@broberg/ai-sdk 0.7.0 → 0.9.0

package/dist/index.js

@@ -280,7 +280,10 @@ var PRICING = {
   "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 }
+  "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}`];
@@ -1028,6 +1031,11 @@ function openrouterAdapter(config = {}) {
 
 // src/providers/mistral.ts
 var MISTRAL_OCR_PRICE_PER_PAGE = 2e-3;
+var VOXTRAL_PRICE_PER_MIN = {
+  "voxtral-mini-latest": 2e-3,
+  "voxtral-mini-2507": 2e-3,
+  "voxtral-mini-2602": 2e-3
+};
 function mistralAdapter(config = {}) {
   const baseUrl = config.baseUrl ?? "https://api.mistral.ai/v1";
   const base = makeOpenAICompatibleAdapter({ name: "mistral", baseUrl, apiKey: config.apiKey });
@@ -1097,7 +1105,175 @@ function mistralAdapter(config = {}) {
     });
     return { results, usage };
   }
-  return { ...base, ocr, moderate };
+  async function embedding(req) {
+    const res = await fetchImpl(`${baseUrl}/embeddings`, {
+      method: "POST",
+      headers: { "content-type": "application/json", authorization: `Bearer ${key()}` },
+      body: JSON.stringify({ model: req.spec.model, input: req.input })
+    });
+    if (!res.ok) {
+      const body = await res.text().catch(() => "");
+      throw new Error(`mistral embeddings ${res.status}: ${body.slice(0, 300)}`);
+    }
+    const data = await res.json();
+    const vectors = (data.data ?? []).map((d) => d.embedding);
+    const usage = freshUsage({
+      provider: "mistral",
+      model: req.spec.model,
+      transport: "http",
+      capability: "embedding",
+      inputTokens: data.usage?.prompt_tokens ?? data.usage?.total_tokens ?? 0,
+      outputTokens: 0
+    });
+    return { vectors, usage };
+  }
+  async function transcribe(req) {
+    const form = new FormData();
+    form.append("file", new Blob([req.audio]), "audio");
+    form.append("model", req.spec.model);
+    if (req.language) form.append("language", req.language);
+    const res = await fetchImpl(`${baseUrl}/audio/transcriptions`, {
+      method: "POST",
+      headers: { authorization: `Bearer ${key()}` },
+      body: form
+    });
+    if (!res.ok) {
+      const body = await res.text().catch(() => "");
+      throw new Error(`mistral transcribe ${res.status}: ${body.slice(0, 300)}`);
+    }
+    const data = await res.json();
+    const usage = freshUsage({
+      provider: "mistral",
+      model: req.spec.model,
+      transport: "http",
+      capability: "transcribe",
+      inputTokens: 0,
+      outputTokens: 0
+    });
+    if (req.durationSec !== void 0) {
+      usage.costUsd = req.durationSec / 60 * (VOXTRAL_PRICE_PER_MIN[req.spec.model] ?? 0);
+    }
+    return { text: data.text ?? "", usage };
+  }
+  async function batchSubmit(req) {
+    const jsonl = req.items.map(
+      (it) => JSON.stringify({
+        custom_id: it.customId,
+        body: { model: req.spec.model, messages: [{ role: "user", content: it.prompt }] }
+      })
+    ).join("\n");
+    const form = new FormData();
+    form.append("purpose", "batch");
+    form.append("file", new Blob([jsonl], { type: "application/jsonl" }), "batch.jsonl");
+    const up = await fetchImpl(`${baseUrl}/files`, {
+      method: "POST",
+      headers: { authorization: `Bearer ${key()}` },
+      body: form
+    });
+    if (!up.ok) throw new Error(`mistral batch upload ${up.status}: ${(await up.text().catch(() => "")).slice(0, 200)}`);
+    const fileId = (await up.json()).id;
+    const job = await fetchImpl(`${baseUrl}/batch/jobs`, {
+      method: "POST",
+      headers: { "content-type": "application/json", authorization: `Bearer ${key()}` },
+      body: JSON.stringify({ input_files: [fileId], model: req.spec.model, endpoint: "/v1/chat/completions" })
+    });
+    if (!job.ok) throw new Error(`mistral batch job ${job.status}: ${(await job.text().catch(() => "")).slice(0, 200)}`);
+    const data = await job.json();
+    return { jobId: data.id ?? "", status: data.status ?? "queued", total: data.total_requests };
+  }
+  async function batchStatus(req) {
+    const res = await fetchImpl(`${baseUrl}/batch/jobs/${req.jobId}`, { headers: { authorization: `Bearer ${key()}` } });
