@broberg/ai-sdk 0.17.1 → 0.19.0

package/dist/index.js

@@ -6,6 +6,9 @@ import {
   resolveModel,
   setAvailability
 } from "./chunk-HVZSYNZ5.js";
+import {
+  getPrice
+} from "./chunk-V5AQYES2.js";
 
 // src/routing/tier-map.ts
 var DEFAULT_TIER_MAP = {
@@ -214,101 +217,6 @@ function parseArgs(raw) {
   return {};
 }
 
-// src/cost/pricing.ts
-var V = "2026-06-02";
-var MS = "2026-06-04-mistral.ai";
-var PRICING = {
-  // Anthropic (direct API). DEFAULT_TIER_MAP: fast/cheap=haiku, smart/vision=sonnet, powerful=opus.
-  "anthropic:claude-haiku-4-5": {
-    inputPer1M: 0.8,
-    outputPer1M: 4,
-    cacheReadPer1M: 0.08,
-    cacheWritePer1M: 1,
-    version: V
-  },
-  "anthropic:claude-sonnet-4-6": {
-    inputPer1M: 3,
-    outputPer1M: 15,
-    cacheReadPer1M: 0.3,
-    cacheWritePer1M: 3.75,
-    version: V
-  },
-  "anthropic:claude-opus-4-8": {
-    inputPer1M: 15,
-    outputPer1M: 75,
-    cacheReadPer1M: 1.5,
-    cacheWritePer1M: 18.75,
-    version: V
-  },
-  // OpenAI. embedding default tier = text-embedding-3-small (no output tokens).
-  "openai:text-embedding-3-small": { inputPer1M: 0.02, outputPer1M: 0, version: V },
-  "openai:text-embedding-3-large": { inputPer1M: 0.13, outputPer1M: 0, version: V },
-  "openai:gpt-4o": { inputPer1M: 2.5, outputPer1M: 10, version: V },
-  "openai:gpt-4o-mini": { inputPer1M: 0.15, outputPer1M: 0.6, version: V },
-  // Whisper is priced per minute, not per token — not representable here; transcribe
-  // (F5.6) computes its own cost. Listed as 0 so token-based compute never charges it.
-  "openai:whisper-1": { inputPer1M: 0, outputPer1M: 0, version: V },
-  // OpenRouter (meta-router — model slugs include the upstream vendor). Slugs use
-  // dots (claude-sonnet-4.6) to match OpenRouter's live ids; the dashed forms
-  // never matched a real call. Caught by the F014 catalogue research.
-  "openrouter:anthropic/claude-sonnet-4.6": { inputPer1M: 3, outputPer1M: 15, version: V },
-  // OpenRouter ground-truth $1/$5 — a markup over Anthropic-direct's $0.8/$4
-  // (the `anthropic:` entry above). Was masked while the slug used dashes.
-  "openrouter:anthropic/claude-haiku-4.5": { inputPer1M: 1, outputPer1M: 5, version: "2026-06-04" },
-  "openrouter:google/gemini-2.5-flash": { inputPer1M: 0.3, outputPer1M: 2.5, version: V },
-  // Ground-truth from OpenRouter /api/v1/models (was a 0.3 estimate; now 0.279).
-  "openrouter:minimax/minimax-m2.7": {
-    inputPer1M: 0.279,
-    outputPer1M: 1.2,
-    version: "2026-06-04"
-  },
-  // DeepSeek V4 (CN-hosted — NOT GDPR-safe; non-personal-data workloads only).
-  // On 2026-05-22 DeepSeek made the "75% off" promo the permanent official price.
-  // V4-Pro $0.435/$0.87 is ~34x cheaper than GPT-5.5 on output; flash is cheaper
-  // still. Numbers match OpenRouter /api/v1/models 1:1 (no router markup). A strong
-  // cheap route for fleet background work once `claude -p` is API-billed (15 Jun).
-  "openrouter:deepseek/deepseek-v4-pro": { inputPer1M: 0.435, outputPer1M: 0.87, version: "2026-05-22-deepseek-official" },
-  "openrouter:deepseek/deepseek-v4-flash": { inputPer1M: 0.0983, outputPer1M: 0.1966, version: "2026-05-22-deepseek-official" },
