> {\r\n if (!stream || typeof stream !== \"object\") {\r\n return undefined;\r\n }\r\n\r\n const candidate = stream as {\r\n finalChatCompletion?: () => Promise<unknown>;\r\n finalCompletion?: () => Promise<unknown>;\r\n finalResponse?: () => Promise<unknown>;\r\n finalContent?: () => Promise<unknown>;\r\n };\r\n\r\n if (typeof candidate.finalChatCompletion === \"function\") {\r\n return candidate.finalChatCompletion();\r\n }\r\n\r\n if (typeof candidate.finalResponse === \"function\") {\r\n return candidate.finalResponse();\r\n }\r\n\r\n if (typeof candidate.finalCompletion === \"function\") {\r\n return candidate.finalCompletion();\r\n }\r\n\r\n if (typeof candidate.finalContent === \"function\") {\r\n return candidate.finalContent();\r\n }\r\n\r\n return undefined;\r\n}\r\n\r\nfunction ensureAsyncIterable(value: unknown, label: string): asserts value is AsyncIterable<unknown> {\r\n if (!value || typeof value !== \"object\" || typeof (value as AsyncIterable<unknown>)[Symbol.asyncIterator] !== \"function\") {\r\n throw new UsageTapError(\r\n \"USAGETAP_BAD_REQUEST\",\r\n `${label} expected an async iterable stream but received ${typeof value}`,\r\n );\r\n }\r\n}\r\n\r\nfunction chunkToText(chunk: unknown): string {\r\n if (chunk === undefined || chunk === null) {\r\n return \"\";\r\n }\r\n\r\n if (typeof chunk === \"string\") {\r\n return chunk;\r\n }\r\n\r\n if (typeof chunk === \"object\") {\r\n const candidate = chunk as {\r\n choices?: Array<{\r\n delta?: {\r\n content?: string | Array<{ text?: string }>;\r\n };\r\n }>;\r\n content?: string;\r\n };\r\n\r\n const delta = candidate.choices?.[0]?.delta;\r\n const content = delta?.content ?? candidate.content;\r\n\r\n if (typeof content === \"string\") {\r\n return content;\r\n }\r\n\r\n if (Array.isArray(content)) {\r\n return content\r\n .map((entry) => {\r\n if (!entry) return \"\";\r\n if (typeof entry === \"string\") return entry;\r\n if (typeof entry.text === \"string\") return entry.text;\r\n return \"\";\r\n })\r\n .join(\"\");\r\n }\r\n }\r\n\r\n return String(chunk);\r\n}\r\n\r\nfunction formatSsePayload(\r\n text: string,\r\n options: StreamToResponseOptions[\"sse\"],\r\n): string {\r\n if (!text) {\r\n return \"\";\r\n }\r\n\r\n const lines = text.split(/\\r?\\n/);\r\n const eventLine = options?.event ? `event: ${options.event}\\n` : \"\";\r\n const retryLine = options?.retry ? `retry: ${options.retry}\\n` : \"\";\r\n const dataLines = lines.map((line) => `data: ${line}`).join(\"\\n\");\r\n return `${eventLine}${retryLine}${dataLines}\\n\\n`;\r\n}\r\n\r\nfunction setHeaderIfPossible(res: NodeResponseLike, key: string, value: string): void {\r\n if (typeof res.setHeader === \"function\" && res.headersSent !== true) {\r\n res.setHeader(key, value);\r\n }\r\n}\r\n\r\nfunction tryInferUsage<TResponse>(\r\n response: TResponse,\r\n hints: VendorHints | undefined,\r\n extractor: OpenAIInvokeParams<TResponse>[\"extractUsage\"],\r\n ctx: WithUsageContext,\r\n): void {\r\n const explicit = extractor?.(response);\r\n const inferred = explicit ?? inferUsageFromResponse(response, hints);\r\n\r\n if (inferred) {\r\n ctx.setUsage(inferred);\r\n }\r\n}\r\n\r\nfunction inferUsageFromResponse(\r\n response: unknown,\r\n hints: VendorHints | undefined,\r\n): Partial<Omit<EndCallRequest, \"callId\" | \"error\">> | undefined {\r\n if (!response || typeof response !