+    if (!res.ok) throw new Error(`mistral batch status ${res.status}`);
+    const d = await res.json();
+    return { jobId: d.id ?? req.jobId, status: d.status ?? "unknown", total: d.total_requests, completed: d.succeeded_requests };
+  }
+  async function batchResults(req) {
+    const job = await fetchImpl(`${baseUrl}/batch/jobs/${req.jobId}`, { headers: { authorization: `Bearer ${key()}` } });
+    if (!job.ok) throw new Error(`mistral batch results ${job.status}`);
+    const outputFile = (await job.json()).output_file;
+    if (!outputFile) throw new Error("mistral batch: job has no output_file yet (not finished)");
+    const content = await fetchImpl(`${baseUrl}/files/${outputFile}/content`, { headers: { authorization: `Bearer ${key()}` } });
+    if (!content.ok) throw new Error(`mistral batch download ${content.status}`);
+    const lines = (await content.text()).trim().split("\n").filter(Boolean);
+    return lines.map((line) => {
+      const row = JSON.parse(line);
+      return { customId: row.custom_id ?? "", text: row.response?.body?.choices?.[0]?.message?.content ?? "" };
+    });
+  }
+  return { ...base, ocr, moderate, embedding, transcribe, batchSubmit, batchStatus, batchResults };
+}
+
+// src/providers/elevenlabs.ts
+var ELEVENLABS_PRICE_PER_1K_CHARS = 0.15;
+var ELEVENLABS_DANISH_VOICES = {
+  soren: "xj6X4BCUsv9oxohm1E8o",
+  jesper: "Bl1YwS3uJac5zEOSNESn",
+  mads: "BIWC0507fYMfhPcAEIRP",
+  noam: "V34B5u5UbLdNJVEkcgXp",
+  camilla: "4RklGmuxoAskAbGXplXN"
+};
+function resolveVoice(nameOrId) {
+  return ELEVENLABS_DANISH_VOICES[nameOrId] ?? nameOrId;
+}
+function elevenlabsAdapter(config = {}) {
+  const baseUrl = config.baseUrl ?? "https://api.elevenlabs.io/v1";
+  const fetchImpl = config.fetch ?? fetch;
+  function key() {
+    const k = config.apiKey ?? process.env.ELEVENLABS_API_KEY;
+    if (!k) throw new Error("elevenlabs adapter: API key not set (env ELEVENLABS_API_KEY)");
+    return k;
+  }
+  function priceFor(chars, model) {
+    const usage = freshUsage({
+      provider: "elevenlabs",
+      model,
+      transport: "http",
+      capability: "podcast",
+      inputTokens: 0,
+      outputTokens: 0
+    });
+    usage.costUsd = chars / 1e3 * (config.pricePer1kChars ?? ELEVENLABS_PRICE_PER_1K_CHARS);
+    return usage;
+  }
+  async function dialogue(req) {
+    const res = await fetchImpl(`${baseUrl}/text-to-dialogue`, {
+      method: "POST",
+      headers: { "xi-api-key": key(), "content-type": "application/json", accept: "audio/mpeg" },
+      body: JSON.stringify({
+        model_id: req.spec.model,
+        inputs: req.inputs.map((t) => ({ text: t.text, voice_id: t.voiceId })),
+        ...req.format ? { output_format: req.format } : {}
+      })
+    });
+    if (!res.ok) {
+      const body = await res.text().catch(() => "");
+      throw new Error(`elevenlabs dialogue ${res.status}: ${body.slice(0, 300)}`);
+    }
+    const audio = new Uint8Array(await res.arrayBuffer());
+    const chars = req.inputs.reduce((n, t) => n + t.text.length, 0);
+    return { audio, mimeType: "audio/mpeg", usage: priceFor(chars, req.spec.model) };
+  }
+  async function tts(req) {
+    const model = req.spec.model;
+    const res = await fetchImpl(`${baseUrl}/text-to-speech/${req.voiceId}`, {
+      method: "POST",
+      headers: { "xi-api-key": key(), "content-type": "application/json", accept: "audio/mpeg" },
+      body: JSON.stringify({ text: req.text, model_id: model })
+    });
+    if (!res.ok) {
+      const body = await res.text().catch(() => "");
+      throw new Error(`elevenlabs tts ${res.status}: ${body.slice(0, 300)}`);
+    }
+    const audio = new Uint8Array(await res.arrayBuffer());
+    return { audio, mimeType: "audio/mpeg", usage: priceFor(req.text.length, model) };
+  }
+  async function listVoices() {
+    const res = await fetchImpl(`${baseUrl}/voices`, { headers: { "xi-api-key": key() } });
+    if (!res.ok) throw new Error(`elevenlabs voices ${res.status}`);
+    const data = await res.json();
+    return (data.voices ?? []).map((v) => ({ voiceId: v.voice_id, name: v.name, language: v.labels?.language }));
+  }
+  return { name: "elevenlabs", dialogue, tts, listVoices };
 }
 