-  // Google Gemini (direct). Provider key is "gemini" — matches the adapter's
-  // usage.provider + the override.provider callers pass. (Image-gen models are
-  // priced per-image in the adapter, not here.)
-  "gemini:gemini-2.5-flash": { inputPer1M: 0.3, outputPer1M: 2.5, version: V },
-  // flash-lite is the default `video` tier (F019) — cheap native video understanding.
-  "gemini:gemini-2.5-flash-lite": { inputPer1M: 0.1, outputPer1M: 0.4, version: "2026-06-04-or-xref" },
-  // Mistral (direct, La Plateforme). Official prices from mistral.ai/pricing
-  // (2026-06-04, per Christian's CD report). EU/Paris-hosted — the designated
-  // GDPR-safe provider for client/personal-data workloads (see F015). NB:
-  // medium-3.5 is the premium "Vibe" coding tier ($1.5/$7.5); Large 3 ($0.5/$1.5)
-  // is the cheaper frontier general-purpose model despite the higher number.
-  "mistral:mistral-large-latest": { inputPer1M: 0.5, outputPer1M: 1.5, version: MS },
-  "mistral:mistral-large-2512": { inputPer1M: 0.5, outputPer1M: 1.5, version: MS },
-  "mistral:mistral-medium-latest": { inputPer1M: 1.5, outputPer1M: 7.5, version: MS },
-  "mistral:mistral-medium-3.5": { inputPer1M: 1.5, outputPer1M: 7.5, version: MS },
-  "mistral:mistral-medium-3": { inputPer1M: 0.4, outputPer1M: 2, version: "2026-06-04-or-xref" },
-  "mistral:mistral-small-latest": { inputPer1M: 0.1, outputPer1M: 0.3, version: MS },
-  "mistral:mistral-small-2603": { inputPer1M: 0.1, outputPer1M: 0.3, version: MS },
-  "mistral:ministral-3b-latest": { inputPer1M: 0.1, outputPer1M: 0.1, version: MS },
-  "mistral:ministral-8b-latest": { inputPer1M: 0.15, outputPer1M: 0.15, version: MS },
-  "mistral:ministral-14b-latest": { inputPer1M: 0.2, outputPer1M: 0.2, version: MS },
-  "mistral:magistral-medium-latest": { inputPer1M: 2, outputPer1M: 5, version: MS },
-  "mistral:magistral-small-latest": { inputPer1M: 0.5, outputPer1M: 1.5, version: MS },
-  "mistral:devstral-latest": { inputPer1M: 0.4, outputPer1M: 2, version: MS },
-  "mistral:codestral-latest": { inputPer1M: 0.3, outputPer1M: 0.9, version: MS },
-  "mistral:open-mistral-nemo": { inputPer1M: 0.15, outputPer1M: 0.15, version: MS },
-  // Moderation (F016.4) — per input token; output 0. (OCR is per-page in the adapter.)
-  "mistral:mistral-moderation-latest": { inputPer1M: 0.1, outputPer1M: 0, version: MS },
-  // Embeddings (F016.5) — per input token.
-  "mistral:mistral-embed": { inputPer1M: 0.1, outputPer1M: 0, version: MS },
-  "mistral:codestral-embed": { inputPer1M: 0.15, outputPer1M: 0, version: MS }
-};
-function getPrice(provider, model) {
-  const exact = PRICING[`${provider}:${model}`];
-  if (exact) return exact;
-  const base = model.replace(/-\d{8}$/, "");
-  if (base !== model) return PRICING[`${provider}:${base}`];
-  return void 0;
-}
-
 // src/cost/usage.ts
 function computeCost(provider, model, inputTokens, outputTokens, cacheReadTokens = 0, cacheCreationTokens = 0) {
   const price = getPrice(provider, model);
@@ -1395,6 +1303,86 @@ function elevenlabsAdapter(config = {}) {
   return { name: "elevenlabs", dialogue, tts, listVoices };
 }
 
