feat(ai-gateway): add organization auto model routing

apps/web/src/app/(app)/organizations/[id]/custom-modes/page.tsx

@@ -9,7 +9,13 @@ export default async function OrganizationCustomModesPage({
   return (
     <OrganizationByPageLayout
       params={params}
-      render={({ organization }) => <CustomModesLayout organizationId={organization.id} />}
+      render={({ organization, role, isGlobalAdmin }) => (
+        <CustomModesLayout
+          organizationId={organization.id}
+          role={role}
+          isGlobalAdmin={isGlobalAdmin}
+        />
+      )}
     />
   );
 }

apps/web/src/app/api/openrouter/[...path]/route.test.ts

@@ -13,6 +13,7 @@ import type { Provider } from '@/lib/ai-gateway/providers/types';
 import { fetchEfficientAutoDecision } from '@/lib/ai-gateway/auto-routing-decision';
 import { logMicrodollarUsage } from '@/lib/ai-gateway/processUsage';
 import { applyResolvedAutoModel } from '@/lib/ai-gateway/auto-model/resolution';
+import { getDirectByokModel } from '@/lib/ai-gateway/providers/direct-byok';
 
 jest.mock('next/server', () => {
   return {
@@ -40,6 +41,9 @@ jest.mock('@/lib/ai-gateway/abuse-service', () => {
   };
 });
 jest.mock('@/lib/ai-gateway/providers/get-provider');
+jest.mock('@/lib/ai-gateway/providers/direct-byok', () => ({
+  getDirectByokModel: jest.fn(async () => ({ provider: null, model: null })),
+}));
 jest.mock('@/lib/ai-gateway/providers/upstream-request');
 jest.mock('@/lib/ai-gateway/providers/gateway-models-cache');
 jest.mock('@/lib/redis', () => ({
@@ -90,6 +94,7 @@ const mockedRedisSet = jest.mocked(redisClient.set);
 const mockedFetchEfficientAutoDecision = jest.mocked(fetchEfficientAutoDecision);
 const mockedLogMicrodollarUsage = jest.mocked(logMicrodollarUsage);
 const mockedApplyResolvedAutoModel = jest.mocked(applyResolvedAutoModel);
+const mockedGetDirectByokModel = jest.mocked(getDirectByokModel);
 
 const provider = {
   id: 'openrouter',
@@ -413,7 +418,9 @@ describe('POST /api/openrouter/v1/chat/completions rules-engine actions', () =>
 describe('kilo-auto/efficient classifier billing', () => {
   beforeEach(() => {
     jest.clearAllMocks();
+    mockedGetDirectByokModel.mockResolvedValue({ provider: null, model: null });
     setUserAuth();
+
     mockedGetProvider.mockResolvedValue({
       kind: 'provider',
       provider,
@@ -443,6 +450,48 @@ describe('kilo-auto/efficient classifier billing', () => {
     });
   });
 
+  it('rejects Organization Auto direct-BYOK routes when provider selection falls through', async () => {
+    mockedGetUserFromAuth.mockResolvedValue({
+      user: {
+        id: 'user-123',
+        google_user_email: 'test@example.com',
+        microdollars_used: 0,
+      } as User,
+      authFailedResponse: null,
+      organizationId: 'org-1',
+    });
+    mockedGetBalanceAndOrgSettings.mockResolvedValue({
+      balance: 1000,
+      settings: {
+        default_model: 'kilo-auto/org',
+        org_auto_model: { routes: {}, fallback_model: 'kilo-auto/balanced' },
+      },
+      plan: 'enterprise',
+    });
+    mockedApplyResolvedAutoModel.mockImplementation(async (_params, request) => {
+      request.body.model = 'martian/moonshotai/kimi-k2.6';
+      return {
+        kind: 'ok',
+        resolved: { model: 'martian/moonshotai/kimi-k2.6' },
+        routingTarget: 'martian/moonshotai/kimi-k2.6',
+      };
+    });
+    mockedGetDirectByokModel.mockResolvedValue({
+      provider: { id: 'martian' } as never,
+      model: {} as never,
+    });
+
+    const { POST } = await import('./route');
+    const response = await POST(makeRequest(makeBody('kilo-auto/org')) as never);
+
+    expect(response.status).toBe(400);
+    expect(await response.json()).toMatchObject({
+      error_type: 'organization_auto_configuration',
+      message: expect.stringContaining('does not have an enabled BYOK credential for martian'),
+    });
+    expect(mockedUpstreamRequest).not.toHaveBeenCalled();
+  });
+
   it('bills classifier cost when cost > 0 and user is non-BYOK', async () => {
     mockedFetchEfficientAutoDecision.mockResolvedValue({
       decision: {

