@shop-os/modelkit 0.3.8 → 0.4.0

package/dist/index.cjs

@@ -3404,6 +3404,36 @@ function describeThrown(cause) {
 
 // src/retry.ts
 var textBody = import_zod6.z.string();
+var BILLING_CODES = /* @__PURE__ */ new Set([
+  "insufficient_quota",
+  "billing_hard_limit_reached",
+  "billing_not_active",
+  "payment_required",
+  "insufficient_balance",
+  "insufficient_credits",
+  "insufficient_credit"
+]);
+var errorCodes = import_zod6.z.object({
+  code: import_zod6.z.string().optional(),
+  type: import_zod6.z.string().optional(),
+  error: import_zod6.z.object({ code: import_zod6.z.string().optional(), type: import_zod6.z.string().optional() }).optional()
+});
+var BILLING_TEXT = /insufficient_quota|payment[ _]required|exhausted balance|insufficient (?:credits?|balance|funds)|billing_hard_limit/i;
+var BILLING_TEXT_STATUSES = /* @__PURE__ */ new Set([403, 429]);
+function billingCode(body) {
+  const parsed = errorCodes.safeParse(body);
+  if (!parsed.success) return false;
+  const { code, type, error } = parsed.data;
+  return [code, type, error?.code, error?.type].some(
+    (value) => value !== void 0 && BILLING_CODES.has(value.toLowerCase())
+  );
+}
+function isBillingFailure(status, body, text2) {
+  if (status === 402) return true;
+  if (status === void 0 || status < 400 || status >= 500) return false;
+  if (billingCode(body)) return true;
+  return BILLING_TEXT_STATUSES.has(status) && BILLING_TEXT.test(text2);
+}
 function classifyHttpError(status, body, provider, model) {
   const asText = textBody.safeParse(body);
   const text2 = asText.success ? asText.data : JSON.stringify(body);
@@ -3412,6 +3442,7 @@ function classifyHttpError(status, body, provider, model) {
   );
   let code = "provider_rejected";
   if (blocked) code = "content_blocked";
+  else if (isBillingFailure(status, body, text2)) code = "provider_billing";
   else if (status === 401 || status === 403) code = "auth";
   else if (status === 429) code = "rate_limited";
   else if (status === 408 || status === 504) code = "timeout";
@@ -3474,12 +3505,11 @@ function sleep(ms, signal) {
       rejectAborted();
       return;
     }
-    let timer;
     const onAbort = () => {
-      if (timer !== void 0) clearTimeout(timer);
+      clearTimeout(timer);
       rejectAborted();
     };
-    timer = setTimeout(() => {
+    const timer = setTimeout(() => {
       signal?.removeEventListener("abort", onAbort);
       resolve();
     }, ms);
@@ -9260,6 +9290,13 @@ function createModelKit(config, adapters) {
       model: model.id
     });
   }
+  function timed(event, startedAt) {
+    if (event.latencyMs !== void 0) return event;
+    return { ...event, latencyMs: Math.max(0, Date.now() - startedAt) };
+  }
+  function textTimed(event, stepCount, startedAt) {
+    return stepCount === 1 ? timed(event, startedAt) : event;
+  }
   async function emit(event) {
     try {
       await config.onUsage?.(event);
@@ -9343,17 +9380,21 @@ function createModelKit(config, adapters) {
     const abortSignal = combineSignals(req.abortSignal, call.signal);
     if (abortSignal) options.abortSignal = abortSignal;
     if (call.maxRetries !== void 0) options.maxRetries = call.maxRetries;
+    const startedAt = Date.now();
     const result = await (0, import_ai5.generateImage)(options);
     const images = result.images.map(
       (file) => toGeneratedImage({ bytes: file.uint8Array, mime: file.mediaType })
     );
-    const usage2 = {
-      model: model.modelId,
-      provider: model.provider,
-      modality: "IMAGE",
-      unit: "images",
-      quantity: images.length
-    };
+    const usage2 = timed(
+      {
+        model: model.modelId,
+        provider: model.provider,
+        modality: "IMAGE",
+        unit: "images",
+        quantity: images.length
+      },
+      startedAt
+    );
     await emit(usage2);
     return { image: images[0], images, usage: usage2, raw: result, warnings };
   }
@@ -9426,18 +9467,22 @@ function createModelKit(config, adapters) {
