@broberg/ai-sdk 0.58.1 → 0.59.1

package/dist/index.d.ts

@@ -52,13 +52,33 @@ interface SearchCache {
     get(key: string): SearchResult | undefined;
     set(key: string, value: SearchResult, ttlMs: number): void;
 }
+/** Running spend per key (tenant + day). Async allowed, so it can live in a database.
+ *  `reserve` must CHECK AND ADD in one atomic step (e.g. `UPDATE … SET spent = spent + ?
+ *  WHERE spent + ? <= cap`) and say whether it fitted — a separate read-then-write lets
+ *  parallel calls all pass a cap only one of them fits under. `release` gives a
+ *  reservation back when the call fails. */
+interface SearchCapStore {
+    reserve(key: string, usd: number, capUsd: number): boolean | Promise<boolean>;
+    release(key: string, usd: number): void | Promise<void>;
+    getSpent(key: string): number | Promise<number>;
+}
 interface SearchOptions {
     /** false = no cache. Default: one in-process memory cache shared by all calls. */
     cache?: SearchCache | false;
     /** Where each call (and each failed attempt) is recorded. ai.search() passes the client's sink. */
     costSink?: CostSink;
-    /** Explicit keys (BYOK); missing fields fall back to env. */
+    /** Explicit keys (BYOK); missing fields fall back to env — unless `byok` is set. */
     credentials?: SearchCredentials;
+    /** F078.3 — customer-key mode: keys come ONLY from `credentials`, env is never read,
+     *  so a tenant whose key is missing cannot run on the fleet's account. */
+    byok?: boolean;
+    /** F078.3 — hard ceiling in USD per tenant (`labels.tenantId`) per calendar day in
+     *  Europe/Copenhagen. A call that would cross it throws before any request. */
+    dailyCapUsd?: number;
+    /** Where the day's spend is kept. Default: in-process memory shared by all calls. */
+    capStore?: SearchCapStore;
+    /** Clock for the day boundary (tests). */
+    now?: () => Date;
     /** Injectable fetch (tests, proxies) — same contract as the model adapters (F073). */
     fetch?: typeof fetch;
 }
@@ -68,6 +88,15 @@ declare class SearchKeyMissingError extends Error {
     readonly envVar: string;
     constructor(provider: SearchProviderId, envVar: string);
 }
+/** Thrown before any request when the tenant's daily ceiling would be crossed. */
+declare class SearchBudgetExceededError extends Error {
+    readonly tenant: string;
+    readonly day: string;
+    readonly capUsd: number;
+    readonly spentUsd: number;
+    readonly requestedUsd: number;
+    constructor(tenant: string, day: string, capUsd: number, spentUsd: number, requestedUsd: number);
+}
 
 interface MockupInput {
     description: string;
@@ -2301,8 +2330,8 @@ declare const falStubAdapter: ProviderAdapter;
  *  wires the live adapters. */
 declare const stubProviders: Record<string, ProviderAdapter>;
 
-declare const VERSION: "0.58.1";
-declare const SDK_TAG: "@broberg/ai-sdk@0.58.1";
+declare const VERSION: "0.59.1";
+declare const SDK_TAG: "@broberg/ai-sdk@0.59.1";
 
 /** Built-in defaults. Every entry is overridable via AiConfig.defaults or a
  *  per-call override.
@@ -2849,4 +2878,4 @@ declare function memorySearchCache(max?: number): SearchCache;
 
 declare function search(req: SearchRequest, opts?: SearchOptions): Promise<SearchResult>;
 