apps/web/src/app/api/openrouter/[...path]/route.ts

@@ -17,6 +17,7 @@ import type {
 } from '@/lib/ai-gateway/providers/openrouter/types';
 import { applyProviderSpecificLogic } from '@/lib/ai-gateway/providers/apply-provider-specific-logic';
 import { getProvider } from '@/lib/ai-gateway/providers/get-provider';
+import { getDirectByokModel } from '@/lib/ai-gateway/providers/direct-byok';
 import { buildExperimentPromptCapture } from '@/lib/ai-gateway/experiments/persist';
 import { isPublicIdExperimented } from '@/lib/ai-gateway/experiments/membership';
 import { upstreamRequest } from '@/lib/ai-gateway/providers/upstream-request';
@@ -48,6 +49,7 @@ import {
   modelNotAllowedResponse,
   extractHeaderAndLimitLength,
   noFreeModelsAvailableResponse,
+  organizationAutoConfigurationResponse,
   temporarilyUnavailableResponse,
   usageLimitExceededResponse,
   wrapInSafeNextResponse,
@@ -95,6 +97,7 @@ import {
   isKiloAutoModel,
   KILO_AUTO_FREE_MODEL,
   KILO_AUTO_EFFICIENT_MODEL,
+  ORG_AUTO_MODEL,
 } from '@/lib/ai-gateway/auto-model';
 import { applyResolvedAutoModel } from '@/lib/ai-gateway/auto-model/resolution';
 import { fetchEfficientAutoDecision } from '@/lib/ai-gateway/auto-routing-decision';
@@ -243,6 +246,13 @@ export async function POST(request: NextRequest): Promise<NextResponseType<unkno
       ? getBalanceAndOrgSettings(res.organizationId, res.user)
       : { balance: 0, settings: undefined, plan: undefined }
   );
+  const organizationContextPromise = Promise.all([authPromise, balanceAndSettingsPromise]).then(
+    ([auth, balanceAndSettings]) => ({
+      organizationId: auth.organizationId,
+      settings: balanceAndSettings.settings,
+      plan: balanceAndSettings.plan,
+    })
+  );
 
   // Extract IP early (needed for free model routing fallback and rate limiting)
   const ipAddress = request.headers.get('x-forwarded-for')?.split(',')[0]?.trim();
@@ -272,6 +282,7 @@ export async function POST(request: NextRequest): Promise<NextResponseType<unkno
   }
 
   let autoModel: string | null = null;
+  let routingTarget: string | null = null;
   let classifierCostUsd = 0;
   if (isKiloAutoModel(requestedModelLowerCased)) {
     autoModel = requestedModelLowerCased;
@@ -312,6 +323,7 @@ export async function POST(request: NextRequest): Promise<NextResponseType<unkno
         apiKind: requestBodyParsed.kind,
         clientIp: ipAddress ?? null,
         efficientDecision,
+        organizationContext: organizationContextPromise,
       },
       requestBodyParsed,
       authPromise.then(res => res.user),
@@ -320,6 +332,10 @@ export async function POST(request: NextRequest): Promise<NextResponseType<unkno
     if (autoResult.kind === 'no_free_models_available') {
       return noFreeModelsAvailableResponse();
     }
+    if (autoResult.kind === 'organization_auto_configuration_error') {
+      return organizationAutoConfigurationResponse(autoResult.message);
+    }
+    routingTarget = autoResult.routingTarget ?? null;
   }
 