== \"object\") {\r\n return undefined;\r\n }\r\n\r\n const candidate = response as {\r\n usage?: {\r\n prompt_tokens?: number;\r\n completion_tokens?: number;\r\n total_tokens?: number;\r\n cached_tokens?: number;\r\n prompt_tokens_details?: {\r\n cached_tokens?: number;\r\n };\r\n };\r\n model?: string;\r\n };\r\n\r\n if (!candidate.usage) {\r\n return undefined;\r\n }\r\n\r\n const cachedInputTokens =\r\n candidate.usage.prompt_tokens_details?.cached_tokens ??\r\n candidate.usage.cached_tokens;\r\n\r\n return {\r\n modelUsed: candidate.model ?? hints?.preferredModel,\r\n inputTokens: candidate.usage.prompt_tokens,\r\n responseTokens: candidate.usage.completion_tokens,\r\n cachedInputTokens,\r\n } satisfies Partial<Omit<EndCallRequest, \"callId\" | \"error\">>;\r\n}\r\n\r\nfunction wrapStreamForUsageTap<TStream>(\r\n source: UsageTapStream<TStream> | AsyncIterable<TStream>,\r\n finalize: () => Promise<void> | void,\r\n ctx: WithUsageContext,\r\n): AsyncIterable<TStream> & { __usageTapFinalize: () => Promise<void> } {\r\n const getIterator = source[Symbol.asyncIterator];\r\n if (typeof getIterator !== \"function\") {\r\n throw new TypeError(\"Stream is not async iterable\");\r\n }\r\n\r\n const iterator = getIterator.call(source) as AsyncIterator<TStream>;\r\n let completed = false;\r\n\r\n const invokeFinalize = async (): Promise<void> => {\r\n if (completed) return;\r\n completed = true;\r\n try {\r\n await finalize();\r\n } catch (error) {\r\n ctx.setError({\r\n code: \"USAGE_FINALIZE_ERROR\",\r\n message: error instanceof Error ? error.message : String(error),\r\n });\r\n throw error;\r\n }\r\n };\r\n\r\n const prototype = (Object.getPrototypeOf(source as object) as object | null) ?? Object.prototype;\r\n const wrapped = Object.create(prototype) as unknown as UsageTapIterableIterator<TStream>;\r\n\r\n for (const key of Reflect.ownKeys(source as object)) {\r\n try {\r\n const descriptor = Object.getOwnPropertyDescriptor(source as object, key);\r\n if (descriptor) {\r\n Object.defineProperty(wrapped, key, descriptor);\r\n }\r\n } catch {\r\n /* ignore non-configurable properties */\r\n }\r\n }\r\n\r\n Object.defineProperty(wrapped, Symbol.asyncIterator, {\r\n value(): UsageTapIterableIterator<TStream> {\r\n return this as UsageTapIterableIterator<TStream>;\r\n },\r\n configurable: true,\r\n });\r\n\r\n Object.defineProperty(wrapped, \"next\", {\r\n value: async (...args: Parameters<AsyncIterator<TStream>[\"next\"]>): Promise<IteratorResult<TStream>> => {\r\n try {\r\n const result = await iterator.next(...args);\r\n if (result.done) {\r\n await invokeFinalize();\r\n }\r\n return result;\r\n } catch (error) {\r\n await invokeFinalize().catch(() => undefined);\r\n throw error;\r\n }\r\n },\r\n configurable: true,\r\n writable: true,\r\n });\r\n\r\n Object.defineProperty(wrapped, \"return\", {\r\n value: async (value?: TStream): Promise<IteratorResult<TStream>> => {\r\n if (typeof iterator.return === \"function\") {\r\n const rawResult: unknown = await iterator.return(value);\r\n if (!isIteratorResult<TStream>(rawResult)) {\r\n throw new TypeError(\"Iterator.return() returned an invalid result\");\r\n }\r\n await invokeFinalize();\r\n return rawResult;\r\n }\r\n await invokeFinalize();\r\n