fix(ai-gateway): populate is_byok from Vercel gateway modelAttempts

apps/web/src/lib/ai-gateway/processUsage.messages.test.ts

@@ -91,6 +91,97 @@ describe('processMessagesApiUsage', () => {
     expect(result.cacheWriteTokens).toBe(0);
   });
 
+  test('extracts is_byok=true from Vercel modelAttempts credentialType', () => {
+    const usage = {
+      input_tokens: 10,
+      output_tokens: 5,
+      cache_read_input_tokens: 0,
+      cache_creation_input_tokens: 0,
+      server_tool_use: { input_tokens: 0, web_fetch_requests: 0, web_search_requests: 0 },
+    };
+    const providerMetadata = {
+      gateway: {
+        routing: {
+          finalProvider: 'bedrock',
+          modelAttempts: [
+            {
+              success: true,
+              providerAttempts: [{ provider: 'bedrock', credentialType: 'byok', success: true }],
+            },
+          ],
+        },
+        cost: '0',
+        marketCost: '0.000402',
+      },
+    };
+
+    const result = processMessagesApiUsage(usage, providerMetadata, coreProps);
+
+    expect(result.is_byok).toBe(true);
+    expect(result.cost_mUsd).toBe(402);
+  });
+
+  test('extracts is_byok=false from Vercel system credentialType', () => {
+    const usage = {
+      input_tokens: 10,
+      output_tokens: 5,
+      cache_read_input_tokens: 0,
+      cache_creation_input_tokens: 0,
+      server_tool_use: { input_tokens: 0, web_fetch_requests: 0, web_search_requests: 0 },
+    };
+    const providerMetadata = {
+      gateway: {
+        routing: {
+          finalProvider: 'bedrock',
+          modelAttempts: [
+            {
+              success: true,
+              providerAttempts: [{ provider: 'bedrock', credentialType: 'system', success: true }],
+            },
+          ],
+        },
+        cost: '0',
+        marketCost: '0.000402',
+      },
+    };
+
+    const result = processMessagesApiUsage(usage, providerMetadata, coreProps);
+
+    expect(result.is_byok).toBe(false);
+  });
+
+  test('picks the successful provider attempt when earlier attempts failed', () => {
+    const usage = {
+      input_tokens: 10,
+      output_tokens: 5,
+      cache_read_input_tokens: 0,
+      cache_creation_input_tokens: 0,
+      server_tool_use: { input_tokens: 0, web_fetch_requests: 0, web_search_requests: 0 },
+    };
+    const providerMetadata = {
+      gateway: {
+        routing: {
+          finalProvider: 'anthropic',
+          modelAttempts: [
+            {
+              success: true,
+              providerAttempts: [
+                { provider: 'bedrock', credentialType: 'byok', success: false },
+                { provider: 'anthropic', credentialType: 'system', success: true },
+              ],
+            },
+          ],
+        },
+        cost: '0',
+        marketCost: '0.000402',
+      },
+    };
+
+    const result = processMessagesApiUsage(usage, providerMetadata, coreProps);
+
+    expect(result.is_byok).toBe(false);
+  });
+
   test('returns zero cost when no usage or metadata is provided', () => {
     const result = processMessagesApiUsage(null, null, coreProps);
 

apps/web/src/lib/ai-gateway/processUsage.messages.ts

@@ -15,6 +15,7 @@ import {
   computeOpenRouterCostFields,
   computeVercelCostMicrodollars,
   drainSseStream,
+  extractVercelIsByok,
 } from '@/lib/ai-gateway/processUsage.shared';
 import type Anthropic from '@anthropic-ai/sdk';
 
@@ -61,7 +62,7 @@ export function processMessagesApiUsage(
       cacheHitTokens,
       cacheWriteTokens,
       cost_mUsd,
-      is_byok: null,
+      is_byok: extractVercelIsByok(vercelGateway),
     };
   }
 

apps/web/src/lib/ai-gateway/processUsage.responses.test.ts

@@ -97,6 +97,36 @@ describe('processResponsesApiUsage', () => {
     expect(result.inputTokens).toBe(0);
     expect(result.outputTokens).toBe(0);
   });
+
+  test('extracts is_byok=true from Vercel modelAttempts credentialType', () => {
+    const usage = {
+      input_tokens: 10,
+      output_tokens: 5,
+      total_tokens: 15,
+      input_tokens_details: { cached_tokens: 0 },
+      output_tokens_details: { reasoning_tokens: 0 },
+    };
+    const providerMetadata = {
+      gateway: {
+        routing: {
+          finalProvider: 'openai',
+          modelAttempts: [
+            {
+              success: true,
+              providerAttempts: [{ provider: 'openai', credentialType: 'byok', success: true }],
+            },
+          ],
+        },
+        cost: '0',
+        marketCost: '0.0001',
+      },
+    };
+
+    const result = processResponsesApiUsage(usage, providerMetadata, coreProps);
+
+    expect(result.is_byok).toBe(true);
+    expect(result.cost_mUsd).toBe(100);
+  });
 });
 
