@usagetap/sdk 1.3.1 → 1.4.0

package/dist/index.cjs

@@ -245,6 +245,11 @@ async function compressMessagesWithUsageTap(options) {
     aggressiveness,
     "UsageTap prompt message compression"
   );
+  if (options.latencyBudgetMs !== void 0 && (!Number.isFinite(options.latencyBudgetMs) || options.latencyBudgetMs < 0)) {
+    throw new Error(
+      "UsageTap prompt message compression latencyBudgetMs must be a non-negative number"
+    );
+  }
   const original = stableStringifyInput(options.input);
   const headers = {
     "content-type": "application/json"
@@ -259,7 +264,15 @@ async function compressMessagesWithUsageTap(options) {
       headers,
       body: JSON.stringify({
         ...cloneInputRecord(options.input),
-        compression_settings: { aggressiveness }
+        compression_settings: {
+          aggressiveness,
+          ...options.mode === void 0 ? {} : { mode: options.mode },
+          ...options.latencyBudgetMs === void 0 ? {} : { latency_budget_ms: options.latencyBudgetMs },
+          ...options.compactEmptyUserMessages === void 0 ? {} : { compact_empty_user_messages: options.compactEmptyUserMessages },
+          ...options.compactDuplicateUserTextParts === void 0 ? {} : {
+            compact_duplicate_user_text_parts: options.compactDuplicateUserTextParts
+          }
+        }
       }),
       signal: options.signal
     }
@@ -755,10 +768,358 @@ function scalarToToon(value) {
   return JSON.stringify(text);
 }
 
+// src/resources.ts
+var CANONICAL_MEDIA_TYPE = "application/vnd.usagetap.v1+json";
+var DEFAULT_GATEWAY_BASE_URL = "https://gateway.usagetap.com";
+function normalizedBaseUrl(value) {
+  return `${value.replace(/\/+$/, "")}/`;
+}
+function errorMessage(payload, status) {
+  if (payload && typeof payload === "object") {
+    const record = payload;
+    const error = record.error;
+    if (error && typeof error === "object") {
+      const message2 = error.message;
+      if (typeof message2 === "string" && message2) return message2;
+    }
+    const message = record.message;
+    if (typeof message === "string" && message) return message;
+  }
+  return `UsageTap request failed with HTTP ${status}`;
+}
+function errorCode(status) {
+  if (status === 401 || status === 403) return "USAGETAP_AUTH_ERROR";
+  if (status === 429) return "USAGETAP_RATE_LIMITED";
+  if (status >= 500) return "USAGETAP_SERVER_ERROR";
+  return "USAGETAP_BAD_REQUEST";
+}
+var ResourceTransport = class {
+  baseUrl;
+  apiKey;
+  fetchImpl;
+  defaultHeaders;
+  sdkVersion;
+  constructor(baseUrl, config) {
+    this.baseUrl = normalizedBaseUrl(baseUrl);
+    this.apiKey = config.apiKey;
+    this.fetchImpl = config.fetchImpl;
+    this.defaultHeaders = config.headers ?? {};
+    this.sdkVersion = config.sdkVersion;
+  }
+  async request(request) {
+    const body = request.body === void 0 ? void 0 : JSON.stringify(request.body);
+    const headers = {
+      ...this.defaultHeaders,
+      accept: request.response === "data" ? CANONICAL_MEDIA_TYPE : request.response === "ndjson" ? "application/x-ndjson" : "application/json",
+      authorization: `Bearer ${this.apiKey}`,
+      "x-usage-sdk": `js/${this.sdkVersion}`,
+      ...body ? { "content-type": "application/json" } : {},
+      ...request.options?.idempotencyKey ? { "idempotency-key": request.options.idempotencyKey } : {},
+      ...request.options?.headers
+    };
+    let response;
+    try {
+      response = await this.fetchImpl(
+        new URL(request.path.replace(/^\/+/, ""), this.baseUrl),
+        {
+          method: request.method,
+          headers,
+          body,
+          signal: request.options?.signal
+        }
+      );
+    } catch (error) {
+      throw new UsageTapError(
+        "USAGETAP_NETWORK_ERROR",
+        "Failed to reach UsageTap",
+        { retryable: true, cause: error }
+      );
+    }
+    const requestId = response.headers.get("x-request-id") ?? response.headers.get("x-usage-correlation-id") ?? void 0;
+    const text = await response.text();
+    let payload;
+    if (text && request.response !== "ndjson") {
+      try {
+        payload = JSON.parse(text);
+      } catch (error) {
+        throw new UsageTapError(
+          "USAGETAP_INVALID_RESPONSE",
+          "UsageTap returned invalid JSON",
+          { status: response.status, correlationId: requestId, cause: error }