return { done: true, value } as IteratorResult<TStream>;\r\n },\r\n configurable: true,\r\n writable: true,\r\n });\r\n\r\n Object.defineProperty(wrapped, \"throw\", {\r\n value: async (error?: unknown): Promise<IteratorResult<TStream>> => {\r\n if (typeof iterator.throw === \"function\") {\r\n const rawResult: unknown = await iterator.throw(error);\r\n if (!isIteratorResult<TStream>(rawResult)) {\r\n throw new TypeError(\"Iterator.throw() returned an invalid result\");\r\n }\r\n await invokeFinalize();\r\n return rawResult;\r\n }\r\n await invokeFinalize();\r\n throw error;\r\n },\r\n configurable: true,\r\n writable: true,\r\n });\r\n\r\n Object.defineProperty(wrapped, \"__usageTapFinalize\", {\r\n value: async (): Promise<void> => {\r\n await invokeFinalize();\r\n },\r\n configurable: true,\r\n });\r\n\r\n return wrapped;\r\n}\r\n\r\nfunction isIteratorResult<T>(value: unknown): value is IteratorResult<T> {\r\n return isObjectRecord(value) && \"done\" in value;\r\n}\r\n","import type OpenAI from \"openai\";\r\nimport { UsageTapClient } from \"../client\";\r\nimport type { OpenAIAdapter } from \"./openai\";\r\nimport { createOpenAIAdapter } from \"./openai\";\r\n\r\nexport interface OpenRouterAdapterInit {\r\n client: OpenAI;\r\n usageTap: UsageTapClient;\r\n}\r\n\r\nexport function createOpenRouterAdapter(init: OpenRouterAdapterInit): OpenAIAdapter {\r\n return createOpenAIAdapter(init);\r\n}\r\n"]}
|
|
@@ -21,15 +21,36 @@ interface EntitlementHints {
|
|
|
21
21
|
downgrade?: EntitlementDowngradeHint;
|
|
22
22
|
}
|
|
23
23
|
interface MeterSummary {
|
|
24
|
-
|
|
24
|
+
/** Always numeric. Check the `unlimited` flag to detect unbounded meters. */
|
|
25
|
+
remaining: number;
|
|
25
26
|
limit: number | null;
|
|
26
|
-
|
|
27
|
+
/** Always numeric (0 when no usage recorded). */
|
|
28
|
+
used: number;
|
|
29
|
+
/** When true this meter is unbounded; `remaining` is informational only. */
|
|
27
30
|
unlimited: boolean;
|
|
31
|
+
/** remaining / limit (0-1). null when unlimited or limit is null/0. */
|
|
28
32
|
ratio: number | null;
|
|
29
33
|
label?: string;
|
|
30
34
|
}
|
|
31
35
|
type MeterSnapshot = Record<string, MeterSummary>;
|
|
32
36
|
type RemainingRatios = Record<string, number | null | undefined>;
|
|
37
|
+
interface RollingCallsRateLimitState {
|
|
38
|
+
windowType: "rolling_calls";
|
|
39
|
+
limit: number;
|
|
40
|
+
used: number;
|
|
41
|
+
remaining: number;
|
|
42
|
+
windowSeconds: number;
|
|
43
|
+
state: "fresh" | "estimated" | "stale" | "unknown";
|
|
44
|
+
source?: "current_rolling_check" | "call_records";
|
|
45
|
+
nextAvailableAt?: string;
|
|
46
|
+
scope?: {
|
|
47
|
+
callType?: "standard" | "API";
|
|
48
|
+
tier?: "standard" | "premium";
|
|
49
|
+
};
|
|
50
|
+
}
|
|
51
|
+
interface RateLimitsSnapshot {
|
|
52
|
+
rollingCalls?: RollingCallsRateLimitState | Record<string, RollingCallsRateLimitState>;
|
|
53
|
+
}
|
|
33
54
|
interface SubscriptionSnapshot {
|
|
34
55
|
id: string | null;
|
|
35
56
|
usagePlanVersionId: string | null;
|
|
@@ -82,6 +103,14 @@ interface BeginCallRequest {
|
|
|
82
103
|
customerName?: string;
|
|
83
104
|
customerEmail?: string;
|
|
84
105
|
stripeCustomerId?: string;
|
|
106
|
+
/**
|
|
107
|
+
* Whether this call uses batch pricing (typically 50% of standard rates).
