Skip to content

[AISAGE-733] fix(ai-gateway): only expect is_byok from upstreams that report it #8875

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
chrarnoldus merged 1 commit into from
Oct 26, 2026
Merged

[AISAGE-733] fix(ai-gateway): only expect is_byok from upstreams that report it #8875

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 10 additions & 5 deletions packages/web-shared/src/lib/ai-gateway/llm-proxy-helpers.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -765,7 +765,8 @@ describe('parseTranscriptionUsageFromResponse', () => {
const result = parseTranscriptionUsageFromResponse(
makeResponse(),
200,
'openai/gpt-4o-mini-transcribe'
'openai/gpt-4o-mini-transcribe',
'openrouter'
);

expect(result.cost_mUsd).toBe(20);
Expand All @@ -787,7 +788,8 @@ describe('parseTranscriptionUsageFromResponse', () => {
},
}),
200,
'openai/gpt-4o-mini-transcribe'
'openai/gpt-4o-mini-transcribe',
'openrouter'
);

expect(result.cost_mUsd).toBe(40);
Expand All @@ -798,7 +800,8 @@ describe('parseTranscriptionUsageFromResponse', () => {
const result = parseTranscriptionUsageFromResponse(
makeResponse(),
200,
'openai/gpt-4o-mini-transcribe'
'openai/gpt-4o-mini-transcribe',
'openrouter'
);

expect(result.generation_time).toBe(2.5);
Expand All @@ -811,7 +814,8 @@ describe('parseTranscriptionUsageFromResponse', () => {
const result = parseTranscriptionUsageFromResponse(
JSON.stringify(parsed),
200,
'openai/whisper-1'
'openai/whisper-1',
'openrouter'
);

expect(result.model).toBe('openai/whisper-1');
Expand All @@ -824,7 +828,8 @@ describe('parseTranscriptionUsageFromResponse', () => {
const result = parseTranscriptionUsageFromResponse(
JSON.stringify(parsed),
200,
'openai/gpt-4o-mini-transcribe'
'openai/gpt-4o-mini-transcribe',
'openrouter'
);

expect(result.hasError).toBe(true);
Expand Down
18 changes: 14 additions & 4 deletions packages/web-shared/src/lib/ai-gateway/llm-proxy-helpers.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,10 @@ import type {
GatewayChatApiKind,
ProviderId,
} from '@kilocode/web-shared/lib/ai-gateway/providers/types';
import { computeOpenRouterCostFields } from '@kilocode/web-shared/lib/ai-gateway/processUsage.shared';
import {
computeOpenRouterCostFields,
providerReportsIsByok,
} from '@kilocode/web-shared/lib/ai-gateway/processUsage.shared';
import { ProxyErrorType } from '@kilocode/web-shared/lib/proxy-error-types';
import type { UserByokProviderId } from '@kilocode/web-shared/lib/ai-gateway/providers/openrouter/inference-provider-id';
import {
Expand Down Expand Up @@ -1370,7 +1373,8 @@ export function parseEmbeddingUsageFromResponse(
export function parseTranscriptionUsageFromResponse(
responseText: string,
statusCode: number,
requestedModel: string
requestedModel: string,
provider: ProviderId
): MicrodollarUsageStats {
const json = JSON.parse(responseText) as TranscriptionResponse;
const base = {
Expand All @@ -1391,7 +1395,8 @@ export function parseTranscriptionUsageFromResponse(
const cost = computeOpenRouterCostFields(
json.usage ?? {},
base,
'transcription_usage_processing'
'transcription_usage_processing',
providerReportsIsByok(provider)
);

return {
Expand Down Expand Up @@ -1473,7 +1478,12 @@ export function countAndStoreTranscriptionUsage(
: clonedResponse
.text()
.then(text =>
parseTranscriptionUsageFromResponse(text, statusCode, usageContext.requested_model)
parseTranscriptionUsageFromResponse(
text,
statusCode,
usageContext.requested_model,
usageContext.provider
)
)
.catch(() => null);

Expand Down
16 changes: 8 additions & 8 deletions packages/web-shared/src/lib/ai-gateway/processUsage.messages.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,7 +57,7 @@ describe('processMessagesApiUsage', () => {
cost_details: { upstream_inference_cost: 0.7 },
};

const result = processMessagesApiUsage(usage, null, coreProps);
const result = processMessagesApiUsage(usage, null, coreProps, 'openrouter');

expect(result.cost_mUsd).toBe(1000);
expect(result.is_byok).toBe(false);
Expand All @@ -80,7 +80,7 @@ describe('processMessagesApiUsage', () => {
cost_details: { upstream_inference_cost: 0.02 },
};