     if (providerOptions) options.providerOptions = providerOptions;
     if (abortSignal) options.abortSignal = abortSignal;
     logCoercions(logger, model.modelId, warnings);
+    const startedAt = Date.now();
     const pending = (0, import_ai5.experimental_generateVideo)(options);
     let settled;
     let failed = false;
     const done = pending.then(async (result) => {
-      const usage2 = {
-        model: model.modelId,
-        provider: model.provider,
-        modality: "VIDEO",
-        unit: "seconds",
-        quantity: seconds2 ?? 0,
-        ...seconds2 !== void 0 && { durationSeconds: seconds2 }
-      };
+      const usage2 = timed(
+        {
+          model: model.modelId,
+          provider: model.provider,
+          modality: "VIDEO",
+          unit: "seconds",
+          quantity: seconds2 ?? 0,
+          ...seconds2 !== void 0 && { durationSeconds: seconds2 }
+        },
+        startedAt
+      );
       settled = {
         video: { bytes: result.video.uint8Array, mime: result.video.mediaType },
         usage: usage2,
@@ -9494,8 +9539,12 @@ function createModelKit(config, adapters) {
           message: `${adapter.id} cannot generate images`,
           model: model.id
         });
+      const startedAt = Date.now();
       const result = await adapter.generateImage(request, ctx);
-      const usage2 = { ...result.usage, model: model.id, unit: model.usage.unit };
+      const usage2 = timed(
+        { ...result.usage, model: model.id, unit: model.usage.unit },
+        startedAt
+      );
       await emit(usage2);
       const images = result.images.map(toGeneratedImage);
       const first = images[0];
@@ -9538,6 +9587,7 @@ function createModelKit(config, adapters) {
           message: `${adapter.id} cannot generate video`,
           model: model.id
         });
+      const startedAt = Date.now();
       const job = await adapter.generateVideo(request, ctx);
       const result = job.result.bind(job);
       let settled;
@@ -9547,11 +9597,14 @@ function createModelKit(config, adapters) {
         result(opts) {
           if (settled) return settled;
           const pending = result(opts).then(
-            (out) => emit({
-              ...out.usage,
-              model: model.id,
-              unit: model.usage.unit
-            }).then(() => out),
+            async (out) => {
+              const usage2 = timed(
+                { ...out.usage, model: model.id, unit: model.usage.unit },
+                startedAt
+              );
+              await emit(usage2);
+              return { ...out, usage: usage2 };
+            },
             (cause) => {
               settled = void 0;
               throw cause;
@@ -9588,9 +9641,14 @@ function createModelKit(config, adapters) {
           model: model.id
         });
       }
+      const startedAt = Date.now();
       const result = await adapter.generateAudio(resolved, ctx);
-      await emit({ ...result.usage, model: model.id, unit: model.usage.unit });
-      return result;
+      const usage2 = timed(
+        { ...result.usage, model: model.id, unit: model.usage.unit },
+        startedAt
+      );
+      await emit(usage2);
+      return { ...result, usage: usage2 };
     },
     async prepareMediaPart(modelId, ref, mediaType, call = {}) {
       mediaRetryOptions(call);
@@ -9619,11 +9677,13 @@ function createModelKit(config, adapters) {
       const { webhookUrl, ...rest } = call;
       const { adapter, ctx } = videoToolTarget(req.model, rest);
       if (!adapter.submitVideoTool) throw noVideoTools(adapter, ctx.model);
-      return adapter.submitVideoTool(
+      const submittedAt = Date.now();
+      const submission = await adapter.submitVideoTool(
         req,
         ctx,
         webhookUrl !== void 0 ? { webhookUrl } : {}
       );
+      return { ...submission, submittedAt };
     },
     async videoToolStatus(sub, call = {}) {
       const { adapter, ctx } = videoToolTarget(sub.model, call);