|
|
108
|
+
*/
|
|
109
|
+
batch?: boolean;
|
|
110
|
+
/**
|
|
111
|
+
* Explicit pricing mode for this call.
|
|
112
|
+
*/
|
|
113
|
+
pricingMode?: "batch" | "standard";
|
|
85
114
|
}
|
|
86
115
|
interface BalanceSummary {
|
|
87
116
|
standardCallsRemaining?: number;
|
|
@@ -92,6 +121,23 @@ interface BalanceSummary {
|
|
|
92
121
|
customMeter1Remaining?: number;
|
|
93
122
|
customMeter2Remaining?: number;
|
|
94
123
|
}
|
|
124
|
+
type SpendVelocityWindowKey = "hour" | "day";
|
|
125
|
+
interface SpendVelocityWindow {
|
|
126
|
+
bucket: string;
|
|
127
|
+
windowMinutes: number;
|
|
128
|
+
startedAt: string;
|
|
129
|
+
endedAt: string;
|
|
130
|
+
completedCostUsd: number;
|
|
131
|
+
completedCalls: number;
|
|
132
|
+
}
|
|
133
|
+
interface SpendVelocitySnapshot {
|
|
134
|
+
currency: "USD";
|
|
135
|
+
source: "usage_aggregate";
|
|
136
|
+
generatedAt: string;
|
|
137
|
+
customerId: string;
|
|
138
|
+
currentCallCostUsd: number;
|
|
139
|
+
windows: Record<SpendVelocityWindowKey, SpendVelocityWindow>;
|
|
140
|
+
}
|
|
95
141
|
interface PlanSummary {
|
|
96
142
|
id: string | null;
|
|
97
143
|
name: string | null;
|
|
@@ -107,6 +153,7 @@ interface BeginCallResponseBody {
|
|
|
107
153
|
policy: LimitType;
|
|
108
154
|
allowed: AllowedEntitlements;
|
|
109
155
|
entitlementHints: EntitlementHints;
|
|
156
|
+
rateLimits?: RateLimitsSnapshot;
|
|
110
157
|
meters: MeterSnapshot;
|
|
111
158
|
remainingRatios: RemainingRatios;
|
|
112
159
|
subscription: SubscriptionSnapshot;
|
|
@@ -116,6 +163,8 @@ interface BeginCallResponseBody {
|
|
|
116
163
|
plan?: PlanSummary;
|
|
117
164
|
balances?: BalanceSummary;
|
|
118
165
|
stripeCustomerId?: string | null;
|
|
166
|
+
batch?: boolean;
|
|
167
|
+
pricingMode?: "batch" | "standard";
|
|
119
168
|
}
|
|
120
169
|
interface EndCallRequest {
|
|
121
170
|
callId: string;
|
|
@@ -126,9 +175,16 @@ interface EndCallRequest {
|
|
|
126
175
|
/** Optional tags for metric tracking. Not sent to API, used for onUsageMetric callback. */
|
|
127
176
|
tags?: string[];
|
|
128
177
|
modelUsed?: string;
|
|
178
|
+
/**
|
|
179
|
+
* Total prompt/input tokens for the call (including any cached prompt tokens).
|
|
180
|
+
*/
|
|
129
181
|
inputTokens?: number;
|
|
130
182
|
responseTokens?: number;
|
|
131
|
-
|
|
183
|
+
/**
|
|
184
|
+
* Prompt/input tokens served from cache.
|
|
185
|
+
* Preferred over legacy aliases.
|
|
186
|
+
*/
|
|
187
|
+
cachedInputTokens?: number;
|
|
132
188
|
reasoningTokens?: number;
|
|
133
189
|
searches?: number;
|
|
134
190
|
audio?: number;
|
|
@@ -138,6 +194,14 @@ interface EndCallRequest {
|
|
|
138
194
|
message: string;
|
|
139
195
|
};
|
|
140
196
|
stripeCustomerId?: string;
|
|
197
|
+
/**
|
|
198
|
+
* Whether this call uses batch pricing (typically 50% of standard rates).
|
|
199
|
+
*/
|
|
200
|
+
batch?: boolean;
|
|
201
|
+
/**
|
|
202
|
+
* Explicit pricing mode for this call.