const result = processMessagesApiUsage(usage, null, coreProps);
const result = processMessagesApiUsage(usage, null, coreProps, 'openrouter');

expect(result.cost_mUsd).toBe(20000);
expect(result.is_byok).toBe(true);
Expand All @@ -99,7 +99,7 @@ describe('processMessagesApiUsage', () => {
gateway: { routing: { finalProvider: 'anthropic' }, cost: '0', marketCost: '0.000375' },
};

const result = processMessagesApiUsage(usage, providerMetadata, coreProps);
const result = processMessagesApiUsage(usage, providerMetadata, coreProps, 'openrouter');

expect(result.cost_mUsd).toBe(375); // toMicrodollars(0.000375)
expect(result.is_byok).toBeNull();
Expand Down Expand Up @@ -134,7 +134,7 @@ describe('processMessagesApiUsage', () => {
},
};

const result = processMessagesApiUsage(usage, providerMetadata, coreProps);
const result = processMessagesApiUsage(usage, providerMetadata, coreProps, 'openrouter');

expect(result.is_byok).toBe(true);
expect(result.cost_mUsd).toBe(402);
Expand Down Expand Up @@ -165,7 +165,7 @@ describe('processMessagesApiUsage', () => {
},
};

const result = processMessagesApiUsage(usage, providerMetadata, coreProps);
const result = processMessagesApiUsage(usage, providerMetadata, coreProps, 'openrouter');

expect(result.is_byok).toBe(false);
});
Expand Down Expand Up @@ -198,13 +198,13 @@ describe('processMessagesApiUsage', () => {
},
};

const result = processMessagesApiUsage(usage, providerMetadata, coreProps);
const result = processMessagesApiUsage(usage, providerMetadata, coreProps, 'openrouter');

expect(result.is_byok).toBe(false);
});

test('returns zero cost when no usage or metadata is provided', () => {
const result = processMessagesApiUsage(null, null, coreProps);
const result = processMessagesApiUsage(null, null, coreProps, 'openrouter');

expect(result.cost_mUsd).toBe(0);
expect(result.is_byok).toBeNull();
Expand Down Expand Up @@ -474,7 +474,7 @@ describe('parseMessagesMicrodollarUsageFromString approval tests', () => {
test(vercelMessagesJson, async () => {
const inputFile = join(sampleDir, vercelMessagesJson);
const jsonString = await readFile(inputFile, 'utf-8');
const result = parseMessagesMicrodollarUsageFromString(jsonString, 200);
const result = parseMessagesMicrodollarUsageFromString(jsonString, 200, 'openrouter');
const resultString = JSON.stringify(result, null, 2);
const approvalFilePath = inputFile + '.approved.json';
await verifyApproval(resultString, approvalFilePath);
Expand Down
14 changes: 9 additions & 5 deletions packages/web-shared/src/lib/ai-gateway/processUsage.messages.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@ import type {
import type { GatewayMessagesRequest } from '@kilocode/web-shared/lib/ai-gateway/providers/openrouter/types';
import {
computeOpenRouterCostFields,
providerReportsIsByok,
computeVercelCostMicrodollars,
drainSseStream,
extractVercelGenerationTime,
Expand Down Expand Up @@ -84,7 +85,8 @@ function getVercelResolvedModel(
export function processMessagesApiUsage(
usage: MessagesApiUsage | null | undefined,
providerMetadata: VercelProviderMetaData | null | undefined,
coreProps: NotYetCostedUsageStats
coreProps: NotYetCostedUsageStats,
provider: ProviderId
): JustTheCostsUsageStats {
const cacheHitTokens = usage?.cache_read_input_tokens ?? 0;
const cacheWriteTokens = usage?.cache_creation_input_tokens ?? 0;
Expand All @@ -96,7 +98,8 @@ export function processMessagesApiUsage(
const { cost_mUsd, is_byok } = computeOpenRouterCostFields(
usage,
coreProps,
'messages_sse_processing'
'messages_sse_processing',
providerReportsIsByok(provider)
);
return { inputTokens, outputTokens, cacheHitTokens, cacheWriteTokens, cost_mUsd, is_byok };
}
Expand Down Expand Up @@ -258,13 +261,14 @@ export async function parseMessagesMicrodollarUsageFromStream(
} satisfies NotYetCostedUsageStats;

const usage = deltaUsage ? mergeMessageDeltaUsage(startUsage, deltaUsage) : startUsage;
const costs = processMessagesApiUsage(usage, providerMetadata, coreProps);
const costs = processMessagesApiUsage(usage, providerMetadata, coreProps, provider);
return { ...coreProps, ...costs };
}

export function parseMessagesMicrodollarUsageFromString(
fullResponse: string,
statusCode: number
statusCode: number,
provider: ProviderId
): MicrodollarUsageStats {
const responseJson = JSON.parse(fullResponse) as
| (Anthropic.Messages.Message & MaybeHasVercelProviderMetadata & MaybeHasOpenRouterProvider)
Expand Down Expand Up @@ -297,7 +301,7 @@ export function parseMessagesMicrodollarUsageFromString(
status_code: statusCode,
} satisfies NotYetCostedUsageStats;

const costs = processMessagesApiUsage(usage, providerMetadata, coreProps);
const costs = processMessagesApiUsage(usage, providerMetadata, coreProps, provider);
return { ...coreProps, ...costs };
}