-export { AZURE_DANISH_VOICES, AZURE_DANISH_VOICE_LIST, type AiClient, type AiConfig, type AzureVoiceInfo, BatchJob, BatchRequestItem, BatchResultItem, type BflAdapterConfig, type BflCredits, BudgetConfig, BudgetExceededError, BudgetGuard, BudgetStore, CONFIG_DERIVED_HOSTS, Capability, type ChatInput, ChatResult, ChatStreamEvent, type CheckVoiceOptions, type ClassifyInput, type ClassifyResult, type Contracts, type CostQuery, CostSink, type CostSummary, type CostSummaryQuery, type CostTimeseriesQuery, DEFAULT_BASE_URLS, DEFAULT_SEARCH_ROUTES, DEFAULT_TIER_MAP, type DesignInput, type DesignResult, type DiscordSinkConfig, ELEVENLABS_DANISH_VOICES, type EmbeddingInput, EmbeddingResult, type ExtractInput, type ExtractResult, type FalAdapterConfig, type FixedHostProvider, type HttpResponse, type ImageInput, ImageResult, type JudgeInput, JudgeResult, LOCALLY_PINNED_HOSTS, type MockupInput, type MockupResult, type ModerationInput, ModerationResult, type OcrInput, OcrResult, type OpenAICompatibleConfig, type PodcastInput, PodcastResult, type PriceDriftResult, type PriceDriftRow, type PricingEntry, ProviderAdapter, type RefreshOptions, type RefreshResult, Region, type RerankInput, type RerankResult, type RouteForecast, SDK_TAG, SEARCH_CACHE_TTL_MS, SEARCH_PRICE_USD, type SearchCache, type SearchCredentials, type SearchItem, SearchKeyMissingError, type SearchOptions, type SearchProviderId, type SearchPurpose, type SearchRequest, type SearchResult, type SearchRoute, type SqliteBudgetStoreConfig, type SqliteSinkConfig, StreamHttpError, type SubprocessResponse, Tier, TierSpec, Tool, ToolCall, type TrainStyleInput, TrainStyleResult, type TranscribeInput, TranscribeResult, type TranslateInput, TranslateResult, Transport, type TransportRequest, type TransportResponse, type TtsInput, type UpmetricsCostClientConfig, UpmetricsCostError, type UpmetricsCostRow, type UpmetricsCostSummary, type UpmetricsCostTimeseries, type UpmetricsSink, type UpmetricsSinkConfig, type UpmetricsSinkStats, Usage, VERSION, type VideoInput, type VisionInput, type VoiceInfo, type VoiceProvider, type VoiceResolveResult, type VoiceStatus, VoiceUnavailableError, ZDR_PROVIDERS, aiConfigSchema, anthropicAdapter, anthropicApiAdapter, anthropicSubprocessAdapter, azureAdapter, bflAdapter, bflCredits, byokAdapter, chatInputSchema, checkPriceDrift, checkVoice, computeCost, createAI, deepinfraAdapter, deeplAdapter, deepseekAdapter, defaultBaseUrl, defaultProviders, discordSink, elevenlabsAdapter, embeddingInputSchema, falAdapter, falStubAdapter, freshUsage, fromProviderToolCall, geminiAdapter, getCostSummary, getPrice, httpTransport, imageInputSchema, isByokKeyed, listAzureDanishVoices, listVoices, makeContracts, makeOpenAICompatibleAdapter, matchLabel, memorySearchCache, messageSchema, mistralAdapter, mistralStubAdapter, multiSink, noopSink, openaiAdapter, openaiStubAdapter, openrouterAdapter, parseClaudeCliJson, parseJsonLoose, refreshAvailability, requestyAdapter, resetRefreshState, resetRegistry, resolveAzureVoice, resolveTier, resolveVoice, routeSearch, search, sqliteBudgetStore, sqliteSink, streamTransport, stubProviders, subprocessTransport, tierSpecSchema, toProviderTools, toolSchema, translateInputSchema, typesafeAdapter, upmetricsCostClient, upmetricsSink, usdFromMicro, vertexAdapter, visionInputSchema, wouldProviderRouteTo, wouldRouteTo };
