@usagetap/sdk 1.3.2 → 1.4.0

package/dist/adapters/openai.d.cts

@@ -1,9 +1,11 @@
 import OpenAI from 'openai';
-import { B as BeginCallRequest, P as PromptCompressionProvider, U as UsageTapClient, W as WithUsageOptions, a as PromptCompressionTelemetry, V as VendorHints, b as UsageTapSuccessResponse, c as BeginCallResponseBody, E as EndCallRequest } from '../client-BD8O2J8Z.cjs';
+import { B as BeginCallRequest, P as PromptCompressionProvider, a as PromptCompressionMode, S as SamplingOptions, U as UsageTapClient, W as WithUsageOptions, b as PromptCompressionTelemetry, V as VendorHints, c as UsageTapSuccessResponse, d as BeginCallResponseBody, E as EndCallRequest } from '../client-CExQ8e1T.cjs';
 
 interface OpenAIAdapterInit {
     client: OpenAI;
     usageTap: UsageTapClient;
+    /** Vendor label recorded at call_end. Defaults to `openai`. */
+    provider?: string;
 }
 interface OpenAIRequestContext {
     hints?: VendorHints;
@@ -65,6 +67,14 @@ interface OpenAIPromptCompressionOptions {
     /** Minimum estimated tokens for each individual text segment. */
     minTokens?: number;
     failOpen?: boolean;
+    /** Select deterministic-only, automatic model gating, or a forced model pass. */
+    mode?: PromptCompressionMode;
+    /** Maximum compression latency the service should target. */
+    latencyBudgetMs?: number;
+    /** Remove empty user messages before compression. */
+    compactEmptyUserMessages?: boolean;
+    /** Deduplicate repeated text parts within user messages. */
+    compactDuplicateUserTextParts?: boolean;
     tokenCompanyModel?: string;
     aggressiveness?: number | Partial<Record<OpenAIPromptCompressionRole, number>>;
     /** @deprecated Use aggressiveness instead. */
@@ -74,7 +84,13 @@ interface OpenAIPromptCompressionOptions {
     /** @deprecated Use aggressiveness instead. */
     usageTapCompressionAggressiveness?: number | Partial<Record<OpenAIPromptCompressionRole, number>>;
 }
-interface OpenAIPromptCompressionTurnStats extends PromptCompressionTelemetry {
+interface PromptCompressionTelemetryWithDiagnostics extends PromptCompressionTelemetry {
+    originalCharacters: number;
+    compressedCharacters: number;
+    savedCharacters: number;
+    savingsRatio: number;
+}
+interface OpenAIPromptCompressionTurnStats extends PromptCompressionTelemetryWithDiagnostics {
     callId: string;
     operation: "chat.completions.create" | "chat.completions.stream" | "responses.create";
     messagesCompressed: number;
@@ -105,16 +121,37 @@ declare class OpenAIPromptCompressionStats {
     get savingsRatio(): number;
 }
 interface WrapOpenAIOptions {
+    /** Vendor label recorded at call_end. Defaults to `openai`. */
+    provider?: string;
     defaultContext?: Partial<WrapOpenAIContext>;
     applyVendorHints?: boolean;
     promptCompression?: boolean | OpenAIPromptCompressionOptions;
+    /** Retain eligible metered calls with an exact link to their UsageTap call. */
+    sampling?: MeteredOpenAISamplingOptions | true;
 }
 interface OpenAICompressionOptions extends OpenAIPromptCompressionOptions {
     /** UsageTap API key. Defaults to USAGETAP_API_KEY. */
     apiKey?: string;
     /** Reuse an existing client when custom transport or endpoints are configured. */
     usageTapClient?: UsageTapClient;
+    /** Also retain eligible calls after the provider responds. */
+    sampling?: SamplingOptions | true;
 }
+type OpenAISamplingOptions = Partial<SamplingOptions> & {
+    /** UsageTap API key. Defaults to USAGETAP_API_KEY. */
+    apiKey?: string;
+    /** Reuse an existing UsageTap client. */
+    usageTapClient?: UsageTapClient;
+    /** Provider label stored with each sample. */
+    provider?: string;
+};
+type MeteredOpenAISamplingOptions = Partial<SamplingOptions> & {
+    /** Provider label stored with each retained sample. */
+    provider?: string;
+};
+type SamplingWrappedOpenAI = OpenAI & {
+    unwrap: () => OpenAI;
+};
 type CompressionWrappedOpenAI = OpenAI & {
     /** Return the original, unwrapped OpenAI-compatible client. */
     unwrap: () => OpenAI;
@@ -124,8 +161,12 @@ type WithMeteringOpenAIOptions = BeginCallRequest & {
     apiKey?: string;
     /** Reuse an existing client when custom transport or endpoints are configured. */
     usageTapClient?: UsageTapClient;
+    /** Vendor label recorded at call_end. Defaults to `openai`. */
+    provider?: string;
     applyVendorHints?: boolean;
     promptCompression?: boolean | OpenAIPromptCompressionOptions;
+    /** Retain eligible calls using the same ID as the metered UsageTap call. */
+    sampling?: MeteredOpenAISamplingOptions | true;
 };
 type ChatCompletionsResource = OpenAI["chat"]["completions"];
 type ChatCompletionCreate = ChatCompletionsResource["create"];
@@ -208,6 +249,8 @@ type UsageTapStream<T> = AsyncIterable<T> & {
 };
 declare function toNextResponse<T>(stream: UsageTapStream<T>, options?: StreamToResponseOptions): Response;
 declare function pipeToResponse<T>(stream: UsageTapStream<T>, res: NodeResponseLike, options?: StreamToResponseOptions): Promise<void>;
+/** Retain locally-selected OpenAI-compatible calls without changing call sites. */
+declare function withSampling(client: OpenAI, options?: OpenAISamplingOptions): SamplingWrappedOpenAI;
 /**
  * Add hosted prompt compression to an OpenAI-compatible client without
  * changing any of its call sites. Works with OpenAI and OpenRouter clients.
@@ -231,4 +274,4 @@ interface NodeResponseLike {
     flush?(): void;
 }
 
