Refactor AI gateway provider definitions

ENVIRONMENT.md

@@ -264,11 +264,11 @@ Manage shared web env var additions and rotations with `pnpm web:env set <VARIAB
 
 ### AI Providers
 
-- `OPENROUTER_API_KEY` - Primary OpenRouter API key for model inference through the AI gateway; provider definition in `apps/web/src/lib/ai-gateway/providers/provider-definitions.ts` pointing to `https://openrouter.ai/api/v1`. `[SECRET]`
+- `OPENROUTER_API_KEY` - Primary OpenRouter API key for model inference through the AI gateway; provider definition in `apps/web/src/lib/ai-gateway/providers/definitions/openrouter.ts` pointing to `https://openrouter.ai/api/v1`. `[SECRET]`
 - `OPENAI_API_KEY` - OpenAI API key supplied as a managed BYOK credential when managed inference requests route through the Vercel AI Gateway and permit the OpenAI provider. `[SECRET]`
 - `MISTRAL_API_KEY` - Mistral API key; used in `apps/web/src/lib/ai-gateway/embeddings/embedding-providers.ts` for `codestral-embed-2505` and `mistral-embed` embeddings, in the FIM completions proxy at `apps/web/src/app/api/fim/completions/route.ts` (routes Mistral Codestral vs. La Plateforme keys), and as a provider config in `apps/web/src/lib/config.server.ts`. `[SECRET]`
-- `LONGCAT_API_KEY` - LongCat API key for model inference through the AI gateway; provider definition in `apps/web/src/lib/ai-gateway/providers/provider-definitions.ts`. `[SECRET]`
-- `STREAMLAKE_API_KEY` - StreamLake API key for model inference through the AI gateway; provider definition in `apps/web/src/lib/ai-gateway/providers/provider-definitions.ts`. `[SECRET]`
+- `LONGCAT_API_KEY` - LongCat API key for model inference through the AI gateway; provider definition in `apps/web/src/lib/ai-gateway/providers/definitions/longcat.ts`. `[SECRET]`
+- `STREAMLAKE_API_KEY` - StreamLake API key for model inference through the AI gateway; provider definition in `apps/web/src/lib/ai-gateway/providers/definitions/streamlake.ts`. `[SECRET]`
 - `PERPLEXITY_API_KEY` - Perplexity API key for percentage-routed Kimi K3 inference through the AI gateway; provider definition in `apps/web/src/lib/ai-gateway/providers/partner/providers.ts`. `[SECRET]`
 - `INCEPTION_API_KEY` - Inception Labs API key; used in `apps/web/src/app/api/fim/completions/route.ts` and `apps/web/src/app/api/edit/completions/route.ts` as a fill-in-the-middle (FIM) provider, with endpoint `https://api.inceptionlabs.ai/v1/fim/completions`. Defined in `apps/web/src/lib/config.server.ts`. `[SECRET]`
 - `AI_ATTRIBUTION_ADMIN_SECRET` - Admin secret for the AI Attribution service (`apps/web/src/lib/ai-attribution-service.ts`); sent as `X-Admin-Secret` header. `[SECRET]`

apps/web/src/lib/ai-gateway/auto-model/resolution.ts

@@ -30,7 +30,7 @@ import {
   selectAutoFreeCandidate,
 } from '@/lib/ai-gateway/models';
 import { getOpenRouterModelsFromDatabase } from '@/lib/ai-gateway/providers/gateway-models-cache';
-import { tryGetProviderById } from '@/lib/ai-gateway/providers/provider-definitions';
+import { tryGetProviderById } from '@/lib/ai-gateway/providers/definitions/try-get-provider-by-id';
 import type { ProviderId } from '@/lib/ai-gateway/providers/types';
 import {
   getOrganizationAutoRoute,

apps/web/src/lib/ai-gateway/local-fake-llm.ts

@@ -1,5 +1,5 @@
 import { buildDirectProvider } from '@/lib/ai-gateway/experiments/build-direct-provider';
-import { OPENROUTER } from '@/lib/ai-gateway/providers/openrouter-definition';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
 import type { Provider } from '@/lib/ai-gateway/providers/types';
 import type { OpenRouterModel } from '@/lib/organizations/organization-types';
 

apps/web/src/lib/ai-gateway/model-api-kinds.ts

@@ -1,5 +1,6 @@
 import { findKiloExclusiveModel } from '@/lib/ai-gateway/models';
-import { OPENROUTER, tryGetProviderById } from '@/lib/ai-gateway/providers/provider-definitions';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
+import { tryGetProviderById } from '@/lib/ai-gateway/providers/definitions/try-get-provider-by-id';
 import type { GatewayChatApiKind } from '@/lib/ai-gateway/providers/types';
 