 // src/providers/fal.ts
@@ -1188,6 +1364,7 @@ var defaultProviders = {
   deepinfra: deepinfraAdapter(),
   openrouter: openrouterAdapter(),
   mistral: mistralAdapter(),
+  elevenlabs: elevenlabsAdapter(),
   fal: falAdapter()
 };
 
@@ -1510,6 +1687,17 @@ var moderationInputSchema = z.object({
   input: z.union([z.string(), z.array(z.string())]),
   ...callOptions
 });
+var podcastInputSchema = z.object({
+  script: z.array(z.object({ speaker: z.string(), text: z.string() })).min(1),
+  voices: z.record(z.string(), z.string()),
+  format: z.string().optional(),
+  ...callOptions
+});
+var ttsInputSchema = z.object({
+  text: z.string(),
+  voice: z.string(),
+  ...callOptions
+});
 var budgetSchema = z.object({
   perCallUsd: z.number().positive().optional(),
   rollingUsd: z.number().positive().optional()
@@ -1531,6 +1719,9 @@ var DEFAULT_IMAGE_SPEC = {
 };
 var DEFAULT_OCR_SPEC = { provider: "mistral", model: "mistral-ocr-latest", transport: "http" };
 var DEFAULT_MODERATION_SPEC = { provider: "mistral", model: "mistral-moderation-latest", transport: "http" };
+var DEFAULT_PODCAST_SPEC = { provider: "elevenlabs", model: "eleven_v3", transport: "http" };
+var DEFAULT_TTS_SPEC = { provider: "elevenlabs", model: "eleven_multilingual_v2", transport: "http" };
+var DEFAULT_BATCH_SPEC = { provider: "mistral", model: "mistral-small-latest", transport: "http" };
 function createAI(config = {}) {
   const cfg = aiConfigSchema.parse(config);
   const providers = cfg.providers ?? defaultProviders;
@@ -1810,6 +2001,48 @@ function createAI(config = {}) {
         }
       });
     },
+    async podcast(input) {
+      input = podcastInputSchema.parse(input);
+      const inputs = input.script.map((turn) => {
+        const mapped = input.voices[turn.speaker];
+        if (!mapped) throw new Error(`ai.podcast: no voice mapped for speaker "${turn.speaker}"`);
+        return { text: turn.text, voiceId: resolveVoice(mapped) };
+      });
+      const chars = input.script.reduce((n, t) => n + t.text.length, 0);
+      return runCapability({
+        primary: { ...DEFAULT_PODCAST_SPEC, ...input.override },
+        fallback: input.fallback,
+        capability: "podcast",
+        purpose: input.purpose,
+        labels: input.labels,
+        estIn: chars,
+        // per-character cost (not token-based)
+        estOut: 0,
+        invoke: async (spec) => {
+          const adapter = pickProvider(spec.provider);
+          if (!adapter.dialogue) throw new Error(`createAI: provider "${spec.provider}" does not support podcast/dialogue`);
+          return adapter.dialogue({ inputs, format: input.format, spec });
+        }
+      });
+    },
+    async tts(input) {
+      input = ttsInputSchema.parse(input);
