@broberg/ai-sdk 0.51.0 → 0.53.0

package/dist/index.js

@@ -18,7 +18,7 @@ import {
   listMediaPrices,
   pricingFreshness,
   pricingGeneratedAt
-} from "./chunk-O4LR6X4U.js";
+} from "./chunk-ZSLRCOYM.js";
 
 // src/cost/default-hosts.ts
 var DEFAULT_BASE_URLS = {
@@ -517,6 +517,7 @@ function anthropicAdapter(config = {}) {
       cacheReadTokens: data.usage?.cache_read_input_tokens ?? 0,
       cacheCreationTokens: data.usage?.cache_creation_input_tokens ?? 0
     });
+    if (typeof data.model === "string" && data.model) usage.servedModel = data.model;
     const result = { text, usage };
     if (toolCalls.length > 0) result.toolCalls = toolCalls;
     return result;
@@ -809,6 +810,7 @@ function makeOpenAICompatibleAdapter(config) {
     if (config.costFromResponseField && typeof data.usage?.cost === "number") {
       usage.costUsd = data.usage.cost;
     }
+    if (typeof data.model === "string" && data.model) usage.servedModel = data.model;
     const result = { text, usage };
     if (toolCalls && toolCalls.length > 0) result.toolCalls = toolCalls;
     return result;
@@ -876,6 +878,7 @@ function makeOpenAICompatibleAdapter(config) {
         if (config.costFromResponseField && typeof chunk.usage.cost === "number") {
           usage.costUsd = chunk.usage.cost;
         }
+        if (typeof chunk.model === "string" && chunk.model) usage.servedModel = chunk.model;
         yield { type: "usage", costUsd: usage.costUsd, model: usage.model, usage };
       }
     }
@@ -3498,6 +3501,9 @@ var aiConfigSchema = z.object({
   /** F069.1 — customer-key mode. Every provider must be built with byokAdapter(), so no
    *  call can fall back to the fleet's env keys; createAI throws at setup otherwise. */
   byok: z.boolean().optional(),
+  /** F069.2 — attribution labels stamped on EVERY call's usage (and so on every cost-sink
+   *  row), e.g. { byok: "true", tenantId }. A call's own `labels` win on the same key. */
+  labels: z.record(z.string(), z.string()).optional(),
   /** F039.2 — client-wide opt-out from prompt caching (default: on where the
    *  provider supports it). A per-call `promptCache` wins over this. */
   promptCache: z.boolean().optional(),
@@ -3517,8 +3523,8 @@ var aiConfigSchema = z.object({
 import { randomUUID } from "crypto";
 
 // src/version.ts
-var VERSION = "0.51.0";
-var SDK_TAG = "@broberg/ai-sdk@0.51.0";
+var VERSION = "0.53.0";
+var SDK_TAG = "@broberg/ai-sdk@0.53.0";
 
 // src/cost/sinks/upmetrics.ts
 function isRetryableStatus(status) {
@@ -3568,6 +3574,9 @@ function upmetricsSink(config) {
         // capability/transport — no ingest-schema change, and without it the one
         // field built for auditability existed only in memory.
         region: usage.region,
+        // F071.2 — the model that answered (contract with upmetrics F048). Omitted when
+        // the provider did not say, rather than echoing the requested model as a guess.
+        ...usage.servedModel ? { served_model: usage.servedModel } : {},
         // F050: HOW cost_usd was arrived at. upmetrics already distinguishes
         // reported / computed / unpriced — we were sending an assumed number in
         // the same field as a measured one, so their labels could not be right
@@ -3594,6 +3603,36 @@ function upmetricsSink(config) {
     void config.complianceMode;
     return JSON.stringify(body);
   }
+  function buildFailureBody(f) {
+    const endedAt = new Date(new Date(f.ts).getTime()).toISOString();
+    const startedAt = new Date(new Date(f.ts).getTime() - (f.latencyMs || 0)).toISOString();
+    const body = {
+      mode: "record",
+      agent_kind: config.agentKind ?? (f.capability === "embedding" ? "embedding" : "chatbot"),
+      agent_name: config.agentName,
+      provider: f.provider,
+      model: f.model,
+      status: "error",
+      input_tokens: 0,
+      output_tokens: 0,
+      cost_usd: 0,
+      duration_ms: f.latencyMs,
+      started_at: startedAt,
+      ended_at: endedAt,
+      tags: {
+        ...f.labels,
+        capability: f.capability,
+        transport: f.transport,
+        error_code: f.errorCode,
+        error_kind: f.errorKind,
+        sdk: SDK_TAG,
+        idempotencyKey: randomUUID()
+      }
+    };
+    if (f.tier !== void 0) body.tier = f.tier;
+    if (f.purpose !== void 0) body.purpose = f.purpose;
+    return JSON.stringify(body);
+  }
   async function send(body) {
     try {
       const res = await doFetch(url, {
@@ -3687,6 +3726,30 @@ function upmetricsSink(config) {
         config.onError?.(err);
       }
     },
+    async recordFailure(failure) {
+      try {
+        const body = buildFailureBody(failure);
+        const r = await send(body);
+        if (r.kind === "ok") {
+          counts.sent += 1;
+          return;
+        }
+        if (r.kind === "reject") {
+          counts.rejected += 1;
+          config.onError?.(r.err);
+          return;
+        }
+        if (!retry) {
+          lose(r.err);
+          return;
+        }
+        queue.push({ body, attempts: 1 });
+        enforceCap();
+        schedule();
+      } catch (err) {
+        config.onError?.(err);
+      }
+    },
     async flush() {
       if (timer) {
         clearTimeout(timer);
@@ -3742,6 +3805,29 @@ function assertByokProviders(providers) {
   }
 }
 