 describe('parseMicrodollarUsageFromStream approval tests', () => {

apps/web/src/lib/ai-gateway/processUsage.responses.ts

@@ -16,6 +16,7 @@ import {
   computeOpenRouterCostFields,
   computeVercelCostMicrodollars,
   drainSseStream,
+  extractVercelIsByok,
 } from '@/lib/ai-gateway/processUsage.shared';
 
 // OpenRouter adds cost fields to the standard Responses API usage object.
@@ -72,7 +73,7 @@ export function processResponsesApiUsage(
       cacheHitTokens,
       cacheWriteTokens: 0,
       cost_mUsd,
-      is_byok: null,
+      is_byok: extractVercelIsByok(vercelGateway),
     };
   }
 

apps/web/src/lib/ai-gateway/processUsage.shared.ts

@@ -61,6 +61,29 @@ export function computeVercelCostMicrodollars(
   return toMicrodollars(isNaN(marketCost_USD) ? 0 : marketCost_USD);
 }
 
+/**
+ * Extracts whether the Vercel AI Gateway served the request with BYOK credentials.
+ *
+ * The gateway reports per-attempt `credentialType` ("byok" | "system") in
+ * `provider_metadata.gateway.routing.modelAttempts[].providerAttempts[]`. We look
+ * at the successful provider attempt within the successful model attempt. Returns
+ * `null` when credentialType is absent or no successful attempt is found.
+ */
+export function extractVercelIsByok(
+  vercelGateway: NonNullable<VercelProviderMetaData['gateway']> | undefined | null
+): boolean | null {
+  const modelAttempts = vercelGateway?.routing?.modelAttempts;
+  if (!modelAttempts) return null;
+  const successfulModel = modelAttempts.find(m => m.success) ?? modelAttempts.at(-1);
+  const providerAttempts = successfulModel?.providerAttempts;
+  if (!providerAttempts) return null;
+  const successfulProvider = providerAttempts.find(p => p.success) ?? providerAttempts.at(-1);
+  const credentialType = successfulProvider?.credentialType;
+  if (credentialType === 'byok') return true;
+  if (credentialType === 'system') return false;
+  return null;
+}
+
 /**
  * Drains a ReadableStream of binary chunks, calling `onTextChunk` for each
  * decoded piece of text. Handles client-abort (`ResponseAborted`) gracefully

apps/web/src/lib/ai-gateway/processUsage.ts

@@ -43,6 +43,7 @@ import type {
   OpenRouterUsage,
   PromptInfo,
   UsageMetaData,
+  VercelProviderMetaData,
 } from '@/lib/ai-gateway/processUsage.types';
 import {
   parseResponsesMicrodollarUsageFromStream,
@@ -53,7 +54,11 @@ import {
   parseMessagesMicrodollarUsageFromString,
 } from '@/lib/ai-gateway/processUsage.messages';
 import { OPENROUTER_BYOK_COST_MULTIPLIER } from '@/lib/ai-gateway/processUsage.constants';
-import { computeOpenRouterCostFields, drainSseStream } from '@/lib/ai-gateway/processUsage.shared';
+import {
+  computeOpenRouterCostFields,
+  drainSseStream,
+  extractVercelIsByok,
+} from '@/lib/ai-gateway/processUsage.shared';
 import { isClaudeModel } from '@/lib/ai-gateway/providers/anthropic.constants';
 import { isMinimaxModel } from '@/lib/ai-gateway/providers/minimax';
 import type { KiloExclusiveModel } from '@/lib/ai-gateway/providers/kilo-exclusive-model';
@@ -653,7 +658,8 @@ export function countAndStoreUsage(
 
 export function processOpenRouterUsage(
   usage: OpenRouterUsage | null | undefined,
-  coreProps: NotYetCostedUsageStats
+  coreProps: NotYetCostedUsageStats,
+  vercelProviderMetadata?: VercelProviderMetaData | null
 ): JustTheCostsUsageStats {
   // usage may be null when there's no response (e.g. error), so default to empty object
   const { cost_mUsd, is_byok } = computeOpenRouterCostFields(
@@ -671,7 +677,7 @@ export function processOpenRouterUsage(
       0,
     outputTokens: usage?.completion_tokens ?? 0,
     cost_mUsd,
-    is_byok,
+    is_byok: is_byok ?? extractVercelIsByok(vercelProviderMetadata?.gateway),
   };
 }
 