+// src/providers/azure.ts
+var AZURE_TTS_PRICE_PER_1K_CHARS = 0.016;
+var DEFAULT_REGION = "westeurope";
+var DEFAULT_FORMAT = "audio-24khz-48kbitrate-mono-mp3";
+var AZURE_DANISH_VOICE_LIST = [
+  { name: "christel", voiceId: "da-DK-ChristelNeural", gender: "female", display: "Christel", native: true, defaultRate: 0.85 },
+  { name: "seraphina", voiceId: "de-DE-SeraphinaMultilingualNeural", gender: "female", display: "Seraphina", native: false },
+  { name: "ava", voiceId: "en-US-AvaMultilingualNeural", gender: "female", display: "Ava", native: false },
+  { name: "jeppe", voiceId: "da-DK-JeppeNeural", gender: "male", display: "Jeppe", native: true },
+  { name: "florian", voiceId: "de-DE-FlorianMultilingualNeural", gender: "male", display: "Florian", native: false },
+  { name: "andrew", voiceId: "en-US-AndrewMultilingualNeural", gender: "male", display: "Andrew", native: false }
+];
+var AZURE_DANISH_VOICES = Object.fromEntries(
+  AZURE_DANISH_VOICE_LIST.map((v) => [v.name, v.voiceId])
+);
+function listAzureDanishVoices() {
+  return AZURE_DANISH_VOICE_LIST;
+}
+function resolveAzureVoice(nameOrVoice) {
+  return AZURE_DANISH_VOICES[nameOrVoice] ?? nameOrVoice;
+}
+function xmlEscape(s) {
+  return s.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;").replace(/'/g, "&apos;");
+}
+function localeOf(voice) {
+  const parts = voice.split("-");
+  return parts.length >= 2 ? `${parts[0]}-${parts[1]}` : "en-US";
+}
+function azureAdapter(config = {}) {
+  const fetchImpl = config.fetch ?? fetch;
+  function key() {
+    const k = config.apiKey ?? process.env.AZURE_SPEECH_KEY;
+    if (!k) throw new Error("azure adapter: API key not set (env AZURE_SPEECH_KEY)");
+    return k;
+  }
+  function region() {
+    return config.region ?? process.env.AZURE_SPEECH_REGION ?? DEFAULT_REGION;
+  }
+  function priceFor(chars, model) {
+    const usage = freshUsage({
+      provider: "azure",
+      model,
+      transport: "http",
+      capability: "tts",
+      inputTokens: 0,
+      outputTokens: 0
+    });
+    usage.costUsd = chars / 1e3 * (config.pricePer1kChars ?? AZURE_TTS_PRICE_PER_1K_CHARS);
+    return usage;
+  }
+  async function tts(req) {
+    const voice = resolveAzureVoice(req.voiceId);
+    const lang = req.lang ?? localeOf(voice);
+    const format = req.format ?? DEFAULT_FORMAT;
+    const escaped = xmlEscape(req.text);
+    const effRate = req.rate ?? AZURE_DANISH_VOICE_LIST.find((v) => v.voiceId === voice)?.defaultRate;
+    const inner = effRate != null && effRate !== 1 ? `<prosody rate='${effRate}'>${escaped}</prosody>` : escaped;
+    const ssml = `<speak version='1.0' xml:lang='${lang}'><voice name='${voice}'>${inner}</voice></speak>`;
+    const res = await fetchImpl(
+      `https://${region()}.tts.speech.microsoft.com/cognitiveservices/v1`,
+      {
+        method: "POST",
+        headers: {
+          "Ocp-Apim-Subscription-Key": key(),
+          "Content-Type": "application/ssml+xml",
+          "X-Microsoft-OutputFormat": format
+        },
+        body: ssml
+      }
+    );
+    if (!res.ok) {
+      const body = await res.text().catch(() => "");
+      throw new Error(`azure tts ${res.status}: ${body.slice(0, 300)}`);
+    }
+    const audio = new Uint8Array(await res.arrayBuffer());
+    return { audio, mimeType: "audio/mpeg", usage: priceFor(req.text.length, req.spec.model) };
+  }
+  return { name: "azure", tts };
+}
+
 // src/providers/fal.ts
 import { deflateRawSync, crc32 } from "zlib";
 var FAL_IMAGE_PRICE_ESTIMATE = {
@@ -1822,6 +1810,7 @@ var defaultProviders = {
   openrouter: openrouterAdapter(),
   mistral: mistralAdapter(),
   elevenlabs: elevenlabsAdapter(),
+  azure: azureAdapter(),
   fal: falAdapter(),
   bfl: bflAdapter()
 };
@@ -2206,6 +2195,9 @@ var podcastInputSchema = z.object({
 var ttsInputSchema = z.object({
   text: z.string(),
   voice: z.string(),
+  lang: z.string().optional(),
+  format: z.string().optional(),
+  rate: z.number().positive().optional(),
   ...callOptions
 });
 var budgetSchema = z.object({
@@ -2659,7 +2651,7 @@ function createAI(config = {}) {
         invoke: async (spec) => {
           const adapter = pickProvider(spec.provider);
           if (!adapter.tts) throw new Error(`createAI: provider "${spec.provider}" does not support tts`);
-          return adapter.tts({ text: input.text, voiceId: resolveVoice(input.voice), spec });
+          return adapter.tts({ text: input.text, voiceId: resolveVoice(input.voice), lang: input.lang, format: input.format, rate: input.rate, spec });
         }
       });
     },
@@ -2809,8 +2801,8 @@ var stubProviders = {
 };
 