-export { type CompressionWrappedOpenAI, type NodeResponseLike, type OpenAIAdapter, type OpenAIAdapterInit, type OpenAICompressionOptions, type OpenAIInvokeParams, type OpenAIInvokeResult, type OpenAIPromptCompressionFailure, type OpenAIPromptCompressionOptions, type OpenAIPromptCompressionRole, type OpenAIPromptCompressionRoleOptions, type OpenAIPromptCompressionRoleSetting, OpenAIPromptCompressionStats, type OpenAIPromptCompressionTurnStats, type OpenAIRequestContext, type OpenAIStreamCallResult, type OpenAIStreamParams, type OpenAIStreamResult, type StreamMode, type StreamOpenAIRouteOptions, type StreamToResponseOptions, type UsageTapStream, type WithMeteringOpenAIOptions, type WrapOpenAICallOptions, type WrapOpenAIContext, type WrapOpenAIOptions, type WrapOpenAIResponseCallOptions, type WrappedOpenAI, createOpenAIAdapter, pipeToResponse, streamOpenAIRoute, toNextResponse, withCompression, withMetering, wrapOpenAI };
+export { type CompressionWrappedOpenAI, type MeteredOpenAISamplingOptions, type NodeResponseLike, type OpenAIAdapter, type OpenAIAdapterInit, type OpenAICompressionOptions, type OpenAIInvokeParams, type OpenAIInvokeResult, type OpenAIPromptCompressionFailure, type OpenAIPromptCompressionOptions, type OpenAIPromptCompressionRole, type OpenAIPromptCompressionRoleOptions, type OpenAIPromptCompressionRoleSetting, OpenAIPromptCompressionStats, type OpenAIPromptCompressionTurnStats, type OpenAIRequestContext, type OpenAISamplingOptions, type OpenAIStreamCallResult, type OpenAIStreamParams, type OpenAIStreamResult, type SamplingWrappedOpenAI, type StreamMode, type StreamOpenAIRouteOptions, type StreamToResponseOptions, type UsageTapStream, type WithMeteringOpenAIOptions, type WrapOpenAICallOptions, type WrapOpenAIContext, type WrapOpenAIOptions, type WrapOpenAIResponseCallOptions, type WrappedOpenAI, createOpenAIAdapter, pipeToResponse, streamOpenAIRoute, toNextResponse, withCompression, withMetering, withSampling, wrapOpenAI };