Expand Down
14 changes: 7 additions & 7 deletions packages/web-shared/src/lib/ai-gateway/processUsage.responses.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -119,7 +119,7 @@ describe('processResponsesApiUsage', () => {
cost_details: { upstream_inference_cost: 0.7 },
};

const result = processResponsesApiUsage(usage, null, coreProps);
const result = processResponsesApiUsage(usage, null, coreProps, 'openrouter');

expect(result.cost_mUsd).toBe(1000);
expect(result.is_byok).toBe(false);
Expand All @@ -141,7 +141,7 @@ describe('processResponsesApiUsage', () => {
cost_details: { upstream_inference_cost: 0.02 },
};

const result = processResponsesApiUsage(usage, null, coreProps);
const result = processResponsesApiUsage(usage, null, coreProps, 'openrouter');

expect(result.cost_mUsd).toBe(20000);
expect(result.is_byok).toBe(true);
Expand All @@ -159,7 +159,7 @@ describe('processResponsesApiUsage', () => {
gateway: { routing: { finalProvider: 'openai' }, cost: '0', marketCost: '0.0962825' },
};

const result = processResponsesApiUsage(usage, providerMetadata, coreProps);
const result = processResponsesApiUsage(usage, providerMetadata, coreProps, 'openrouter');

expect(result.cost_mUsd).toBe(96283); // toMicrodollars(0.0962825)
expect(result.is_byok).toBeNull();
Expand All @@ -170,7 +170,7 @@ describe('processResponsesApiUsage', () => {
});

test('returns zero cost when no usage or metadata is provided', () => {
const result = processResponsesApiUsage(null, null, coreProps);
const result = processResponsesApiUsage(null, null, coreProps, 'openrouter');

expect(result.cost_mUsd).toBe(0);
expect(result.is_byok).toBeNull();
Expand Down Expand Up @@ -202,7 +202,7 @@ describe('processResponsesApiUsage', () => {
},
};

const result = processResponsesApiUsage(usage, providerMetadata, coreProps);
const result = processResponsesApiUsage(usage, providerMetadata, coreProps, 'openrouter');

expect(result.is_byok).toBe(true);
expect(result.cost_mUsd).toBe(100);
Expand Down Expand Up @@ -285,7 +285,7 @@ describe('parseMicrodollarUsageFromString approval tests', () => {
test(openrouterResponsesJson, async () => {
const inputFile = join(sampleDir, openrouterResponsesJson);
const jsonString = await readFile(inputFile, 'utf-8');
const result = parseResponsesMicrodollarUsageFromString(jsonString, 200);
const result = parseResponsesMicrodollarUsageFromString(jsonString, 200, 'openrouter');
const resultString = JSON.stringify(result, null, 2);
const approvalFilePath = inputFile + '.approved.json';
await verifyApproval(resultString, approvalFilePath);
Expand All @@ -295,7 +295,7 @@ describe('parseMicrodollarUsageFromString approval tests', () => {
test(vercelResponsesJson, async () => {
const inputFile = join(sampleDir, vercelResponsesJson);
const jsonString = await readFile(inputFile, 'utf-8');
const result = parseResponsesMicrodollarUsageFromString(jsonString, 200);
const result = parseResponsesMicrodollarUsageFromString(jsonString, 200, 'openrouter');
const resultString = JSON.stringify(result, null, 2);
const approvalFilePath = inputFile + '.approved.json';
await verifyApproval(resultString, approvalFilePath);
Expand Down
14 changes: 9 additions & 5 deletions packages/web-shared/src/lib/ai-gateway/processUsage.responses.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@ import type {
import type { GatewayResponsesRequest } from '@kilocode/web-shared/lib/ai-gateway/providers/openrouter/types';
import {
computeOpenRouterCostFields,
providerReportsIsByok,
computeVercelCostMicrodollars,
drainSseStream,
extractVercelGenerationTime,
Expand Down Expand Up @@ -51,7 +52,8 @@ type ResponsesApiStreamEvent = {
export function processResponsesApiUsage(
usage: ResponsesApiUsage | null | undefined,
providerMetadata: VercelProviderMetaData | null | undefined,
coreProps: NotYetCostedUsageStats
coreProps: NotYetCostedUsageStats,
provider: ProviderId
): JustTheCostsUsageStats {
// OpenAI-style input_tokens already includes cached and cache-write tokens.
const inputTokens = usage?.input_tokens ?? 0;
Expand All @@ -64,7 +66,8 @@ export function processResponsesApiUsage(
const { cost_mUsd, is_byok } = computeOpenRouterCostFields(
usage,
coreProps,
'responses_sse_processing'
'responses_sse_processing',
providerReportsIsByok(provider)
);
return { inputTokens, outputTokens, cacheHitTokens, cacheWriteTokens, cost_mUsd, is_byok };
}
Expand Down Expand Up @@ -229,13 +232,14 @@ export async function parseResponsesMicrodollarUsageFromStream(
status_code: statusCode,
} satisfies NotYetCostedUsageStats;