+        );
+      }
+    }
+    if (!response.ok) {
+      throw new UsageTapError(
+        errorCode(response.status),
+        errorMessage(payload, response.status),
+        {
+          status: response.status,
+          retryable: response.status === 429 || response.status >= 500,
+          correlationId: requestId,
+          details: payload && typeof payload === "object" ? payload : void 0
+        }
+      );
+    }
+    if (request.response === "ndjson") {
+      if (!text.trim()) return [];
+      try {
+        return text.trim().split(/\r?\n/).filter(Boolean).map((line) => JSON.parse(line));
+      } catch (error) {
+        throw new UsageTapError(
+          "USAGETAP_INVALID_RESPONSE",
+          "UsageTap returned invalid NDJSON",
+          { status: response.status, correlationId: requestId, cause: error }
+        );
+      }
+    }
+    if (request.response === "data") {
+      if (!payload || typeof payload !== "object" || !("data" in payload)) {
+        throw new UsageTapError(
+          "USAGETAP_INVALID_RESPONSE",
+          "UsageTap response missing data",
+          { status: response.status, correlationId: requestId }
+        );
+      }
+      return payload.data;
+    }
+    if (payload === void 0) {
+      throw new UsageTapError(
+        "USAGETAP_INVALID_RESPONSE",
+        "UsageTap response was empty",
+        { status: response.status, correlationId: requestId }
+      );
+    }
+    return payload;
+  }
+};
+function resourceId(value, keys, label) {
+  const id = typeof value === "string" ? value : keys.map((key) => value[key]).find((candidate) => Boolean(candidate?.trim()));
+  if (!id?.trim()) {
+    throw new UsageTapError(
+      "USAGETAP_BAD_REQUEST",
+      `${label} requires a non-empty ID`
+    );
+  }
+  return id.trim();
+}
+function terminalSummary(status) {
+  return status === "COMPLETE" || status === "FAILED";
+}
+function terminalGatewayBatch(status) {
+  return ["completed", "failed", "expired", "cancelled"].includes(status);
+}
+function validateWaitOptions(options) {
+  const pollIntervalMs = options.pollIntervalMs ?? 1500;
+  const timeoutMs = options.timeoutMs ?? 3 * 6e4;
+  if (!Number.isFinite(pollIntervalMs) || pollIntervalMs < 0) {
+    throw new UsageTapError(
+      "USAGETAP_BAD_REQUEST",
+      "pollIntervalMs must be a non-negative number"
+    );
+  }
+  if (!Number.isFinite(timeoutMs) || timeoutMs < 1) {
+    throw new UsageTapError(
+      "USAGETAP_BAD_REQUEST",
+      "timeoutMs must be a positive number"
+    );
+  }
+  return { pollIntervalMs, timeoutMs };
+}
+var SummarizationResource = class {
+  summaries;
+  batches;
+  profiles;
+  measurements;
+  transport;
+  constructor(config) {
+    this.transport = new ResourceTransport(config.apiBaseUrl, config);
+    this.summaries = {
+      create: (params, options) => this.transport.request({
+        method: "POST",
+        path: "/v1/compression/summaries",
+        body: params,
+        options,
+        response: "data"
+      }),
+      retrieve: (jobId, options) => this.transport.request({
+        method: "GET",
+        path: `/v1/compression/jobs/${encodeURIComponent(
+          resourceId(jobId, ["jobId"], "summaries.retrieve")
+        )}`,
+        options,
+        response: "data"
+      }),
+      wait: (job, options) => this.waitForSummary(job, options)
+    };
+    this.batches = {
+      create: (params, options) => this.transport.request({
+        method: "POST",
+        path: "/v1/compression/batches",
+        body: params,
+        options,
+        response: "data"
+      }),
+      retrieve: (batchId, options) => this.transport.request({
+        method: "GET",
+        path: `/v1/compression/batches/${encodeURIComponent(
+          resourceId(batchId, ["batchId"], "summarization.batches.retrieve")
+        )}`,
+        options,
+        response: "data"
+      }),
+      wait: (batch, options) => this.waitForBatch(batch, options)
+    };
+    this.profiles = {
+      retrieve: (profile, options) => this.transport.request({
+        method: "GET",
+        path: `/v1/compression/profiles/${encodeURIComponent(
+          resourceId(profile, [], "summarization.profiles.retrieve")
+        )}`,
+        options,
+        response: "data"
+      })
+    };
+    this.measurements = {
+      create: (params, options) => this.transport.request({
