Skip to content

The open alternative to Salesforce, designed for AI.

fix(ai): stop double-counting cache-creation tokens in reported token totals (#23405)

Etienne74260a161d9a
diff --git a/packages/twenty-front/src/modules/ai/components/AiChatInitialLoadingIndicator.tsx b/packages/twenty-front/src/modules/ai/components/AiChatInitialLoadingIndicator.tsx
index d68648310cc..184f96b6575 100644
--- a/packages/twenty-front/src/modules/ai/components/AiChatInitialLoadingIndicator.tsx
+++ b/packages/twenty-front/src/modules/ai/components/AiChatInitialLoadingIndicator.tsx
@@ -10,6 +10,7 @@ const StyledLoadingIconContainer = styled.div`
display: flex;
justify-content: center;
padding-inline: ${themeCssVariables.spacing[1]};
+ width: fit-content;
`;
const StyledLoadingIconWrapper = styled.span`
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/__tests__/agent-async-executor.service.spec.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/__tests__/agent-async-executor.service.spec.ts
index d647d8d971b..b34ee889775 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/__tests__/agent-async-executor.service.spec.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/__tests__/agent-async-executor.service.spec.ts
@@ -200,6 +200,47 @@ describe('AgentAsyncExecutorService — workflow agent role-scoped tool resoluti
expect(result.creditsUsedMicro).toBe(4200);
});
+ it('emits the token total without re-adding cache-creation tokens', async () => {
+ roleTargetRepository.findOne.mockResolvedValueOnce({
+ roleId: agentRoleId,
+ });
+ aiBillingService.calculateCost.mockReturnValue(0.0042);
+ generateTextMock.mockResolvedValueOnce({
+ text: '',
+ steps: [
+ {
+ toolCalls: [],
+ providerMetadata: {
+ anthropic: { cacheCreationInputTokens: 30 },
+ },
+ },
+ ],
+ usage: {
+ ...baseUsage,
+ // inputTokens (100) is the full prompt: noCache(60) + cacheRead(10) +
+ // cacheCreation(30) — the emitted total must not add the 30 again
+ inputTokenDetails: {
+ noCacheTokens: 60,
+ cacheReadTokens: 10,
+ cacheWriteTokens: 30,
+ },
+ },
+ } as unknown as Awaited<ReturnType<typeof generateText>>);
+
+ await service.executeAgent({
+ agent: buildAgent(),
+ userPrompt: 'test',
+ workspaceId,
+ });
+
+ expect(aiBillingService.emitAiTokenUsageEvent).toHaveBeenCalledTimes(1);
+
+ const [, , emittedTotalTokens] =
+ aiBillingService.emitAiTokenUsageEvent.mock.calls[0];
+
+ expect(emittedTotalTokens).toBe(150);
+ });
+
it('folds native web search dollars into totalCostInDollars and creditsUsedMicro', async () => {
roleTargetRepository.findOne.mockResolvedValueOnce({
roleId: agentRoleId,
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/agent-async-executor.service.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/agent-async-executor.service.ts
index 944b5f67150..3efaf642c4c 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/agent-async-executor.service.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/agent-async-executor.service.ts
@@ -426,8 +426,7 @@ export class AgentAsyncExecutorService {
);
const totalTokens =
(accumulatedUsage.inputTokens ?? 0) +
- (accumulatedUsage.outputTokens ?? 0) +
- cacheCreationTokens;
+ (accumulatedUsage.outputTokens ?? 0);
void this.aiBillingService.emitAiTokenUsageEvent(
workspaceId,
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/__tests__/ai-billing.service.spec.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/__tests__/ai-billing.service.spec.ts
index 0d47a1b92ee..eef55b946a8 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/__tests__/ai-billing.service.spec.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/__tests__/ai-billing.service.spec.ts
@@ -384,5 +384,49 @@ describe('AiBillingService', () => {
'workspace-1',
);
});
+
+ it('should not add cache-creation tokens to the emitted token quantity', async () => {
+ mockAiModelRegistryService.getEffectiveModelConfig.mockReturnValue(
+ anthropicModelConfig as ReturnType<
+ AiModelRegistryService['getEffectiveModelConfig']
+ >,
+ );
+
+ await service.calculateAndBillUsage(
+ 'claude-sonnet-4-5-20250929',
+ {
+ usage: {