const costs = processResponsesApiUsage(usage, providerMetadata, coreProps);
const costs = processResponsesApiUsage(usage, providerMetadata, coreProps, provider);
return { ...coreProps, ...costs };
}

export function parseResponsesMicrodollarUsageFromString(
fullResponse: string,
statusCode: number
statusCode: number,
provider: ProviderId
): MicrodollarUsageStats {
const responseJson = JSON.parse(fullResponse) as ResponsesApiResponse | null;

Expand All @@ -260,7 +264,7 @@ export function parseResponsesMicrodollarUsageFromString(
status_code: statusCode,
} satisfies NotYetCostedUsageStats;

const costs = processResponsesApiUsage(usage, providerMetadata, coreProps);
const costs = processResponsesApiUsage(usage, providerMetadata, coreProps, provider);
return { ...coreProps, ...costs };
}

Expand Down
20 changes: 18 additions & 2 deletions packages/web-shared/src/lib/ai-gateway/processUsage.shared.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,7 @@ import type { Span } from '@sentry/nextjs';
import { toMicrodollars } from '@kilocode/web-shared/lib/microdollars';
import { errorExceptInTest } from '@kilocode/web-shared/lib/utils.server';
import { OPENROUTER_BYOK_COST_MULTIPLIER } from '@kilocode/web-shared/lib/ai-gateway/processUsage.constants';
import type { ProviderId } from '@kilocode/web-shared/lib/ai-gateway/providers/types';
import type {
NotYetCostedUsageStats,
VercelProviderAttempt,
Expand All @@ -15,14 +16,29 @@ type OpenRouterCostFields = {
cost_details?: { upstream_inference_cost: number };
};

/**
* Whether the provider reports `usage.is_byok` next to `usage.cost`. OpenRouter
* does on chat completions, Messages, Responses and embeddings, and Vercel AI
* Gateway does on chat completions. Martian reports `cost` in the same shape but
* never `is_byok`.
*/
export function providerReportsIsByok(provider: ProviderId): boolean {
return provider === 'openrouter' || provider === 'vercel';
}

/**
* Computes cost_mUsd and is_byok from OpenRouter usage cost fields, and emits a
* Sentry warning when the accounting looks suspicious.
*
* With BYOK, OpenRouter's `cost` is only its 5% fee and the real cost is
* `cost_details.upstream_inference_cost`, so a cost without a BYOK status cannot
* be trusted. `requireIsByok` says whether that is suspicious for this response.
*/
export function computeOpenRouterCostFields(
usage: OpenRouterCostFields,
coreProps: NotYetCostedUsageStats,
source: string
source: string,
requireIsByok: boolean
): { cost_mUsd: number; is_byok: boolean | null } {
const is_byok = usage.is_byok ?? null;
const openrouterCost_USD = usage.cost ?? 0;
Expand All @@ -31,7 +47,7 @@ export function computeOpenRouterCostFields(
const inferredUpstream_USD = openrouterCost_USD * OPENROUTER_BYOK_COST_MULTIPLIER;
const microdollar_error = (inferredUpstream_USD - upstream_inference_cost_USD) * 1000000;
if (
(is_byok == null && (openrouterCost_USD || upstream_inference_cost_USD)) || // unknown byok status but known non-zero costs? We're borked!
(requireIsByok && is_byok == null && (openrouterCost_USD || upstream_inference_cost_USD)) || // unknown byok status but known non-zero costs? We're borked!
(is_byok && usage.cost !== 0 && 1.1 < Math.abs(microdollar_error)) // byok and cost is not 5% of upstream? Weird, EXCEPT sometimes cost is 0 due to openrouter promo.
) {
const { responseContent: _ignore, ...corePropsCopy } = coreProps;
Expand Down
Loading
Loading