|
|
203
|
+
*/
|
|
204
|
+
pricingMode?: "batch" | "standard";
|
|
141
205
|
}
|
|
142
206
|
interface MeteredUsage {
|
|
143
207
|
calls?: number;
|
|
@@ -150,9 +214,19 @@ interface MeteredUsage {
|
|
|
150
214
|
interface EndCallResponseBody {
|
|
151
215
|
callId: string;
|
|
152
216
|
costUSD: number;
|
|
217
|
+
usage?: {
|
|
218
|
+
inputTokens: number;
|
|
219
|
+
cachedInputTokens: number;
|
|
220
|
+
billableInputTokens: number;
|
|
221
|
+
responseTokens: number;
|
|
222
|
+
reasoningTokens: number;
|
|
223
|
+
};
|
|
153
224
|
metered?: MeteredUsage;
|
|
225
|
+
spendVelocity?: SpendVelocitySnapshot;
|
|
154
226
|
balances?: BalanceSummary;
|
|
155
227
|
stripeCustomerId?: string | null;
|
|
228
|
+
batch?: boolean;
|
|
229
|
+
pricingMode?: "batch" | "standard";
|
|
156
230
|
}
|
|
157
231
|
type UsageTapResultStatus = "ACCEPTED" | "ERROR";
|
|
158
232
|
interface UsageTapResultEnvelope {
|
|
@@ -274,6 +348,7 @@ interface CheckUsageResponseBody {
|
|
|
274
348
|
policy: LimitType;
|
|
275
349
|
allowed: AllowedEntitlements;
|
|
276
350
|
entitlementHints: EntitlementHints;
|
|
351
|
+
rateLimits?: RateLimitsSnapshot;
|
|
277
352
|
meters: MeterSnapshot;
|
|
278
353
|
remainingRatios: RemainingRatios;
|
|
279
354
|
subscription: SubscriptionSnapshot;
|
|
@@ -353,8 +428,8 @@ interface UsageMetricEvent {
|
|
|
353
428
|
/** Usage metrics */
|
|
354
429
|
metrics: {
|
|
355
430
|
inputTokens?: number;
|
|
431
|
+
cachedInputTokens?: number;
|
|
356
432
|
responseTokens?: number;
|
|
357
|
-
cachedTokens?: number;
|
|
358
433
|
reasoningTokens?: number;
|
|
359
434
|
searches?: number;
|
|
360
435
|
audioSeconds?: number;
|
|
@@ -401,4 +476,4 @@ declare class UsageTapClient {
|
|
|
401
476
|
private toApiError;
|
|
402
477
|
}
|
|
403
478
|
|
|
404
|
-
export { type AllowedEntitlements as A, type BeginCallRequest as B, type CreateCustomerOptions as C, type
|
|
479
|
+
export { type AllowedEntitlements as A, type BeginCallRequest as B, type CreateCustomerOptions as C, type RetryOptions as D, type EndCallRequest as E, type UsageTapClientOptions as F, type UsageTapErrorResponse as G, type UsageTapResultEnvelope as H, type IncrementCustomMeterOptions as I, type UsageTapResultStatus as J, type UsageTapLogEntry as K, type WithUsageContext as L, type MeterSummary as M, type LimitType as N, type SubscriptionSnapshot as O, type PlanSummary as P, type ModelHints as Q, type RateLimitsSnapshot as R, type SpendVelocitySnapshot as S, type IdempotencyMetadata as T, UsageTapClient as U, type VendorHints as V, type WithUsageOptions as W, type UsageMetricEvent as X, type UsageTapSuccessResponse as a, type BeginCallResponseBody as b, type BeginCallOptions as c, type