 const GATEWAY_CHAT_API_KINDS: readonly GatewayChatApiKind[] = [

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

@@ -16,7 +16,8 @@ import type {
   OpenRouterGeneration,
 } from './providers/openrouter/types';
 import { fetchGeneration } from './providers/upstream-request';
-import { OPENROUTER, VERCEL_AI_GATEWAY } from './providers/provider-definitions';
+import { OPENROUTER } from './providers/definitions/openrouter';
+import { VERCEL_AI_GATEWAY } from './providers/definitions/vercel';
 import { toMicrodollars } from '../utils';
 import { captureException, captureMessage, startSpan, startInactiveSpan } from '@sentry/nextjs';
 import type { Span } from '@sentry/nextjs';

apps/web/src/lib/ai-gateway/providerHash.test.ts

@@ -1,4 +1,5 @@
-import { OPENROUTER, VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/provider-definitions';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
+import { VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/definitions/vercel';
 import type { GatewayRequest } from '@/lib/ai-gateway/providers/openrouter/types';
 import {
   applyTrackingIds,

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

@@ -1,6 +1,7 @@
 import crypto from 'crypto';
 import { type Provider } from '@/lib/ai-gateway/providers/types';
-import { OPENROUTER, VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/provider-definitions';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
+import { VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/definitions/vercel';
 import { getEnvVariable } from '@/lib/dotenvx';
 import type { GatewayRequest } from '@/lib/ai-gateway/providers/openrouter/types';
 

apps/web/src/lib/ai-gateway/providers/definitions/alibaba.ts

@@ -0,0 +1,19 @@
+import { getEnvVariable } from '@/lib/dotenvx';
+import { isReasoningExplicitlyDisabled } from '@/lib/ai-gateway/providers/openrouter/request-helpers';
+import type { Provider } from '@/lib/ai-gateway/providers/types';
+
+export const ALIBABA = {
+  id: 'alibaba',
+  apiUrl: 'https://dashscope-intl.aliyuncs.com/compatible-mode/v1',
+  apiUrlOverrides: {},
+  apiKey: getEnvVariable('ALIBABA_API_KEY'),
+  apiKeyHeader: null,
+  supportedChatApis: [
+    'chat_completions',
+    // 'responses', // supported, not tested
+  ],
+  responseTransforms: null,
+  async transformRequest(context) {
+    context.request.body.enable_thinking = !isReasoningExplicitlyDisabled(context.request);
+  },
+} as const satisfies Provider;

apps/web/src/lib/ai-gateway/providers/definitions/longcat.ts

@@ -0,0 +1,22 @@
+import { getEnvVariable } from '@/lib/dotenvx';
+import { isReasoningExplicitlyDisabled } from '@/lib/ai-gateway/providers/openrouter/request-helpers';
+import { ReasoningDetailsTransform, type Provider } from '@/lib/ai-gateway/providers/types';
+
+export const LONGCAT = {
+  id: 'longcat',
+  apiUrl: 'https://api.longcat.ai/openai/v1',
+  apiUrlOverrides: {},
+  apiKey: getEnvVariable('LONGCAT_API_KEY'),
+  apiKeyHeader: null,
+  supportedChatApis: ['chat_completions'],
+  responseTransforms: ReasoningDetailsTransform.ReasoningContent,
+  async transformRequest(context) {
+    context.request.body.thinking = {
+      type: isReasoningExplicitlyDisabled(context.request) ? 'disabled' : 'enabled',
+    };
+    delete context.request.body.provider;
+    if (context.request.body.user) {
+      context.extraHeaders['Mt-User-Id'] = context.request.body.user;
+    }
+  },
+} as const satisfies Provider;

apps/web/src/lib/ai-gateway/providers/definitions/martian.ts

@@ -0,0 +1,15 @@
+import { getEnvVariable } from '@/lib/dotenvx';
+import type { Provider } from '@/lib/ai-gateway/providers/types';
+
+export const MARTIAN = {
+  id: 'martian',
+  apiUrl: 'https://api.withmartian.com/v1',
+  apiUrlOverrides: {},
+  apiKey: getEnvVariable('MARTIAN_API_KEY'),
+  apiKeyHeader: null,
+  supportedChatApis: ['chat_completions', 'responses', 'messages'],
+  responseTransforms: null,
+  async transformRequest(context) {
+    delete context.request.body.provider;
+  },
+} as const satisfies Provider;

apps/web/src/lib/ai-gateway/providers/definitions/mistral.ts

@@ -0,0 +1,13 @@
+import { getEnvVariable } from '@/lib/dotenvx';
+import type { Provider } from '@/lib/ai-gateway/providers/types';
+
+export const MISTRAL = {
+  id: 'mistral',
+  apiUrl: 'https://api.mistral.ai/v1',
+  apiUrlOverrides: {},
+  apiKey: getEnvVariable('MISTRAL_API_KEY'),
+  apiKeyHeader: null,
+  supportedChatApis: [],
+  responseTransforms: null,
+  async transformRequest() {},
+} as const satisfies Provider;

apps/web/src/lib/ai-gateway/providers/openrouter-definition.ts → apps/web/src/lib/ai-gateway/providers/definitions/openrouter.ts