+        method: "POST",
+        path: "/v1/compression/measurements",
+        body: params,
+        options,
+        response: "data"
+      })
+    };
+  }
+  async waitForSummary(value, options = {}) {
+    const { pollIntervalMs, timeoutMs } = validateWaitOptions(options);
+    const deadline = Date.now() + timeoutMs;
+    let job = typeof value === "string" ? await this.summaries.retrieve(value, options) : value;
+    while (!terminalSummary(job.status)) {
+      if (Date.now() >= deadline) {
+        throw new UsageTapError(
+          "USAGETAP_RETRY_EXHAUSTED",
+          `Summarization job ${job.jobId} did not finish before timeout`,
+          { retryable: true }
+        );
+      }
+      await sleep(pollIntervalMs, options.signal);
+      job = await this.summaries.retrieve(job.jobId, options);
+    }
+    return job;
+  }
+  async waitForBatch(value, options = {}) {
+    const { pollIntervalMs, timeoutMs } = validateWaitOptions(options);
+    const deadline = Date.now() + timeoutMs;
+    let batch = typeof value === "string" ? await this.batches.retrieve(value, options) : value;
+    while (!terminalSummary(batch.status)) {
+      if (Date.now() >= deadline) {
+        throw new UsageTapError(
+          "USAGETAP_RETRY_EXHAUSTED",
+          `Summarization batch ${batch.batchId} did not finish before timeout`,
+          { retryable: true }
+        );
+      }
+      await sleep(pollIntervalMs, options.signal);
+      batch = await this.batches.retrieve(batch.batchId, options);
+    }
+    return batch;
+  }
+};
+var GatewayResource = class {
+  chat;
+  models;
+  batches;
+  transport;
+  idempotencyGenerator;
+  constructor(config) {
+    this.transport = new ResourceTransport(
+      config.gatewayBaseUrl ?? DEFAULT_GATEWAY_BASE_URL,
+      config
+    );
+    this.idempotencyGenerator = config.idempotencyGenerator ?? createIdempotencyKey;
+    this.chat = {
+      completions: {
+        create: (params, options) => this.transport.request({
+          method: "POST",
+          path: "/v1/chat/completions",
+          body: params,
+          options,
+          response: "json"
+        })
+      }
+    };
+    this.models = {
+      list: (options) => this.transport.request({
+        method: "GET",
+        path: "/v1/models",
+        options,
+        response: "json"
+      })
+    };
+    this.batches = {
+      create: (params, options = {}) => this.transport.request({
+        method: "POST",
+        path: "/v1/batches",
+        body: params,
+        options: {
+          ...options,
+          idempotencyKey: options.idempotencyKey ?? this.idempotencyGenerator()
+        },
+        response: "json"
+      }),
+      retrieve: (batchId, options) => this.transport.request({
+        method: "GET",
+        path: `/v1/batches/${encodeURIComponent(
+          resourceId(batchId, ["id"], "gateway.batches.retrieve")
+        )}`,
+        options,
+        response: "json"
+      }),
+      wait: (batch, options) => this.waitForBatch(batch, options),
+      cancel: (batchId, options) => this.transport.request({
+        method: "POST",
+        path: `/v1/batches/${encodeURIComponent(
+          resourceId(batchId, ["id"], "gateway.batches.cancel")
+        )}/cancel`,
+        options,
+        response: "json"
+      }),
+      results: (batchId, options) => this.transport.request({
+        method: "GET",
+        path: `/v1/batches/${encodeURIComponent(
+          resourceId(batchId, ["id"], "gateway.batches.results")
+        )}/results`,
+        options,
+        response: "ndjson"
+      })
+    };
+  }
+  async waitForBatch(value, options = {}) {
+    const { pollIntervalMs, timeoutMs } = validateWaitOptions(options);
+    const deadline = Date.now() + timeoutMs;
+    let batch = typeof value === "string" ? await this.batches.retrieve(value, options) : value;
+    while (!terminalGatewayBatch(batch.status)) {
+      if (Date.now() >= deadline) {
+        throw new UsageTapError(
+          "USAGETAP_RETRY_EXHAUSTED",
+          `Gateway batch ${batch.id} did not finish before timeout`,
+          { retryable: true }
+        );
+      }
+      await sleep(pollIntervalMs, options.signal);
+      batch = await this.batches.retrieve(batch.id, options);
+    }
+    return batch;
+  }
+};
+
 // src/client.ts
 var CALL_BEGIN_PATH = "call_begin";