CreateCustomerRequest as d, type CreateCustomerResponseBody as e, type CheckUsageOptions as f, type CheckUsageRequest as g, type CheckUsageResponseBody as h, type ChangePlanOptions as i, type ChangePlanRequest as j, type ChangePlanResponseBody as k, type ChangePlanStrategy as l, type IncrementCustomMeterRequest as m, type IncrementCustomMeterResponseBody as n, type CustomMeterSlot as o, type EndCallOptions as p, type EndCallResponseBody as q, type BalanceSummary as r, type SpendVelocityWindow as s, type SpendVelocityWindowKey as t, type EntitlementHints as u, type MeterSnapshot as v, type MeteredUsage as w, type RemainingRatios as x, type RollingCallsRateLimitState as y, type RequestedEntitlements as z };
|
|
@@ -21,15 +21,36 @@ interface EntitlementHints {
|
|
|
21
21
|
downgrade?: EntitlementDowngradeHint;
|
|
22
22
|
}
|
|
23
23
|
interface MeterSummary {
|
|
24
|
-
|
|
24
|
+
/** Always numeric. Check the `unlimited` flag to detect unbounded meters. */
|
|
25
|
+
remaining: number;
|
|
25
26
|
limit: number | null;
|
|
26
|
-
|
|
27
|
+
/** Always numeric (0 when no usage recorded). */
|
|
28
|
+
used: number;
|
|
29
|
+
/** When true this meter is unbounded; `remaining` is informational only. */
|
|
27
30
|
unlimited: boolean;
|
|
31
|
+
/** remaining / limit (0-1). null when unlimited or limit is null/0. */
|
|
28
32
|
ratio: number | null;
|
|
29
33
|
label?: string;
|
|
30
34
|
}
|
|
31
35
|
type MeterSnapshot = Record<string, MeterSummary>;
|
|
32
36
|
type RemainingRatios = Record<string, number | null | undefined>;
|
|
37
|
+
interface RollingCallsRateLimitState {
|
|
38
|
+
windowType: "rolling_calls";
|
|
39
|
+
limit: number;
|
|
40
|
+
used: number;
|
|
41
|
+
remaining: number;
|
|
42
|
+
windowSeconds: number;
|
|
43
|
+
state: "fresh" | "estimated" | "stale" | "unknown";
|
|
44
|
+
source?: "current_rolling_check" | "call_records";
|
|
45
|
+
nextAvailableAt?: string;
|
|
46
|
+
scope?: {
|
|
47
|
+
callType?: "standard" | "API";
|
|
48
|
+
tier?: "standard" | "premium";
|
|
49
|
+
};
|
|
50
|
+
}
|
|
51
|
+
interface RateLimitsSnapshot {
|
|
52
|
+
rollingCalls?: RollingCallsRateLimitState | Record<string, RollingCallsRateLimitState>;
|
|
53
|
+
}
|
|
33
54
|
interface SubscriptionSnapshot {
|
|
34
55
|
id: string | null;
|
|
35
56
|
usagePlanVersionId: string | null;
|
|
@@ -82,6 +103,14 @@ interface BeginCallRequest {
|
|
|
82
103
|
customerName?: string;
|
|
83
104
|
customerEmail?: string;
|
|
84
105
|
stripeCustomerId?: string;
|
|
106
|
+
/**
|
|
107
|
+
* Whether this call uses batch pricing (typically 50% of standard rates).
|
|
108
|
+
*/
|
|
109
|
+
batch?: boolean;
|
|
110
|
+
/**
|
|
111
|
+
* Explicit pricing mode for this call.