+export { AZURE_DANISH_VOICES, AZURE_DANISH_VOICE_LIST, type AiClient, type AiConfig, type AzureVoiceInfo, BatchJob, BatchRequestItem, BatchResultItem, type BflAdapterConfig, type BflCredits, BudgetConfig, BudgetExceededError, BudgetGuard, BudgetStore, CONFIG_DERIVED_HOSTS, Capability, type ChatInput, ChatResult, ChatStreamEvent, type CheckVoiceOptions, type ClassifyInput, type ClassifyResult, type Contracts, type CostQuery, CostSink, type CostSummary, type CostSummaryQuery, type CostTimeseriesQuery, DEFAULT_BASE_URLS, DEFAULT_SEARCH_ROUTES, DEFAULT_TIER_MAP, type DesignInput, type DesignResult, type DiscordSinkConfig, ELEVENLABS_DANISH_VOICES, type EmbeddingInput, EmbeddingResult, type ExtractInput, type ExtractResult, type FalAdapterConfig, type FixedHostProvider, type HttpResponse, type ImageInput, ImageResult, type JudgeInput, JudgeResult, LOCALLY_PINNED_HOSTS, type MockupInput, type MockupResult, type ModerationInput, ModerationResult, type OcrInput, OcrResult, type OpenAICompatibleConfig, type PodcastInput, PodcastResult, type PriceDriftResult, type PriceDriftRow, type PricingEntry, ProviderAdapter, type RefreshOptions, type RefreshResult, Region, type RerankInput, type RerankResult, type RouteForecast, SDK_TAG, SEARCH_CACHE_TTL_MS, SEARCH_PRICE_USD, SearchBudgetExceededError, type SearchCache, type SearchCapStore, type SearchCredentials, type SearchItem, SearchKeyMissingError, type SearchOptions, type SearchProviderId, type SearchPurpose, type SearchRequest, type SearchResult, type SearchRoute, type SqliteBudgetStoreConfig, type SqliteSinkConfig, StreamHttpError, type SubprocessResponse, Tier, TierSpec, Tool, ToolCall, type TrainStyleInput, TrainStyleResult, type TranscribeInput, TranscribeResult, type TranslateInput, TranslateResult, Transport, type TransportRequest, type TransportResponse, type TtsInput, type UpmetricsCostClientConfig, UpmetricsCostError, type UpmetricsCostRow, type UpmetricsCostSummary, type UpmetricsCostTimeseries, type UpmetricsSink, type UpmetricsSinkConfig, type UpmetricsSinkStats, Usage, VERSION, type VideoInput, type VisionInput, type VoiceInfo, type VoiceProvider, type VoiceResolveResult, type VoiceStatus, VoiceUnavailableError, ZDR_PROVIDERS, aiConfigSchema, anthropicAdapter, anthropicApiAdapter, anthropicSubprocessAdapter, azureAdapter, bflAdapter, bflCredits, byokAdapter, chatInputSchema, checkPriceDrift, checkVoice, computeCost, createAI, deepinfraAdapter, deeplAdapter, deepseekAdapter, defaultBaseUrl, defaultProviders, discordSink, elevenlabsAdapter, embeddingInputSchema, falAdapter, falStubAdapter, freshUsage, fromProviderToolCall, geminiAdapter, getCostSummary, getPrice, httpTransport, imageInputSchema, isByokKeyed, listAzureDanishVoices, listVoices, makeContracts, makeOpenAICompatibleAdapter, matchLabel, memorySearchCache, messageSchema, mistralAdapter, mistralStubAdapter, multiSink, noopSink, openaiAdapter, openaiStubAdapter, openrouterAdapter, parseClaudeCliJson, parseJsonLoose, refreshAvailability, requestyAdapter, resetRefreshState, resetRegistry, resolveAzureVoice, resolveTier, resolveVoice, routeSearch, search, sqliteBudgetStore, sqliteSink, streamTransport, stubProviders, subprocessTransport, tierSpecSchema, toProviderTools, toolSchema, translateInputSchema, typesafeAdapter, upmetricsCostClient, upmetricsSink, usdFromMicro, vertexAdapter, visionInputSchema, wouldProviderRouteTo, wouldRouteTo };