 var CALL_END_PATH = "call_end";
 var COMPRESS_PROMPT_PATH = "compress_prompt";
+var SAMPLES_PATH = "samples";
+var SAMPLING_SETTINGS_PATH = "sampling/settings";
+var SAMPLING_DECIDE_PATH = "sampling/decide";
 var CHECK_USAGE_PATH = "customers/{customerId}/usage";
 var CREATE_CUSTOMER_PATH = "customers";
 var CHANGE_PLAN_PATH = "customers/{customerId}/change_plan";
@@ -769,11 +1130,16 @@ var CORRELATION_HEADER = "x-usage-correlation-id";
 var IDEMPOTENCY_HEADER = "idempotency-key";
 var SDK_HEADER = "x-usage-sdk";
 var USER_AGENT = "UsageTapClient";
-var CANONICAL_MEDIA_TYPE = "application/vnd.usagetap.v1+json";
+var CANONICAL_MEDIA_TYPE2 = "application/vnd.usagetap.v1+json";
 var DEFAULT_BASE_URL = "https://api.usagetap.com";
-var SDK_VERSION = "1.3.1" ;
+var DEFAULT_RUN_INACTIVITY_MS = 60 * 60 * 1e3;
+var SDK_VERSION = "1.4.0" ;
 var HAS_WINDOW = typeof globalThis !== "undefined" && typeof globalThis.window !== "undefined";
 var UsageTapClient = class {
+  /** OpenAI-compatible chat, model, and native batch operations. */
+  gateway;
+  /** Published-profile context summarization operations. */
+  summarization;
   apiKey;
   baseUrl;
   fetchImpl;
@@ -798,6 +1164,11 @@ var UsageTapClient = class {
   usageTapCompressionMessagesEndpoint;
   usageTapCompressionModel;
   usageTapCompressionAggressiveness;
+  sampling;
+  samplingSettingsCacheMs;
+  circuitBreaker;
+  circuitBreakerRuns = /* @__PURE__ */ new Map();
+  samplingSettingsCache;
   constructor(options = {}) {
     const apiKey = options.apiKey?.trim() || readEnvironmentVariable("USAGETAP_API_KEY");
     const baseUrl = options.baseUrl?.trim() || readEnvironmentVariable("USAGETAP_BASE_URL") || DEFAULT_BASE_URL;
@@ -820,10 +1191,21 @@ var UsageTapClient = class {
         "A global fetch implementation was not found. Pass fetchImpl in UsageTapClientOptions."
       );
     }
-    const normalizedBaseUrl = normalizeBaseUrl(baseUrl);
-    this.baseUrl = new URL(normalizedBaseUrl);
+    const normalizedBaseUrl2 = normalizeBaseUrl(baseUrl);
+    this.baseUrl = new URL(normalizedBaseUrl2);
     this.apiKey = apiKey;
     this.fetchImpl = wrapFetchImplementation(fetchCandidate, !options.fetchImpl);
+    const resourceConfig = {
+      apiKey,
+      apiBaseUrl: normalizedBaseUrl2,
+      gatewayBaseUrl: options.gatewayBaseUrl?.trim() || readEnvironmentVariable("USAGETAP_GATEWAY_URL"),
+      fetchImpl: this.fetchImpl,
+      headers: options.headers,
+      sdkVersion: SDK_VERSION,
+      idempotencyGenerator: options.idempotencyGenerator
+    };
+    this.gateway = new GatewayResource(resourceConfig);
+    this.summarization = new SummarizationResource(resourceConfig);
     this.defaultFeature = options.defaultFeature;
     this.defaultTags = options.defaultTags?.length ? dedupeStrings(options.defaultTags) : void 0;
     this.defaultHeaders = options.headers ? normalizeHeaderDictionary(options.headers) : {};
@@ -845,11 +1227,122 @@ var UsageTapClient = class {
     this.usageTapCompressionMessagesEndpoint = options.usageTapCompressionMessagesEndpoint;
     this.usageTapCompressionModel = options.usageTapCompressionModel;
     this.usageTapCompressionAggressiveness = options.usageTapCompressionAggressiveness;
+    this.sampling = options.sampling;
+    this.samplingSettingsCacheMs = Number.isFinite(options.samplingSettingsCacheMs) ? Math.max(0, Number(options.samplingSettingsCacheMs)) : 5 * 60 * 1e3;
+    if (options.circuitBreaker) {
+      const maxCallsPerRun = options.circuitBreaker.maxCallsPerRun;
+      if (!Number.isInteger(maxCallsPerRun) || maxCallsPerRun < 1) {
+        throw new UsageTapError(
+          "USAGETAP_BAD_REQUEST",
+          "circuitBreaker.maxCallsPerRun must be a positive integer"
+        );
+      }
+      const runInactivityMs = options.circuitBreaker.runInactivityMs ?? DEFAULT_RUN_INACTIVITY_MS;
+      if (!Number.isFinite(runInactivityMs) || runInactivityMs < 1) {
+        throw new UsageTapError(
+          "USAGETAP_BAD_REQUEST",
+          "circuitBreaker.runInactivityMs must be a positive number"
+        );
+      }
+      this.circuitBreaker = {
+        maxCallsPerRun,
+        runInactivityMs
+      };
+    }
+  }
+  shouldSample(request, policy = this.sampling || void 0) {
+    if (!policy) return false;
+    const rate = Math.min(1, Math.max(0, Number(policy.rate) || 0));
+    if (rate <= 0) return false;