|
|
112
|
+
*/
|
|
113
|
+
pricingMode?: "batch" | "standard";
|
|
85
114
|
}
|
|
86
115
|
interface BalanceSummary {
|
|
87
116
|
standardCallsRemaining?: number;
|
|
@@ -92,6 +121,23 @@ interface BalanceSummary {
|
|
|
92
121
|
customMeter1Remaining?: number;
|
|
93
122
|
customMeter2Remaining?: number;
|
|
94
123
|
}
|
|
124
|
+
type SpendVelocityWindowKey = "hour" | "day";
|
|
125
|
+
interface SpendVelocityWindow {
|
|
126
|
+
bucket: string;
|
|
127
|
+
windowMinutes: number;
|
|
128
|
+
startedAt: string;
|
|
129
|
+
endedAt: string;
|
|
130
|
+
completedCostUsd: number;
|
|
131
|
+
completedCalls: number;
|
|
132
|
+
}
|
|
133
|
+
interface SpendVelocitySnapshot {
|
|
134
|
+
currency: "USD";
|
|
135
|
+
source: "usage_aggregate";
|
|
136
|
+
generatedAt: string;
|
|
137
|
+
customerId: string;
|
|
138
|
+
currentCallCostUsd: number;
|
|
139
|
+
windows: Record<SpendVelocityWindowKey, SpendVelocityWindow>;
|
|
140
|
+
}
|
|
95
141
|
interface PlanSummary {
|
|
96
142
|
id: string | null;
|
|
97
143
|
name: string | null;
|
|
@@ -107,6 +153,7 @@ interface BeginCallResponseBody {
|
|
|
107
153
|
policy: LimitType;
|
|
108
154
|
allowed: AllowedEntitlements;
|
|
109
155
|
entitlementHints: EntitlementHints;
|
|
156
|
+
rateLimits?: RateLimitsSnapshot;
|
|
110
157
|
meters: MeterSnapshot;
|
|
111
158
|
remainingRatios: RemainingRatios;
|
|
112
159
|
subscription: SubscriptionSnapshot;
|
|
@@ -116,6 +163,8 @@ interface BeginCallResponseBody {
|
|
|
116
163
|
plan?: PlanSummary;
|
|
117
164
|
balances?: BalanceSummary;
|
|
118
165
|
stripeCustomerId?: string | null;
|
|
166
|
+
batch?: boolean;
|
|
167
|
+
pricingMode?: "batch" | "standard";
|
|
119
168
|
}
|
|
120
169
|
interface EndCallRequest {
|
|
121
170
|
callId: string;
|
|
@@ -126,9 +175,16 @@ interface EndCallRequest {
|
|
|
126
175
|
/** Optional tags for metric tracking. Not sent to API, used for onUsageMetric callback. */
|
|
127
176
|
tags?: string[];
|
|
128
177
|
modelUsed?: string;
|
|
178
|
+
/**
|
|
179
|
+
* Total prompt/input tokens for the call (including any cached prompt tokens).
|
|
180
|
+
*/
|
|
129
181
|
inputTokens?: number;
|
|
130
182
|
responseTokens?: number;
|
|
131
|
-
|
|
183
|
+
/**
|
|
184
|
+
* Prompt/input tokens served from cache.
|
|
185
|
+
* Preferred over legacy aliases.
|
|
186
|
+
*/
|
|
187
|
+
cachedInputTokens?: number;
|
|
132
188
|
reasoningTokens?: number;
|
|
133
189
|
searches?: number;
|
|
134
190
|
audio?: number;
|
|
@@ -138,6 +194,14 @@ interface EndCallRequest {
|
|
|
138
194
|
message: string;
|
|
139
195
|
};
|
|
140
196
|
stripeCustomerId?: string;
|
|
197
|
+
/**
|
|
198
|
+
* Whether this call uses batch pricing (typically 50% of standard rates).
|
|
199
|
+
*/
|
|
200
|
+
batch?: boolean;
|
|
201
|
+
/**
|
|
202
|
+
* Explicit pricing mode for this call.