package/dist/index.js

@@ -171,20 +171,54 @@ var SearchKeyMissingError = class extends Error {
   provider;
   envVar;
 };
+var SearchBudgetExceededError = class extends Error {
+  constructor(tenant, day, capUsd, spentUsd, requestedUsd) {
+    super(`search: daily cap $${capUsd} for tenant "${tenant}" on ${day} reached ($${spentUsd.toFixed(5)} spent, this call $${requestedUsd})`);
+    this.tenant = tenant;
+    this.day = day;
+    this.capUsd = capUsd;
+    this.spentUsd = spentUsd;
+    this.requestedUsd = requestedUsd;
+    this.name = "SearchBudgetExceededError";
+  }
+  tenant;
+  day;
+  capUsd;
+  spentUsd;
+  requestedUsd;
+};
 
 // src/search/index.ts
 var sharedCache = memorySearchCache();
+var sharedCapStore = /* @__PURE__ */ (() => {
+  const m = /* @__PURE__ */ new Map();
+  return {
+    // Synchronous, so check-and-add cannot interleave with another call.
+    reserve: (k, usd, cap) => {
+      const s = m.get(k) ?? 0;
+      if (s + usd > cap) return false;
+      m.set(k, s + usd);
+      return true;
+    },
+    release: (k, usd) => {
+      m.set(k, (m.get(k) ?? 0) - usd);
+    },
+    getSpent: (k) => m.get(k) ?? 0
+  };
+})();
+var copenhagenDay = (d) => new Intl.DateTimeFormat("en-CA", { timeZone: "Europe/Copenhagen" }).format(d);
 async function callProvider(provider, req, opts) {
   const f = opts.fetch ?? fetch;
   const c = opts.credentials ?? {};
+  const env = opts.byok ? {} : process.env;
   if (provider === "brave") {
-    const key = c.braveApiKey ?? process.env.BRAVE_API_KEY;
+    const key = c.braveApiKey ?? env.BRAVE_API_KEY;
     if (!key) throw new SearchKeyMissingError(provider, "BRAVE_API_KEY");
     return braveSearch({ ...req, provider }, key, f);
   }
   if (provider.startsWith("cloudflare:")) {
-    const accountId = c.cloudflareAccountId ?? process.env.CLOUDFLARE_ACCOUNT_ID;
-    const apiToken = c.cloudflareApiToken ?? process.env.CLOUDFLARE_API_TOKEN;
+    const accountId = c.cloudflareAccountId ?? env.CLOUDFLARE_ACCOUNT_ID;
+    const apiToken = c.cloudflareApiToken ?? env.CLOUDFLARE_API_TOKEN;
     if (!accountId) throw new SearchKeyMissingError(provider, "CLOUDFLARE_ACCOUNT_ID");
     if (!apiToken) throw new SearchKeyMissingError(provider, "CLOUDFLARE_API_TOKEN");
     return cloudflareSearch({ ...req, provider }, { accountId, apiToken }, f);
@@ -217,11 +251,20 @@ async function search(req, opts = {}) {
   const cache = opts.cache === false ? void 0 : opts.cache ?? sharedCache;
   const ttl = SEARCH_CACHE_TTL_MS[req.purpose ?? "agent"];
   const sink = opts.costSink;
+  const capStore = opts.capStore ?? sharedCapStore;
+  const tenant = req.labels?.tenantId ?? "_";
+  const day = copenhagenDay((opts.now ?? (() => /* @__PURE__ */ new Date()))());
+  const capKey = JSON.stringify([tenant, day]);
   let lastErr;
   for (const provider of chain) {
     const key = searchCacheKey(provider, req);
     const hit = cache?.get(key);
     if (hit) return { items: hit.items, meta: { ...hit.meta, cached: true, costUsd: 0, latencyMs: 0 } };
+    const price = SEARCH_PRICE_USD[provider];
+    const cap = opts.dailyCapUsd ?? Infinity;
+    if (!await capStore.reserve(capKey, price, cap)) {
+      throw new SearchBudgetExceededError(tenant, day, cap, await capStore.getSpent(capKey), price);
+    }
     const t0 = performance.now();
     try {
       const out = await callProvider(provider, req, opts);
@@ -238,6 +281,7 @@ async function search(req, opts = {}) {
       return result;
     } catch (e) {
       lastErr = e;
+      await capStore.release(capKey, price);
       if (sink?.recordFailure) {
         try {
           await sink.recordFailure({
@@ -3785,8 +3829,8 @@ var aiConfigSchema = z.object({
 import { randomUUID } from "crypto";
 
 // src/version.ts
-var VERSION = "0.58.1";
-var SDK_TAG = "@broberg/ai-sdk@0.58.1";
+var VERSION = "0.59.1";
+var SDK_TAG = "@broberg/ai-sdk@0.59.1";
 
 // src/cost/sinks/upmetrics.ts
 function isRetryableStatus(status) {
@@ -4580,7 +4624,7 @@ function createAI(config = {}) {
     },
     async search(req, opts) {
       const labels = { ...cfg.labels, ...req.labels };
-      return search({ ...req, ...Object.keys(labels).length ? { labels } : {} }, { ...opts, ...costSink ? { costSink } : {} });
+      return search({ ...req, ...Object.keys(labels).length ? { labels } : {} }, { ...opts, ...cfg.byok ? { byok: true } : {}, ...costSink ? { costSink } : {} });
     },
     async judge(input) {
       input = judgeInputSchema.parse(input);
@@ -5330,6 +5374,7 @@ export {
   SDK_TAG,
   SEARCH_CACHE_TTL_MS,
   SEARCH_PRICE_USD,
+  SearchBudgetExceededError,
   SearchKeyMissingError,
   StreamHttpError,
   UpmetricsCostError,