   let effectiveModelIdLowerCased = requestBodyParsed.body.model.toLowerCase();
@@ -348,6 +364,7 @@ export async function POST(request: NextRequest): Promise<NextResponseType<unkno
   const isRateLimitedFreeModelRequest =
     isKiloExclusiveFreeModel(effectiveModelIdLowerCased) ||
     autoModel === KILO_AUTO_FREE_MODEL.id ||
+    routingTarget === KILO_AUTO_FREE_MODEL.id ||
     (await isPublicIdExperimented(effectiveModelIdLowerCased));
   if (isRateLimitedFreeModelRequest) {
     const rateLimit = await resolveRateLimit(feature, ipAddress, authPromise);
@@ -556,6 +573,29 @@ export async function POST(request: NextRequest): Promise<NextResponseType<unkno
   }
   let effectiveProviderContext = initialProviderResultForAbuseService;
 
+  if (autoModel === ORG_AUTO_MODEL.id && routingTarget) {
+    try {
+      const directByokTarget = await getDirectByokModel(routingTarget);
+      if (directByokTarget.provider && effectiveProviderContext.provider.id !== 'direct-byok') {
+        return organizationAutoConfigurationResponse(
+          `Organization Auto route target '${routingTarget}' is unavailable because this organization does not have an enabled BYOK credential for ${directByokTarget.provider.id}.`
+        );
+      }
+    } catch {
+      return organizationAutoConfigurationResponse(
+        'Organization Auto could not validate this route target against the current model catalog.'
+      );
+    }
+  }
+
+  // Request-level data-collection opt-out: a caller can set
+  // `provider.data_collection: 'deny'` or `provider.zdr: true` on any
+  // request to opt that single request out of training/data-retention.
+  // Direct experiment upstreams ignore those OpenRouter/Vercel flags
+  // (we never reach OpenRouter), but we still capture the prompt to R2
+  // for partner evaluation — which violates the caller's stated
+  // intent. Refuse here regardless of org settings, anon/BYOK status,
+  // or the org-level check below.
   if (
     (await hasBestEffortGuessDataCollectionRequirement(effectiveModelIdLowerCased)) &&
     isDataCollectionExplicitlyDisallowed(requestBodyParsed.body.provider)

apps/web/src/app/api/organizations/[id]/defaults/route.test.ts

@@ -164,6 +164,38 @@ describe('GET /api/organizations/[id]/defaults', () => {
     expect(mockedGetEnhancedOpenRouterModels).not.toHaveBeenCalled();
   });
 
+  test('returns Organization Auto when it is configured as the organization default', async () => {
+    const user = await insertTestUser();
+    const organization = await createOrganization('Test Org', user.id);
+
+    mockedGetEnhancedOpenRouterModels.mockRejectedValue(new Error('should not be called'));
+    mockedGetAuthorizedOrgContext.mockResolvedValue({
+      success: true,
+      data: {
+        user: { ...user, role: 'owner' },
+        organization: {
+          ...organization,
+          plan: 'enterprise' as const,
+          settings: {
+            default_model: 'kilo-auto/org',
+            org_auto_model: {
+              routes: {},
+              fallback_model: 'kilo-auto/balanced',
+            },
+          },
+        },
+      },
+    });
+
+    const response = await GET(new NextRequest('http://localhost:3000'), {
+      params: Promise.resolve({ id: organization.id }),
+    });
+
+    expect(response.status).toBe(200);
+    const body = await response.json();
+    expect(body.defaultModel).toBe('kilo-auto/org');
+  });
+
   test('returns 409 when all available models are blocked by policy', async () => {
     const user = await insertTestUser();
     const organization = await createOrganization('Test Org', user.id);

apps/web/src/app/api/organizations/[id]/defaults/route.ts

@@ -8,8 +8,9 @@ import {
   hasActiveModelRestrictions,
 } from '@/lib/model-allow.server';
 import { getModelIdToProviderSlugsIndex } from '@/lib/ai-gateway/providers/openrouter/models-by-provider-index.server';
-import { KILO_AUTO_FREE_MODEL } from '@/lib/ai-gateway/auto-model';
+import { KILO_AUTO_FREE_MODEL, ORG_AUTO_MODEL } from '@/lib/ai-gateway/auto-model';
 import { getEffectiveModelRestrictions } from '@/lib/organizations/model-restrictions';
+import { isOrganizationAutoConfigured } from '@/lib/organizations/organization-auto-model';
 
 type DefaultsResponse = {
   defaultModel: string;
@@ -67,8 +68,17 @@ export async function GET(
     return undefined;
   };
 