|
|
203
|
+
*/
|
|
204
|
+
pricingMode?: "batch" | "standard";
|
|
141
205
|
}
|
|
142
206
|
interface MeteredUsage {
|
|
143
207
|
calls?: number;
|
|
@@ -150,9 +214,19 @@ interface MeteredUsage {
|
|
|
150
214
|
interface EndCallResponseBody {
|
|
151
215
|
callId: string;
|
|
152
216
|
costUSD: number;
|
|
217
|
+
usage?: {
|
|
218
|
+
inputTokens: number;
|
|
219
|
+
cachedInputTokens: number;
|
|
220
|
+
billableInputTokens: number;
|
|
221
|
+
responseTokens: number;
|
|
222
|
+
reasoningTokens: number;
|
|
223
|
+
};
|
|
153
224
|
metered?: MeteredUsage;
|
|
225
|
+
spendVelocity?: SpendVelocitySnapshot;
|
|
154
226
|
balances?: BalanceSummary;
|
|
155
227
|
stripeCustomerId?: string | null;
|
|
228
|
+
batch?: boolean;
|
|
229
|
+
pricingMode?: "batch" | "standard";
|
|
156
230
|
}
|
|
157
231
|
type UsageTapResultStatus = "ACCEPTED" | "ERROR";
|
|
158
232
|
interface UsageTapResultEnvelope {
|
|
@@ -274,6 +348,7 @@ interface CheckUsageResponseBody {
|
|
|
274
348
|
policy: LimitType;
|
|
275
349
|
allowed: AllowedEntitlements;
|
|
276
350
|
entitlementHints: EntitlementHints;
|
|
351
|
+
rateLimits?: RateLimitsSnapshot;
|
|
277
352
|
meters: MeterSnapshot;
|
|
278
353
|
remainingRatios: RemainingRatios;
|
|
279
354
|
subscription: SubscriptionSnapshot;
|
|
@@ -353,8 +428,8 @@ interface UsageMetricEvent {
|
|
|
353
428
|
/** Usage metrics */
|
|
354
429
|
metrics: {
|
|
355
430
|
inputTokens?: number;
|
|
431
|
+
cachedInputTokens?: number;
|
|
356
432
|
responseTokens?: number;
|
|
357
|
-
cachedTokens?: number;
|
|
358
433
|
reasoningTokens?: number;
|
|
359
434
|
searches?: number;
|
|
360
435
|
audioSeconds?: number;
|
|
@@ -401,4 +476,4 @@ declare class UsageTapClient {
|
|
|
401
476
|
private toApiError;
|
|
402
477
|
}
|
|
403
478
|
|
|
404
|
-
export { type AllowedEntitlements as A, type BeginCallRequest as B, type CreateCustomerOptions as C, type
|
|
479
|
+
export { type AllowedEntitlements as A, type BeginCallRequest as B, type CreateCustomerOptions as C, type RetryOptions as D, type EndCallRequest as E, type UsageTapClientOptions as F, type UsageTapErrorResponse as G, type UsageTapResultEnvelope as H, type IncrementCustomMeterOptions as I, type UsageTapResultStatus as J, type UsageTapLogEntry as K, type WithUsageContext as L, type MeterSummary as M, type LimitType as N, type SubscriptionSnapshot as O, type PlanSummary as P, type ModelHints as Q, type RateLimitsSnapshot as R, type SpendVelocitySnapshot as S, type IdempotencyMetadata as T, UsageTapClient as U, type VendorHints as V, type WithUsageOptions as W, type UsageMetricEvent as X, type UsageTapSuccessResponse as a, type BeginCallResponseBody as b, type BeginCallOptions as c, type CreateCustomerRequest as d, type CreateCustomerResponseBody as e, type CheckUsageOptions as f, type CheckUsageRequest as g, type CheckUsageResponseBody as h, type ChangePlanOptions as i, type ChangePlanRequest as j, type ChangePlanResponseBody as k, type ChangePlanStrategy as l, type IncrementCustomMeterRequest as m, type IncrementCustomMeterResponseBody as n, type CustomMeterSlot as o, type EndCallOptions as p, type EndCallResponseBody as q, type BalanceSummary as r, type SpendVelocityWindow as s, type SpendVelocityWindowKey as t, type EntitlementHints as u, type MeterSnapshot as v, type MeteredUsage as w, type RemainingRatios as x, type RollingCallsRateLimitState as y, type RequestedEntitlements as z };
|
package/dist/express/index.cjs
CHANGED
|
@@ -548,11 +548,12 @@ function inferUsageFromResponse(response, hints) {
|
|
|
548
548
|
if (!candidate.usage) {
|
|
549
549
|
return void 0;
|
|
550
550
|
}
|
|
551
|
+
const cachedInputTokens = candidate.usage.prompt_tokens_details?.cached_tokens ?? candidate.usage.cached_tokens;
|
|
551
552
|
return {
|
|
552
553
|
modelUsed: candidate.model ?? hints?.preferredModel,
|
|
553
554
|
inputTokens: candidate.usage.prompt_tokens,
|
|
554
555
|
responseTokens: candidate.usage.completion_tokens,
|
|
555
|
-
|
|
556
|
+
cachedInputTokens
|
|
556
557
|
};
|
|
557
558
|
}
|
|
558
559
|
function wrapStreamForUsageTap(source, finalize, ctx) {
|