+    const customerId = request.customerId?.trim();
+    if (customerId && policy.customers?.exclude?.includes(customerId)) return false;
+    const feature = request.feature?.trim();
+    if (feature && policy.features?.exclude?.includes(feature)) return false;
+    const included = policy.features?.include?.filter(Boolean) ?? [];
+    if (included.length > 0 && (!feature || !included.includes(feature))) return false;
+    const minimum = Math.max(0, Math.round(policy.minInputTokens ?? 0));
+    if (minimum > 0 && estimatePromptTokens(request.input) < minimum) return false;
+    return (policy.random ?? Math.random)() < rate;
+  }
+  async getSamplingSettings(options = {}) {
+    const now = Date.now();
+    if (!options.forceRefresh && this.samplingSettingsCache && this.samplingSettingsCache.expiresAtMs > now) {
+      return {
+        result: { status: "ACCEPTED", code: "SAMPLING_SETTINGS_CACHED" },
+        data: this.samplingSettingsCache.settings,
+        correlationId: options.correlationId ?? "local-cache"
+      };
+    }
+    const response = await this.requestGet(
+      SAMPLING_SETTINGS_PATH,
+      {
+        signal: options.signal,
+        headers: options.headers,
+        retries: options.retries,
+        correlationId: options.correlationId
+      }
+    );
+    const serverCacheMs = Math.max(0, Number(response.data.cacheSeconds) || 0) * 1e3;
+    const cacheMs = Math.min(this.samplingSettingsCacheMs, serverCacheMs);
+    this.samplingSettingsCache = {
+      settings: response.data,
+      expiresAtMs: now + cacheMs
+    };
+    return response;
+  }
+  async shouldSampleAsync(request, policy) {
+    if (policy) return this.shouldSample(request, policy);
+    if (this.sampling === false) return false;
+    if (this.sampling) return this.shouldSample(request, this.sampling);
+    try {
+      const settings = await this.getSamplingSettings();
+      return this.shouldSample(request, settings.data);
+    } catch {
+      return false;
+    }
+  }
+  async decideSample(request, options = {}) {
+    const hasTokens = Number.isFinite(request.inputTokens) && Number(request.inputTokens) >= 0;
+    const hasCharacters = Number.isFinite(request.inputCharacters) && Number(request.inputCharacters) >= 0;
+    if (!hasTokens && !hasCharacters) {
+      throw new UsageTapError(
+        "USAGETAP_BAD_REQUEST",
+        "decideSample requires inputTokens or inputCharacters"
+      );
+    }
+    return this.request(
+      SAMPLING_DECIDE_PATH,
+      request,
+      options
+    );
+  }
+  async captureSample(request, options = {}) {
+    if (!request || request.input === void 0) {
+      throw new UsageTapError(
+        "USAGETAP_BAD_REQUEST",
+        "captureSample requires input"
+      );
+    }
+    if (!request.provider?.trim()) {
+      throw new UsageTapError(
+        "USAGETAP_BAD_REQUEST",
+        "captureSample requires provider"
+      );
+    }
+    const sampleId = request.sampleId?.trim() || this.idempotencyGenerator();
+    return this.request(
+      SAMPLES_PATH,
+      { ...request, sampleId },
+      { ...options, idempotencyKey: sampleId }
+    );
   }
   async beginCall(request, options = {}) {
     const idempotencyKey = request.idempotencyKey ?? request.idempotency ?? (this.autoIdempotency ? this.idempotencyGenerator() : void 0);
+    this.reserveRunCall(request, idempotencyKey);
+    const apiRequest = { ...request };
+    delete apiRequest.runId;
     const payload = {
-      ...request,
+      ...apiRequest,
       feature: request.feature ?? this.defaultFeature,
       tags: this.mergeTags(request.tags)
     };
@@ -867,6 +1360,28 @@ var UsageTapClient = class {
     );
     return response;
   }
+  /**
+   * Inspect a configured run circuit breaker without consuming another call.
+   */
+  canRunContinue(request) {
+    const identity = this.resolveRunIdentity(request);
+    if (!identity || !this.circuitBreaker) {
+      throw new UsageTapError(
+        "USAGETAP_BAD_REQUEST",
+        "canRunContinue requires circuitBreaker configuration and a non-empty runId"
+      );
+    }
+    this.expireInactiveRuns();
+    const calls = this.circuitBreakerRuns.get(identity.key)?.calls ?? 0;
+    return this.createCircuitBreakerDecision(identity.customerId, identity.runId, calls);
+  }
+  /**
+   * Release local state after a workflow finishes. Returns true when state existed.