 // src/version.ts
-var VERSION = "0.17.1";
-var SDK_TAG = "@broberg/ai-sdk@0.17.1";
+var VERSION = "0.19.0";
+var SDK_TAG = "@broberg/ai-sdk@0.19.0";
 
 // src/availability/refresh.ts
 var NOT_REFRESHED = { refreshed: false, checked: 0, markedUnavailable: [] };
@@ -3083,7 +3075,76 @@ async function getCostSummary(dbPath) {
   }
   return { totalUsd: total?.total ?? 0, byProvider, byCapability };
 }
+
+// src/cost/upmetrics-read.ts
+var UpmetricsCostError = class extends Error {
+  status;
+  code;
+  constructor(message, opts) {
+    super(message);
+    this.name = "UpmetricsCostError";
+    this.status = opts?.status;
+    this.code = opts?.code;
+  }
+};
+var usdFromMicro = (microUsd) => microUsd / 1e6;
+function buildQuery(q) {
+  const p = new URLSearchParams();
+  if (q.bucket) p.set("bucket", q.bucket);
+  if (q.from !== void 0) p.set("from", String(q.from));
+  if (q.to !== void 0) p.set("to", String(q.to));
+  if (q.window && q.from === void 0 && q.to === void 0) p.set("window", q.window);
+  if (q.provider) p.set("provider", q.provider);
+  if (q.model) p.set("model", q.model);
+  if (q.tier) p.set("tier", q.tier);
+  if (q.agentName) p.set("agent_name", q.agentName);
+  if (q.transport) p.set("transport", q.transport);
+  if (q.groupBy) p.set("groupBy", q.groupBy);
+  for (const [k, v] of Object.entries(q.tags ?? {})) p.set(`tag.${k}`, v);
+  const s = p.toString();
+  return s ? `?${s}` : "";
+}
+function upmetricsCostClient(config) {
+  const doFetch = config.fetch ?? fetch;
+  const base = config.baseUrl.replace(/\/$/, "");
+  async function get(path, query) {
+    let res;
+    try {
+      res = await doFetch(`${base}${path}${query}`, {
+        method: "GET",
+        headers: { "X-Upmetrics-Key": config.apiKey }
+      });
+    } catch (err) {
+      throw new UpmetricsCostError(
+        `upmetricsCostClient: ${path} unreachable: ${err?.message ?? String(err)}`
+      );
+    }
+    if (!res.ok) {
+      let code;
+      const text = await res.text().catch(() => "");
+      try {
+        code = JSON.parse(text).error;
+      } catch {
+      }
+      throw new UpmetricsCostError(
+        `upmetricsCostClient: ${path} returned ${res.status}${code ? ` (${code})` : ""}`,
+        { status: res.status, code }
+      );
+    }
+    return await res.json();
+  }
+  return {
+    summary(query = {}) {
+      return get("/api/cost/summary", buildQuery(query));
+    },
+    timeseries(query = {}) {
+      return get("/api/cost/timeseries", buildQuery(query));
+    }
+  };
+}
 export {
+  AZURE_DANISH_VOICES,
+  AZURE_DANISH_VOICE_LIST,
   BudgetExceededError,
   BudgetGuard,
   DEFAULT_TIER_MAP,
@@ -3091,11 +3152,13 @@ export {
   ModelUnavailableError,
   SDK_TAG,
   StreamHttpError,
+  UpmetricsCostError,
   VERSION,
   aiConfigSchema,
   anthropicAdapter,
   anthropicApiAdapter,
   anthropicSubprocessAdapter,
+  azureAdapter,
   bflAdapter,
   bflCredits,
   chatInputSchema,
@@ -3115,6 +3178,7 @@ export {
   getPrice,
   httpTransport,
   imageInputSchema,
+  listAzureDanishVoices,
   listModels,
   makeContracts,
   makeOpenAICompatibleAdapter,
@@ -3130,6 +3194,7 @@ export {
   refreshAvailability,
   resetRefreshState,
   resetRegistry,
+  resolveAzureVoice,
   resolveModel,
   resolveTier,
   resolveVoice,
@@ -3142,7 +3207,9 @@ export {
   toProviderTools,
   toolSchema,
   translateInputSchema,
+  upmetricsCostClient,
   upmetricsSink,
+  usdFromMicro,
   visionInputSchema
 };
 //# sourceMappingURL=index.js.map