+// src/cost/failure.ts
+function classifyFailure(err) {
+  const e = err;
+  const name = e?.name ?? "";
+  const message = String(e?.message ?? "");
+  if (name === "AbortError" || name === "TimeoutError" || /timed? ?out/i.test(message.split("\n")[0] ?? "")) {
+    return { errorCode: "timeout", errorKind: "timeout" };
+  }
+  let status = typeof e?.status === "number" ? e.status : void 0;
+  if (status === void 0) {
+    const m = /(?:^|\s)([45]\d{2})(?=[:\s]|$)/.exec(message.split("\n")[0] ?? "");
+    if (m) status = Number(m[1]);
+  }
+  if (status !== void 0) {
+    const kind = status === 401 || status === 403 ? "auth" : status === 404 ? "not_found" : status === 429 ? "rate_limit" : status >= 500 ? "server" : "other";
+    return { errorCode: String(status), errorKind: kind };
+  }
+  if (name === "TypeError" && /fetch failed|network|ECONN|ENOTFOUND/i.test(message + (e?.cause?.code ?? ""))) {
+    return { errorCode: "network", errorKind: "network" };
+  }
+  return { errorCode: "unknown", errorKind: "other" };
+}
+
 // src/client.ts
 var DEFAULT_IMAGE_SPEC = {
   provider: "fal",
@@ -3831,11 +3917,19 @@ function createAI(config = {}) {
     usage.capability = capability;
     if (tier) usage.tier = tier;
     if (purpose) usage.purpose = purpose;
-    if (labels && Object.keys(labels).length > 0) usage.labels = labels;
+    const merged = { ...cfg.labels, ...labels };
+    if (Object.keys(merged).length > 0) usage.labels = merged;
     usage.latencyMs = Math.round(latencyMs);
     if (!usage.ts) usage.ts = (/* @__PURE__ */ new Date()).toISOString();
     return usage;
   }
+  async function reportFailure(f, err) {
+    if (!costSink?.recordFailure) return;
+    try {
+      await costSink.recordFailure({ ...f, ...classifyFailure(err), ts: (/* @__PURE__ */ new Date()).toISOString() });
+    } catch {
+    }
+  }
   async function report(usage) {
     if (!costSink) return;
     try {
@@ -3871,8 +3965,8 @@ function createAI(config = {}) {
     for (let i = 0; i < routes.length; i++) {
       const spec = routes[i];
       await preflight(spec, opts.estIn, opts.estOut);
+      const t0 = performance.now();
       try {
-        const t0 = performance.now();
         const res = await opts.invoke(spec);
         enrich(res.usage, opts.capability, i === 0 ? opts.tier : void 0, opts.purpose, performance.now() - t0, opts.labels);
         await settle(res.usage);
@@ -3880,6 +3974,20 @@ function createAI(config = {}) {
         return res;
       } catch (e) {
         lastErr = e;
+        const labels = { ...cfg.labels, ...opts.labels };
+        await reportFailure(
+          {
+            provider: spec.provider,
+            model: spec.model,
+            transport: spec.transport,
+            capability: opts.capability,
+            ...i === 0 && opts.tier ? { tier: opts.tier } : {},
+            ...opts.purpose ? { purpose: opts.purpose } : {},
+            ...Object.keys(labels).length ? { labels } : {},
+            latencyMs: Math.round(performance.now() - t0)
+          },
+          e
+        );
       }
     }
     throw lastErr;
@@ -3949,6 +4057,19 @@ function createAI(config = {}) {
         return;
       } catch (e) {
         lastErr = e;
+        await reportFailure(
+          {
+            provider: spec.provider,
+            model: spec.model,
+            transport: spec.transport,
+            capability: "chat",
+            ...i === 0 ? { tier } : {},
+            ...input.purpose ? { purpose: input.purpose } : {},
+            ...Object.keys({ ...cfg.labels, ...input.labels }).length ? { labels: { ...cfg.labels, ...input.labels } } : {},
+            latencyMs: Math.round(performance.now() - t0)
+          },
+          e
+        );
         if (emitted || !eligibleForFallback(e)) {
           yield errorEvent(e);
           return;
@@ -4495,6 +4616,9 @@ function multiSink(sinks) {
   return {
     async record(usage) {
       await Promise.allSettled(sinks.map(async (s) => s.record(usage)));
+    },
+    async recordFailure(failure) {
+      await Promise.allSettled(sinks.map(async (s) => s.recordFailure?.(failure)));
     }
   };
 }