package/dist/adapters/openai.d.ts

@@ -1,9 +1,11 @@
 import OpenAI from 'openai';
-import { B as BeginCallRequest, P as PromptCompressionProvider, U as UsageTapClient, W as WithUsageOptions, a as PromptCompressionTelemetry, V as VendorHints, b as UsageTapSuccessResponse, c as BeginCallResponseBody, E as EndCallRequest } from '../client-BD8O2J8Z.js';
+import { B as BeginCallRequest, P as PromptCompressionProvider, a as PromptCompressionMode, S as SamplingOptions, U as UsageTapClient, W as WithUsageOptions, b as PromptCompressionTelemetry, V as VendorHints, c as UsageTapSuccessResponse, d as BeginCallResponseBody, E as EndCallRequest } from '../client-CExQ8e1T.js';
 
 interface OpenAIAdapterInit {
     client: OpenAI;
     usageTap: UsageTapClient;
+    /** Vendor label recorded at call_end. Defaults to `openai`. */
+    provider?: string;
 }
 interface OpenAIRequestContext {
     hints?: VendorHints;
@@ -65,6 +67,14 @@ interface OpenAIPromptCompressionOptions {
     /** Minimum estimated tokens for each individual text segment. */
     minTokens?: number;
     failOpen?: boolean;
+    /** Select deterministic-only, automatic model gating, or a forced model pass. */
+    mode?: PromptCompressionMode;
+    /** Maximum compression latency the service should target. */
+    latencyBudgetMs?: number;
+    /** Remove empty user messages before compression. */
+    compactEmptyUserMessages?: boolean;
+    /** Deduplicate repeated text parts within user messages. */
+    compactDuplicateUserTextParts?: boolean;
     tokenCompanyModel?: string;
     aggressiveness?: number | Partial<Record<OpenAIPromptCompressionRole, number>>;
     /** @deprecated Use aggressiveness instead. */
@@ -74,7 +84,13 @@ interface OpenAIPromptCompressionOptions {
     /** @deprecated Use aggressiveness instead. */
     usageTapCompressionAggressiveness?: number | Partial<Record<OpenAIPromptCompressionRole, number>>;
 }
-interface OpenAIPromptCompressionTurnStats extends PromptCompressionTelemetry {
+interface PromptCompressionTelemetryWithDiagnostics extends PromptCompressionTelemetry {
+    originalCharacters: number;
+    compressedCharacters: number;
+    savedCharacters: number;
+    savingsRatio: number;
+}
+interface OpenAIPromptCompressionTurnStats extends PromptCompressionTelemetryWithDiagnostics {
     callId: string;
     operation: "chat.completions.create" | "chat.completions.stream" | "responses.create";
     messagesCompressed: number;
@@ -105,16 +121,37 @@ declare class OpenAIPromptCompressionStats {
     get savingsRatio(): number;
 }
 interface WrapOpenAIOptions {
+    /** Vendor label recorded at call_end. Defaults to `openai`. */
+    provider?: string;
     defaultContext?: Partial<WrapOpenAIContext>;
     applyVendorHints?: boolean;
     promptCompression?: boolean | OpenAIPromptCompressionOptions;
+    /** Retain eligible metered calls with an exact link to their UsageTap call. */
+    sampling?: MeteredOpenAISamplingOptions | true;
 }
 interface OpenAICompressionOptions extends OpenAIPromptCompressionOptions {
     /** UsageTap API key. Defaults to USAGETAP_API_KEY. */
     apiKey?: string;
     /** Reuse an existing client when custom transport or endpoints are configured. */
     usageTapClient?: UsageTapClient;
+    /** Also retain eligible calls after the provider responds. */
+    sampling?: SamplingOptions | true;
 }
+type OpenAISamplingOptions = Partial<SamplingOptions> & {
+    /** UsageTap API key. Defaults to USAGETAP_API_KEY. */
+    apiKey?: string;
+    /** Reuse an existing UsageTap client. */
+    usageTapClient?: UsageTapClient;
+    /** Provider label stored with each sample. */
+    provider?: string;
+};
+type MeteredOpenAISamplingOptions = Partial<SamplingOptions> & {
+    /** Provider label stored with each retained sample. */
+    provider?: string;
+};
+type SamplingWrappedOpenAI = OpenAI & {
+    unwrap: () => OpenAI;
+};
 type CompressionWrappedOpenAI = OpenAI & {
     /** Return the original, unwrapped OpenAI-compatible client. */
     unwrap: () => OpenAI;
@@ -124,8 +161,12 @@ type WithMeteringOpenAIOptions = BeginCallRequest & {
     apiKey?: string;
     /** Reuse an existing client when custom transport or endpoints are configured. */
     usageTapClient?: UsageTapClient;
+    /** Vendor label recorded at call_end. Defaults to `openai`. */
+    provider?: string;
     applyVendorHints?: boolean;
     promptCompression?: boolean | OpenAIPromptCompressionOptions;
+    /** Retain eligible calls using the same ID as the metered UsageTap call. */
+    sampling?: MeteredOpenAISamplingOptions | true;
 };
 type ChatCompletionsResource = OpenAI["chat"]["completions"];
 type ChatCompletionCreate = ChatCompletionsResource["create"];
@@ -208,6 +249,8 @@ type UsageTapStream<T> = AsyncIterable<T> & {
 };
 declare function toNextResponse<T>(stream: UsageTapStream<T>, options?: StreamToResponseOptions): Response;
 declare function pipeToResponse<T>(stream: UsageTapStream<T>, res: NodeResponseLike, options?: StreamToResponseOptions): Promise<void>;
+/** Retain locally-selected OpenAI-compatible calls without changing call sites. */
+declare function withSampling(client: OpenAI, options?: OpenAISamplingOptions): SamplingWrappedOpenAI;
 /**
  * Add hosted prompt compression to an OpenAI-compatible client without
  * changing any of its call sites. Works with OpenAI and OpenRouter clients.
@@ -231,4 +274,4 @@ interface NodeResponseLike {
     flush?(): void;
 }
 