@@ -1,11 +1,6 @@
 import { getEnvVariable } from '@/lib/dotenvx';
 import type { Provider } from '@/lib/ai-gateway/providers/types';
 
-/**
- * Self-contained OpenRouter provider definition. Kept in a leaf module so
- * importers that only need `OPENROUTER` (for example the local fake LLM) do
- * not pull the rest of the provider graph into a circular dependency.
- */
 export const OPENROUTER = {
   id: 'openrouter',
   apiUrl: 'https://openrouter.ai/api/v1',

apps/web/src/lib/ai-gateway/providers/definitions/seed.ts

@@ -0,0 +1,32 @@
+import { getEnvVariable } from '@/lib/dotenvx';
+import { isReasoningExplicitlyDisabled } from '@/lib/ai-gateway/providers/openrouter/request-helpers';
+import type { Provider } from '@/lib/ai-gateway/providers/types';
+
+export const SEED = {
+  id: 'seed',
+  apiUrl: 'https://ark.ap-southeast.bytepluses.com/api/v3',
+  apiUrlOverrides: {},
+  apiKey: getEnvVariable('BYTEDANCE_API_KEY'),
+  apiKeyHeader: null,
+  supportedChatApis: [
+    'chat_completions',
+    // 'responses', // supported, not tested
+  ],
+  responseTransforms: null,
+  async transformRequest(context) {
+    if (!isReasoningExplicitlyDisabled(context.request)) {
+      context.request.body.thinking = { type: 'enabled' };
+      if (context.request.kind === 'chat_completions') {
+        context.request.body.reasoning_effort ??= context.request.body.reasoning?.effort;
+      }
+    } else {
+      context.request.body.thinking = { type: 'disabled' };
+    }
+    if (context.request.kind === 'responses') {
+      delete context.request.body.prompt_cache_key;
+      delete context.request.body.safety_identifier;
+      delete context.request.body.user;
+      delete context.request.body.provider;
+    }
+  },
+} as const satisfies Provider;

apps/web/src/lib/ai-gateway/providers/definitions/streamlake.ts

@@ -0,0 +1,15 @@
+import { getEnvVariable } from '@/lib/dotenvx';
+import type { Provider } from '@/lib/ai-gateway/providers/types';
+
+export const STREAMLAKE = {
+  id: 'streamlake',
+  apiUrl: 'https://vanchin.streamlake.ai/api/gateway/v1/endpoints',
+  apiUrlOverrides: {},
+  apiKey: getEnvVariable('STREAMLAKE_API_KEY'),
+  apiKeyHeader: null,
+  supportedChatApis: ['chat_completions'],
+  responseTransforms: null,
+  async transformRequest(context) {
+    delete context.request.body.provider;
+  },
+} as const satisfies Provider;

apps/web/src/lib/ai-gateway/providers/definitions/try-get-provider-by-id.ts

@@ -0,0 +1,32 @@
+import { ALIBABA } from '@/lib/ai-gateway/providers/definitions/alibaba';
+import { LONGCAT } from '@/lib/ai-gateway/providers/definitions/longcat';
+import { MARTIAN } from '@/lib/ai-gateway/providers/definitions/martian';
+import { MISTRAL } from '@/lib/ai-gateway/providers/definitions/mistral';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
+import { SEED } from '@/lib/ai-gateway/providers/definitions/seed';
+import { STREAMLAKE } from '@/lib/ai-gateway/providers/definitions/streamlake';
+import { VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/definitions/vercel';
+import type { Provider, ProviderId } from '@/lib/ai-gateway/providers/types';
+
+export function tryGetProviderById(providerId: ProviderId): Provider | undefined {
+  switch (providerId) {
+    case 'openrouter':
+      return OPENROUTER;
+    case 'alibaba':
+      return ALIBABA;
+    case 'seed':
+      return SEED;
+    case 'longcat':
+      return LONGCAT;
+    case 'martian':
+      return MARTIAN;
+    case 'mistral':
+      return MISTRAL;
+    case 'streamlake':
+      return STREAMLAKE;
+    case 'vercel':
+      return VERCEL_AI_GATEWAY;
+    default:
+      return undefined;
+  }
+}