@@ -703,6 +709,7 @@ export async function parseMicrodollarUsageFromStream(
   let usage: OpenRouterUsage | null = null;
   let inference_provider: string | null = null;
   let finish_reason: string | null = null;
+  let vercelProviderMetadata: VercelProviderMetaData | null = null;
 
   const sseStreamParser = createParser({
     onEvent(event: EventSourceMessage) {
@@ -744,9 +751,13 @@ export async function parseMicrodollarUsageFromStream(
       messageId = json.id ?? messageId;
       usage = json.usage ?? usage;
       const choice = json.choices?.[0];
+      const chunkProviderMetadata = choice?.delta?.provider_metadata;
+      if (chunkProviderMetadata) {
+        vercelProviderMetadata = chunkProviderMetadata;
+      }
       inference_provider =
         json.provider ??
-        choice?.delta?.provider_metadata?.gateway?.routing?.finalProvider ??
+        chunkProviderMetadata?.gateway?.routing?.finalProvider ??
         inference_provider;
       finish_reason = choice?.finish_reason ?? finish_reason;
 
@@ -788,7 +799,7 @@ export async function parseMicrodollarUsageFromStream(
     status_code: effectiveStatusCode,
   };
 
-  const costs = processOpenRouterUsage(usage, coreProps);
+  const costs = processOpenRouterUsage(usage, coreProps, vercelProviderMetadata);
 
   return { ...coreProps, ...costs };
 }
@@ -831,7 +842,11 @@ export function parseMicrodollarUsageFromString(
     status_code: statusCode,
   };
 
-  const costs = processOpenRouterUsage(responseJson?.usage, coreProps);
+  const costs = processOpenRouterUsage(
+    responseJson?.usage,
+    coreProps,
+    choice?.message?.provider_metadata ?? null
+  );
 
   return { ...coreProps, ...costs };
 }

apps/web/src/lib/ai-gateway/processUsage.types.ts

@@ -19,8 +19,26 @@ export type OpenRouterUsage = {
   total_tokens: number;
 }; //ref: https://openrouter.ai/docs/use-cases/usage-accounting#response-format
 
+export type VercelProviderAttempt = {
+  provider?: string;
+  credentialType?: string;
+  success?: boolean;
+};
+
+export type VercelModelAttempt = {
+  success?: boolean;
+  providerAttempts?: VercelProviderAttempt[];
+};
+
 export type VercelProviderMetaData = {
-  gateway?: { routing?: { finalProvider?: string }; cost?: string; marketCost?: string };
+  gateway?: {
+    routing?: {
+      finalProvider?: string;
+      modelAttempts?: VercelModelAttempt[];
+    };
+    cost?: string;
+    marketCost?: string;
+  };
 };
 
 export type MaybeHasVercelProviderMetaData = {

apps/web/src/tests/sample/vercel-messages.log.resp.json.approved.json

@@ -17,5 +17,5 @@
   "cacheHitTokens": 0,
   "cacheWriteTokens": 0,
   "cost_mUsd": 402,
-  "is_byok": null
+  "is_byok": true
 }

apps/web/src/tests/sample/vercel-messages.log.resp.sse.approved.json

@@ -17,5 +17,5 @@
   "cacheHitTokens": 0,
   "cacheWriteTokens": 0,
   "cost_mUsd": 402,
-  "is_byok": null
+  "is_byok": true
 }

apps/web/src/tests/sample/vercel-responses.log.resp.json.approved.json

@@ -17,5 +17,5 @@
   "cacheHitTokens": 0,
   "cacheWriteTokens": 0,
   "cost_mUsd": 6138,
-  "is_byok": null
+  "is_byok": true
 }

apps/web/src/tests/sample/vercel-responses.log.resp.sse.approved.json

@@ -17,5 +17,5 @@
   "cacheHitTokens": 0,
   "cacheWriteTokens": 0,
   "cost_mUsd": 6138,
-  "is_byok": null
+  "is_byok": true
 }