@@ -9635,8 +9695,10 @@ function createModelKit(config, adapters) {
       const { adapter, ctx } = videoToolTarget(sub.model, rest);
       if (!adapter.videoToolResult) throw noVideoTools(adapter, ctx.model);
       const result = await adapter.videoToolResult(sub, ctx, webhook);
-      await emit({ ...result.usage, model: ctx.model.id, unit: ctx.model.usage.unit });
-      return result;
+      const base = { ...result.usage, model: ctx.model.id, unit: ctx.model.usage.unit };
+      const usage2 = sub.submittedAt === void 0 ? base : timed(base, sub.submittedAt);
+      await emit(usage2);
+      return { ...result, usage: usage2 };
     },
     async cancelVideoTool(sub, call = {}) {
       const { adapter, ctx } = videoToolTarget(sub.model, call);
@@ -9658,6 +9720,7 @@ function createModelKit(config, adapters) {
         (model) => textTarget(model, call),
         options.providerOptions
       );
+      const startedAt = Date.now();
       const result = await (0, import_ai5.generateText)({
         ...options,
         ...textAbortOptions(options.abortSignal, call.signal),
@@ -9667,7 +9730,8 @@ function createModelKit(config, adapters) {
           options.prepareStep ?? options.experimental_prepareStep
         )
       });
-      for (const event of steps.usage(result.steps)) await emit(withRowUnit(event));
+      for (const event of steps.usage(result.steps))
+        await emit(textTimed(withRowUnit(event), result.steps.length, startedAt));
       return result;
     },
     streamText(options, call = {}) {
@@ -9678,6 +9742,7 @@ function createModelKit(config, adapters) {
         options.providerOptions
       );
       const callerOnFinish = options.onFinish;
+      const startedAt = Date.now();
       return (0, import_ai5.streamText)({
         ...options,
         ...textAbortOptions(options.abortSignal, call.signal),
@@ -9687,7 +9752,8 @@ function createModelKit(config, adapters) {
           options.prepareStep ?? options.experimental_prepareStep
         ),
         onFinish: async (event) => {
-          for (const usage2 of steps.usage(event.steps)) await emit(withRowUnit(usage2));
+          for (const usage2 of steps.usage(event.steps))
+            await emit(textTimed(withRowUnit(usage2), event.steps.length, startedAt));
           await callerOnFinish?.(event);
         }
       });
@@ -9700,15 +9766,22 @@ function createModelKit(config, adapters) {
         model: languageModel2,
         providerOptions: mergeTextProviderOptions(entry, options.providerOptions)
       };
+      const startedAt = Date.now();
       const result = await (0, import_ai5.generateObject)(sdkOptions);
       await emit(
-        withRowUnit(textUsageEvent(entry, provider, result.usage, result.providerMetadata))
+        timed(
+          withRowUnit(
+            textUsageEvent(entry, provider, result.usage, result.providerMetadata)
+          ),
+          startedAt
+        )
       );
       return result;
     },
     streamObject(options, call = {}) {
       const { entry, languageModel: languageModel2, provider } = textTarget(options.model, call);
       const callerOnFinish = options.onFinish;
+      const startedAt = Date.now();
       const sdkOptions = {
         ...options,
         ...textAbortOptions(options.abortSignal, call.signal),
@@ -9716,8 +9789,16 @@ function createModelKit(config, adapters) {
         providerOptions: mergeTextProviderOptions(entry, options.providerOptions),
         onFinish: async (event) => {
           await emit(
-            withRowUnit(
-              textUsageEvent(entry, provider, event.usage, event.providerMetadata)
+            timed(
+              withRowUnit(
+                textUsageEvent(
+                  entry,
+                  provider,
+                  event.usage,
+                  event.providerMetadata
+                )
+              ),
+              startedAt
             )
           );
           await callerOnFinish?.(event);