+      return runCapability({
+        primary: { ...DEFAULT_TTS_SPEC, ...input.override },
+        fallback: input.fallback,
+        capability: "tts",
+        purpose: input.purpose,
+        labels: input.labels,
+        estIn: input.text.length,
+        // per-character cost
+        estOut: 0,
+        invoke: async (spec) => {
+          const adapter = pickProvider(spec.provider);
+          if (!adapter.tts) throw new Error(`createAI: provider "${spec.provider}" does not support tts`);
+          return adapter.tts({ text: input.text, voiceId: resolveVoice(input.voice), spec });
+        }
+      });
+    },
     async embedding(input) {
       input = embeddingInputSchema.parse(input);
       const tier = input.tier ?? EMBEDDING_DEFAULT_TIER;
@@ -1848,6 +2081,26 @@ function createAI(config = {}) {
         }
       });
     },
+    batch: {
+      async submit(input) {
+        const spec = { ...DEFAULT_BATCH_SPEC, ...input.override };
+        const adapter = pickProvider(spec.provider);
+        if (!adapter.batchSubmit) throw new Error(`createAI: provider "${spec.provider}" does not support batch`);
+        return adapter.batchSubmit({ items: input.requests, spec });
+      },
+      async status(jobId, override) {
+        const spec = { ...DEFAULT_BATCH_SPEC, ...override };
+        const adapter = pickProvider(spec.provider);
+        if (!adapter.batchStatus) throw new Error(`createAI: provider "${spec.provider}" does not support batch`);
+        return adapter.batchStatus({ jobId, spec });
+      },
+      async results(jobId, override) {
+        const spec = { ...DEFAULT_BATCH_SPEC, ...override };
+        const adapter = pickProvider(spec.provider);
+        if (!adapter.batchResults) throw new Error(`createAI: provider "${spec.provider}" does not support batch`);
+        return adapter.batchResults({ jobId, spec });
+      }
+    },
     // Replaced below with the real prompt-contracts (needs the client itself).
     contracts: void 0
   };
@@ -1936,8 +2189,8 @@ var stubProviders = {
 };
 
 // src/version.ts
-var VERSION = "0.7.0";
-var SDK_TAG = "@broberg/ai-sdk@0.7.0";
+var VERSION = "0.9.0";
+var SDK_TAG = "@broberg/ai-sdk@0.9.0";
 
 // src/cost/budget-store.ts
 function sqliteBudgetStore(config) {
@@ -2169,6 +2422,7 @@ export {
   BudgetExceededError,
   BudgetGuard,
   DEFAULT_TIER_MAP,
+  ELEVENLABS_DANISH_VOICES,
   SDK_TAG,
   StreamHttpError,
   VERSION,
@@ -2182,6 +2436,7 @@ export {
   deepinfraAdapter,
   defaultProviders,
   discordSink,
+  elevenlabsAdapter,
   embeddingInputSchema,
   falAdapter,
   falStubAdapter,
@@ -2204,6 +2459,7 @@ export {
   parseClaudeCliJson,
   parseJsonLoose,
   resolveTier,
+  resolveVoice,
   sqliteBudgetStore,
   sqliteSink,
   streamTransport,