+   */
+  resetRun(request) {
+    const identity = this.resolveRunIdentity(request);
+    return identity ? this.circuitBreakerRuns.delete(identity.key) : false;
+  }
   async promptCompress(request, options = {}) {
     if (!request?.callId) {
       throw new UsageTapError(
@@ -964,6 +1479,10 @@ var UsageTapClient = class {
       usageTapCompressionMessagesEndpoint: this.usageTapCompressionMessagesEndpoint,
       aggressiveness: options.aggressiveness ?? this.aggressiveness,
       usageTapCompressionAggressiveness: options.usageTapCompressionAggressiveness ?? this.usageTapCompressionAggressiveness,
+      mode: options.mode,
+      latencyBudgetMs: options.latencyBudgetMs,
+      compactEmptyUserMessages: options.compactEmptyUserMessages,
+      compactDuplicateUserTextParts: options.compactDuplicateUserTextParts,
       fetchImpl: this.fetchImpl,
       signal: options.signal,
       failOpen: options.failOpen
@@ -1005,11 +1524,17 @@ var UsageTapClient = class {
       callId: request.callId,
       feature: feature ?? this.defaultFeature,
       tags: tags ?? this.defaultTags,
+      providerUsed: request.providerUsed,
       modelUsed: request.modelUsed,
+      reasoningEffort: request.reasoningEffort,
+      reasoningEffortSource: request.reasoningEffortSource,
+      reasoningMode: request.reasoningMode,
+      reasoningBudgetTokens: request.reasoningBudgetTokens,
       metrics: {
         inputTokens: request.inputTokens,
         responseTokens: request.responseTokens,
         cachedInputTokens: request.cachedInputTokens,
+        cacheWriteInputTokens: request.cacheWriteInputTokens,
         reasoningTokens: request.reasoningTokens,
         searches: request.searches,
         audioSeconds: request.audioSeconds,
@@ -1117,6 +1642,15 @@ var UsageTapClient = class {
       meterSlot: request.meterSlot,
       amount: request.amount
     };
+    if (request.customerUserId) {
+      payload.customerUserId = request.customerUserId;
+    }
+    if (request.customerUserName) {
+      payload.customerUserName = request.customerUserName;
+    }
+    if (request.customerUserEmail) {
+      payload.customerUserEmail = request.customerUserEmail;
+    }
     if (request.feature) {
       payload.feature = request.feature;
     }
@@ -1153,6 +1687,11 @@ var UsageTapClient = class {
     const beginPayload = idempotencyKey ? { ...beginRequest, idempotencyKey, idempotency: idempotencyKey } : { ...beginRequest };
     const beginResponse = await this.beginCall(beginPayload, options);
     let usage = {};
+    const pricingMode = beginResponse.data.pricingMode ?? beginRequest.pricingMode ?? (beginRequest.batch === true ? "batch" : beginRequest.batch === false ? "standard" : void 0);
+    if (pricingMode) {
+      usage.pricingMode = pricingMode;
+      usage.batch = pricingMode === "batch";
+    }
     const initialStripeCustomerId = typeof beginResponse.data.stripeCustomerId === "string" ? beginResponse.data.stripeCustomerId : typeof beginRequest.stripeCustomerId === "string" ? beginRequest.stripeCustomerId : void 0;
     if (initialStripeCustomerId) {
       usage = { ...usage, stripeCustomerId: initialStripeCustomerId };
@@ -1220,17 +1759,86 @@ var UsageTapClient = class {
   toPromptCompressionTelemetry(result) {
     return {
       provider: result.provider,
-      originalCharacters: result.originalCharacters,
-      compressedCharacters: result.compressedCharacters,
-      savedCharacters: result.savedCharacters,
       originalTokens: result.originalTokens,
       compressedTokens: result.compressedTokens,
       savedTokens: result.savedTokens,
       tokenSavingsRatio: result.tokenSavingsRatio,
-      savingsRatio: result.savingsRatio,
       techniques: result.techniques
     };
   }
+  reserveRunCall(request, idempotencyKey) {
+    const identity = this.resolveRunIdentity(request);
+    if (!identity || !this.circuitBreaker) return;
+    this.expireInactiveRuns();
+    const now = Date.now();
+    const state = this.circuitBreakerRuns.get(identity.key) ?? {
+      calls: 0,
+      reservationKeys: /* @__PURE__ */ new Set(),
+      lastSeenAtMs: now
+    };
+    const reservationKey = idempotencyKey ?? this.idempotencyGenerator();
+    state.lastSeenAtMs = now;
+    if (state.reservationKeys.has(reservationKey)) {
+      this.circuitBreakerRuns.set(identity.key, state);
+      return;
+    }
+    const decision = this.createCircuitBreakerDecision(
+      identity.customerId,