apps/web/src/lib/ai-gateway/providers/definitions/vercel.ts

@@ -0,0 +1,16 @@
+import { getEnvVariable } from '@/lib/dotenvx';
+import type { Provider } from '@/lib/ai-gateway/providers/types';
+import { applyVercelSettings } from '@/lib/ai-gateway/providers/vercel';
+
+export const VERCEL_AI_GATEWAY = {
+  id: 'vercel',
+  apiUrl: 'https://ai-gateway.vercel.sh/v1',
+  apiUrlOverrides: {},
+  apiKey: getEnvVariable('VERCEL_AI_GATEWAY_API_KEY'),
+  apiKeyHeader: null,
+  supportedChatApis: ['chat_completions', 'messages', 'responses'],
+  responseTransforms: null,
+  async transformRequest(context) {
+    await applyVercelSettings(context.model, context.request, context.userByok);
+  },
+} as const satisfies Provider;

apps/web/src/lib/ai-gateway/providers/get-provider.local-fake.test.ts

@@ -1,6 +1,7 @@
 import { afterEach, beforeEach, describe, expect, test } from '@jest/globals';
 import { getProvider, getTranscriptionProvider } from '@/lib/ai-gateway/providers/get-provider';
-import { OPENROUTER, VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/provider-definitions';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
+import { VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/definitions/vercel';
 import { shouldRouteToVercel } from '@/lib/ai-gateway/providers/vercel';
 import type { GatewayRequest } from '@/lib/ai-gateway/providers/openrouter/types';
 import type { User } from '@kilocode/db/schema';

apps/web/src/lib/ai-gateway/providers/get-provider.ts

@@ -16,11 +16,9 @@ import { eq } from 'drizzle-orm';
 import type { AnonymousUserContext } from '@/lib/anonymous';
 import { isAnonymousContext } from '@/lib/anonymous';
 import type { BYOKResult, Provider } from '@/lib/ai-gateway/providers/types';
-import {
-  OPENROUTER,
-  tryGetProviderById,
-  VERCEL_AI_GATEWAY,
-} from '@/lib/ai-gateway/providers/provider-definitions';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
+import { tryGetProviderById } from '@/lib/ai-gateway/providers/definitions/try-get-provider-by-id';
+import { VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/definitions/vercel';
 import { getDirectByokModel } from '@/lib/ai-gateway/providers/direct-byok';
 import { CustomLlmCredentialsSchema, CustomLlmDefinitionSchema } from '@kilocode/db/schema-types';
 import { buildDirectProvider } from '@/lib/ai-gateway/experiments/build-direct-provider';

apps/web/src/lib/ai-gateway/providers/getEmbeddingProvider.test.ts

@@ -3,7 +3,8 @@ import {
   getEmbeddingProvider,
   getTranscriptionProvider,
 } from '@/lib/ai-gateway/providers/get-provider';
-import { OPENROUTER, VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/provider-definitions';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
+import { VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/definitions/vercel';
 import { createAnonymousContext } from '@/lib/anonymous';
 import {
   getModelUserByokProviders,

apps/web/src/lib/ai-gateway/providers/openrouter/index.ts

@@ -8,7 +8,7 @@ import {
   getLocalFakeTranscriptionModelsUrl,
   LOCAL_FAKE_LLM_API_KEY,
 } from '@/lib/ai-gateway/local-fake-llm';
-import { OPENROUTER } from '@/lib/ai-gateway/providers/provider-definitions';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
 import type { OpenRouterModel } from '@/lib/organizations/organization-types';
 import {
   OpenRouterModelsResponseSchema,

apps/web/src/lib/ai-gateway/providers/openrouter/sync-providers.ts

@@ -17,7 +17,8 @@ import { ai_gateway_sync_providers_state, modelsByProvider } from '@kilocode/db/
 import { db } from '@/lib/drizzle';
 import { desc, eq, lt, sql } from 'drizzle-orm';
 import { captureException } from '@sentry/nextjs';
-import { OPENROUTER, VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/provider-definitions';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
+import { VERCEL_AI_GATEWAY } from '@/lib/ai-gateway/providers/definitions/vercel';
 import { logAutoModelChangesForAllOrgs } from '@/lib/organizations/auto-model-change-log';
 import type { Provider } from '@/lib/ai-gateway/providers/types';
 import type { StoredModel } from '@kilocode/db/schema-types';

apps/web/src/lib/ai-gateway/providers/provider-definitions.test.ts

@@ -1,10 +1,8 @@
 import { describe, expect, test } from '@jest/globals';
 
-import {
-  LONGCAT,
-  OPENROUTER,
-  tryGetProviderById,
-} from '@/lib/ai-gateway/providers/provider-definitions';
+import { LONGCAT } from '@/lib/ai-gateway/providers/definitions/longcat';
+import { OPENROUTER } from '@/lib/ai-gateway/providers/definitions/openrouter';
+import { tryGetProviderById } from '@/lib/ai-gateway/providers/definitions/try-get-provider-by-id';
 import type { GatewayRequest } from '@/lib/ai-gateway/providers/openrouter/types';
 import {
   ReasoningDetailsTransform,