-  // If organization has a default model set, validate it against allowed models
-  if (defaultModel && (defaultModel.endsWith('/*') || !(await isAllowed(defaultModel)))) {
+  // If organization has a default model set, validate it against allowed models.
+  // Organization Auto is a virtual organization-only default, so its eligibility
+  // is validated from persisted organization settings rather than provider policy.
+  if (defaultModel === ORG_AUTO_MODEL.id && !isOrganizationAutoConfigured(organization)) {
+    console.warn('organization_auto_invalid_default', { organizationId: organization.id });
+    defaultModel = undefined;
+  } else if (
+    defaultModel &&
+    defaultModel !== ORG_AUTO_MODEL.id &&
+    (defaultModel.endsWith('/*') || !(await isAllowed(defaultModel)))
+  ) {
     // Organization's configured default model is not permitted; fall back to a safe default.
     defaultModel = undefined;
   }

apps/web/src/app/api/organizations/[id]/modes/route.test.ts

@@ -11,7 +11,7 @@ const mockedGetAuthorizedOrgContext = jest.mocked(getAuthorizedOrgContext);
 const mockedGetAllOrganizationModes = jest.mocked(getAllOrganizationModes);
 
 describe('GET /api/organizations/[id]/modes', () => {
-  test('returns defaultModel as part of the additive mode payload', async () => {
+  test('returns the direct mode payload without Organization Auto route projection', async () => {
     mockedGetAuthorizedOrgContext.mockResolvedValue({
       success: true,
       data: {
@@ -30,7 +30,6 @@ describe('GET /api/organizations/[id]/modes', () => {
         config: {
           roleDefinition: 'You are a coding assistant',
           groups: ['read'],
-          defaultModel: 'openai/gpt-4o',
         },
       },
     ]);
@@ -53,57 +52,10 @@ describe('GET /api/organizations/[id]/modes', () => {
           config: {
             roleDefinition: 'You are a coding assistant',
             groups: ['read'],
-            defaultModel: 'openai/gpt-4o',
           },
         },
       ],
     });
     expect(mockedGetAllOrganizationModes).toHaveBeenCalledWith('org-1');
   });
-
-  test('returns a legacy mode row without defaultModel unchanged', async () => {
-    mockedGetAuthorizedOrgContext.mockResolvedValue({
-      success: true,
-      data: {
-        organization: { id: 'org-1' },
-      },
-    } as never);
-    mockedGetAllOrganizationModes.mockResolvedValue([
-      {
-        id: 'mode-1',
-        organization_id: 'org-1',
-        name: 'Code',
-        slug: 'code',
-        created_by: 'user-1',
-        created_at: '2026-01-01T00:00:00.000Z',
-        updated_at: '2026-01-01T00:00:00.000Z',
-        config: {
-          roleDefinition: 'You are a coding assistant',
-          groups: ['read'],
-        },
-      },
-    ]);
-
-    const response = await GET(new NextRequest('http://localhost:3000'), {
-      params: Promise.resolve({ id: 'org-1' }),
-    });
-
-    await expect(response.json()).resolves.toEqual({
-      modes: [
-        {
-          id: 'mode-1',
-          organization_id: 'org-1',
-          name: 'Code',
-          slug: 'code',
-          created_by: 'user-1',
-          created_at: '2026-01-01T00:00:00.000Z',
-          updated_at: '2026-01-01T00:00:00.000Z',
-          config: {
-            roleDefinition: 'You are a coding assistant',
-            groups: ['read'],
-          },
-        },
-      ],
-    });
-  });
 });

apps/web/src/app/api/organizations/[id]/modes/route.ts

@@ -15,9 +15,8 @@ export async function GET(
   }
 
   const { organization } = data;
-  const modes = await getAllOrganizationModes(organization.id);
 
   return NextResponse.json({
-    modes,
+    modes: await getAllOrganizationModes(organization.id),
   });
 }