+      identity.runId,
+      state.calls
+    );
+    if (!decision.allowed) {
+      throw new UsageTapError(
+        "USAGETAP_CIRCUIT_OPEN",
+        `Run ${identity.runId} reached its ${decision.limit}-call circuit-breaker limit`,
+        {
+          details: {
+            reason: decision.reason,
+            customerId: identity.customerId,
+            runId: identity.runId,
+            calls: decision.calls,
+            limit: decision.limit,
+            remaining: decision.remaining
+          }
+        }
+      );
+    }
+    state.calls += 1;
+    state.reservationKeys.add(reservationKey);
+    this.circuitBreakerRuns.set(identity.key, state);
+  }
+  resolveRunIdentity(request) {
+    const customerId = request.customerId?.trim();
+    const runId = request.runId?.trim();
+    if (!customerId || !runId) return void 0;
+    return {
+      key: `${customerId}\0${runId}`,
+      customerId,
+      runId
+    };
+  }
+  createCircuitBreakerDecision(customerId, runId, calls) {
+    const limit = this.circuitBreaker?.maxCallsPerRun ?? 0;
+    const allowed = calls < limit;
+    return {
+      allowed,
+      ...allowed ? {} : { reason: "max_calls_per_run" },
+      customerId,
+      runId,
+      calls,
+      limit,
+      remaining: Math.max(0, limit - calls)
+    };
+  }
+  expireInactiveRuns() {
+    if (!this.circuitBreaker || this.circuitBreakerRuns.size === 0) return;
+    const expiredBefore = Date.now() - this.circuitBreaker.runInactivityMs;
+    for (const [key, state] of this.circuitBreakerRuns) {
+      if (state.lastSeenAtMs < expiredBefore) {
+        this.circuitBreakerRuns.delete(key);
+      }
+    }
+  }
   async request(path, payload, options) {
     const url = new URL(path, this.baseUrl).toString();
     const body = payload !== void 0 ? JSON.stringify(payload) : void 0;
@@ -1418,7 +2026,7 @@ var UsageTapClient = class {
       ...this.defaultHeaders,
       [SDK_HEADER]: `js/${SDK_VERSION}`,
       "content-type": "application/json",
-      accept: CANONICAL_MEDIA_TYPE
+      accept: CANONICAL_MEDIA_TYPE2
     };
     if (!HAS_WINDOW) {
       headers["user-agent"] = `${USER_AGENT}/${SDK_VERSION}`;
@@ -1606,7 +2214,8 @@ function wrapFetch(usageTap, options) {
     defaultContext,
     baseFetch = globalThis.fetch,
     autoIdempotency = true,
-    isOpenAIEndpoint = defaultIsOpenAIEndpoint
+    isOpenAIEndpoint = defaultIsOpenAIEndpoint,
+    provider = "openai"
   } = options;
   return async function wrappedFetch(input, init) {
     const url = typeof input === "string" ? input : input instanceof URL ? input.href : input.url;
@@ -1628,6 +2237,7 @@ function wrapFetch(usageTap, options) {
         const customerIdHeader = headers.get("x-usagetap-customer-id");
         const featureHeader = headers.get("x-usagetap-feature");
         const tagsHeader = headers.get("x-usagetap-tags");
+        const runIdHeader = headers.get("x-usagetap-run-id");
         if (customerIdHeader) {
           contextOverride.customerId = customerIdHeader;
         }
@@ -1640,6 +2250,9 @@ function wrapFetch(usageTap, options) {
             contextOverride.tags = tags;
           }
         }
+        if (runIdHeader) {
+          contextOverride.runId = runIdHeader;
+        }
       }
     } catch {
       return baseFetch(input, init);
@@ -1655,10 +2268,25 @@ function wrapFetch(usageTap, options) {
       callState = {
         callId: beginResponse.data.callId,
         correlationId: beginResponse.correlationId,
-        usage: {},
+        usage: {
+          ...executionMetadataFromOpenAIRequest(body, provider),
+          ...beginResponse.data.pricingMode ? {
+            pricingMode: beginResponse.data.pricingMode,
+            batch: beginResponse.data.pricingMode === "batch"
+          } : context.pricingMode ? {
+            pricingMode: context.pricingMode,
+            batch: context.pricingMode === "batch"
+          } : typeof context.batch === "boolean" ? {
+            batch: context.batch,
+            pricingMode: context.batch ? "batch" : "standard"
+          } : {}
+        },
         finalized: false
       };
     } catch (error) {
+      if (isUsageTapError(error) && error.code === "USAGETAP_CIRCUIT_OPEN") {
+        throw error;
+      }
       console.error("[wrapFetch] Failed to begin call:", error);
       return baseFetch(input, init);
     }
@@ -1699,6 +2327,9 @@ async function wrapNonStreamingResponse(response, callState, usageTap, requestBo