+ // inputTokens is the full prompt: noCache(400) + cacheRead(600) +
+ // cacheCreation(200), so quantity must be 1200 + 500, not + 200 again
+ inputTokens: 1200,
+ outputTokens: 500,
+ totalTokens: 1700,
+ inputTokenDetails: {
+ noCacheTokens: 400,
+ cacheReadTokens: 600,
+ cacheWriteTokens: 200,
+ },
+ outputTokenDetails: { textTokens: 500, reasoningTokens: 0 },
+ },
+ cacheCreationTokens: 200,
+ },
+ 'workspace-1',
+ UsageOperationType.AI_CHAT_TOKEN,
+ 'agent-id-123',
+ );
+
+ expect(
+ mockWorkspaceEventEmitter.emitCustomBatchEvent,
+ ).toHaveBeenCalledWith(
+ USAGE_RECORDED,
+ [
+ expect.objectContaining({
+ creditsUsedMicro: 9630,
+ quantity: 1700,
+ }),
+ ],
+ 'workspace-1',
+ );
+ });
});
});
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/ai-billing.service.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/ai-billing.service.ts
index fef33acdc7f..24bdaab1ce2 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/ai-billing.service.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/ai-billing.service.ts
@@ -74,8 +74,7 @@ export class AiBillingService {
const totalTokens =
(billingInput.usage.inputTokens ?? 0) +
- (billingInput.usage.outputTokens ?? 0) +
- (billingInput.cacheCreationTokens ?? 0);
+ (billingInput.usage.outputTokens ?? 0);
if (this.billingService.isBillingEnabled()) {
await this.billingUsageService.decrementAvailableCreditsInCache({
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/services/chat-execution.service.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/services/chat-execution.service.ts
index eb0b814eade..d74109dff50 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/services/chat-execution.service.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/services/chat-execution.service.ts
@@ -377,10 +377,7 @@ export class ChatExecutionService {
);
const cacheCreationTokens = extractCacheCreationTokensFromSteps(steps);
- const totalTokens =
- (usage.inputTokens ?? 0) +
- (usage.outputTokens ?? 0) +
- cacheCreationTokens;
+ const totalTokens = (usage.inputTokens ?? 0) + (usage.outputTokens ?? 0);
const costInDollars = this.aiBillingService.calculateCost(
registeredModel.modelId,
diff --git a/packages/twenty-server/src/modules/workflow/workflow-executor/workflow-actions/ai-agent/utils/__tests__/build-ai-agent-step-log.util.spec.ts b/packages/twenty-server/src/modules/workflow/workflow-executor/workflow-actions/ai-agent/utils/__tests__/build-ai-agent-step-log.util.spec.ts
index f09dae0c0f3..f5820c8b0db 100644
--- a/packages/twenty-server/src/modules/workflow/workflow-executor/workflow-actions/ai-agent/utils/__tests__/build-ai-agent-step-log.util.spec.ts
+++ b/packages/twenty-server/src/modules/workflow/workflow-executor/workflow-actions/ai-agent/utils/__tests__/build-ai-agent-step-log.util.spec.ts
@@ -60,7 +60,7 @@ describe('buildAiAgentStepLog', () => {
reasoningTokens: 10,
cacheReadTokens: 20,
cacheCreationTokens: 5,
- totalTokens: 155,
+ totalTokens: 150,
});
expect(stepLog.details.cost).toEqual({
totalCostInDollars: 0.012,
diff --git a/packages/twenty-server/src/modules/workflow/workflow-executor/workflow-actions/ai-agent/utils/build-ai-agent-step-log.util.ts b/packages/twenty-server/src/modules/workflow/workflow-executor/workflow-actions/ai-agent/utils/build-ai-agent-step-log.util.ts
index af279bc01b7..5677559d911 100644
--- a/packages/twenty-server/src/modules/workflow/workflow-executor/workflow-actions/ai-agent/utils/build-ai-agent-step-log.util.ts
+++ b/packages/twenty-server/src/modules/workflow/workflow-executor/workflow-actions/ai-agent/utils/build-ai-agent-step-log.util.ts
@@ -31,9 +31,7 @@ export const buildAiAgentStepLog = ({
executionResult.usage.outputTokenDetails?.reasoningTokens,
cacheReadTokens: executionResult.usage.inputTokenDetails?.cacheReadTokens,
cacheCreationTokens: executionResult.cacheCreationTokens,
- totalTokens:
- (executionResult.usage.totalTokens ?? 0) +
- executionResult.cacheCreationTokens,
+ totalTokens: executionResult.usage.totalTokens ?? 0,
},
cost: {
totalCostInDollars: executionResult.totalCostInDollars ?? 0,