-export { type CompressionWrappedOpenAI, type NodeResponseLike, type OpenAIAdapter, type OpenAIAdapterInit, type OpenAICompressionOptions, type OpenAIInvokeParams, type OpenAIInvokeResult, type OpenAIPromptCompressionFailure, type OpenAIPromptCompressionOptions, type OpenAIPromptCompressionRole, type OpenAIPromptCompressionRoleOptions, type OpenAIPromptCompressionRoleSetting, OpenAIPromptCompressionStats, type OpenAIPromptCompressionTurnStats, type OpenAIRequestContext, type OpenAIStreamCallResult, type OpenAIStreamParams, type OpenAIStreamResult, type StreamMode, type StreamOpenAIRouteOptions, type StreamToResponseOptions, type UsageTapStream, type WithMeteringOpenAIOptions, type WrapOpenAICallOptions, type WrapOpenAIContext, type WrapOpenAIOptions, type WrapOpenAIResponseCallOptions, type WrappedOpenAI, createOpenAIAdapter, pipeToResponse, streamOpenAIRoute, toNextResponse, withCompression, withMetering, wrapOpenAI };
+export { type CompressionWrappedOpenAI, type MeteredOpenAISamplingOptions, type NodeResponseLike, type OpenAIAdapter, type OpenAIAdapterInit, type OpenAICompressionOptions, type OpenAIInvokeParams, type OpenAIInvokeResult, type OpenAIPromptCompressionFailure, type OpenAIPromptCompressionOptions, type OpenAIPromptCompressionRole, type OpenAIPromptCompressionRoleOptions, type OpenAIPromptCompressionRoleSetting, OpenAIPromptCompressionStats, type OpenAIPromptCompressionTurnStats, type OpenAIRequestContext, type OpenAISamplingOptions, type OpenAIStreamCallResult, type OpenAIStreamParams, type OpenAIStreamResult, type SamplingWrappedOpenAI, type StreamMode, type StreamOpenAIRouteOptions, type StreamToResponseOptions, type UsageTapStream, type WithMeteringOpenAIOptions, type WrapOpenAICallOptions, type WrapOpenAIContext, type WrapOpenAIOptions, type WrapOpenAIResponseCallOptions, type WrappedOpenAI, createOpenAIAdapter, pipeToResponse, streamOpenAIRoute, toNextResponse, withCompression, withMetering, withSampling, wrapOpenAI };