       if (isJsonRecord(usageBlock)) {
         const usageRecord = usageBlock;
         const promptDetails = isJsonRecord(usageRecord.prompt_tokens_details) ? usageRecord.prompt_tokens_details : void 0;
+        const completionDetails = isJsonRecord(
+          usageRecord.completion_tokens_details ?? usageRecord.output_tokens_details
+        ) ? usageRecord.completion_tokens_details ?? usageRecord.output_tokens_details : void 0;
         const promptTokens = readNumber(usageRecord.prompt_tokens ?? usageRecord.input_tokens);
         if (promptTokens !== void 0) {
           usage.inputTokens = promptTokens;
@@ -1713,7 +2344,9 @@ async function wrapNonStreamingResponse(response, callState, usageTap, requestBo
         if (cachedTokens !== void 0) {
           usage.cachedInputTokens = cachedTokens;
         }
-        const reasoningTokens = readNumber(usageRecord.reasoning_tokens);
+        const reasoningTokens = readNumber(
+          usageRecord.reasoning_tokens ?? completionDetails?.reasoning_tokens
+        );
         if (reasoningTokens !== void 0) {
           usage.reasoningTokens = reasoningTokens;
         }
@@ -1722,6 +2355,14 @@ async function wrapNonStreamingResponse(response, callState, usageTap, requestBo
       if (modelFromResponse) {
         usage.modelUsed = modelFromResponse;
       }
+      const reasoning = isJsonRecord(parsed.reasoning) ? parsed.reasoning : void 0;
+      const responseEffort = normalizeReasoningEffort(reasoning?.effort);
+      if (responseEffort) {
+        usage.reasoningEffort = responseEffort;
+        usage.reasoningEffortSource = "provider_response";
+      }
+      const responseMode = readString(reasoning?.type ?? reasoning?.mode);
+      if (responseMode) usage.reasoningMode = responseMode;
     }
     if (!usage.modelUsed && requestBody) {
       const requestModel = readString(requestBody.model);
@@ -1769,6 +2410,9 @@ function wrapStreamingResponse(response, callState, usageTap) {
             if (isJsonRecord(usageBlock)) {
               const usageRecord = usageBlock;
               const promptDetails = isJsonRecord(usageRecord.prompt_tokens_details) ? usageRecord.prompt_tokens_details : void 0;
+              const completionDetails = isJsonRecord(
+                usageRecord.completion_tokens_details ?? usageRecord.output_tokens_details
+              ) ? usageRecord.completion_tokens_details ?? usageRecord.output_tokens_details : void 0;
               const promptTokens = readNumber(usageRecord.prompt_tokens ?? usageRecord.input_tokens);
               if (promptTokens !== void 0) {
                 accumulatedUsage.inputTokens = promptTokens;
@@ -1783,7 +2427,9 @@ function wrapStreamingResponse(response, callState, usageTap) {
               if (cachedTokens !== void 0) {
                 accumulatedUsage.cachedInputTokens = cachedTokens;
               }
-              const reasoningTokens = readNumber(usageRecord.reasoning_tokens);
+              const reasoningTokens = readNumber(
+                usageRecord.reasoning_tokens ?? completionDetails?.reasoning_tokens
+              );
               if (reasoningTokens !== void 0) {
                 accumulatedUsage.reasoningTokens = reasoningTokens;
               }
@@ -1808,6 +2454,25 @@ function wrapStreamingResponse(response, callState, usageTap) {
     headers: response.headers
   });
 }
+function normalizeReasoningEffort(value) {
+  return value === "none" || value === "minimal" || value === "low" || value === "medium" || value === "high" || value === "xhigh" || value === "max" ? value : void 0;
+}
+function executionMetadataFromOpenAIRequest(body, provider) {
+  const reasoning = isJsonRecord(body?.reasoning) ? body.reasoning : void 0;
+  const effort = normalizeReasoningEffort(
+    body?.reasoning_effort ?? reasoning?.effort ?? body?.thinking_level
+  );
+  const mode = readString(reasoning?.type ?? reasoning?.mode);
+  const budget = readNumber(
+    reasoning?.budget_tokens ?? body?.thinking_budget ?? body?.thinking_budget_tokens
+  );
+  return {
+    providerUsed: provider,
+    ...effort ? { reasoningEffort: effort, reasoningEffortSource: "provider_request" } : {},
+    ...mode ? { reasoningMode: mode } : {},
+    ...budget !== void 0 ? { reasoningBudgetTokens: budget } : {}
+  };
+}
 async function finalizeCall(callState, usageTap, error, usage) {
   if (callState.finalized) return;
   callState.finalized = true;