feat(fim): add Mercury Edit FIM support with provider-based routing

src/app/api/fim/completions/route.ts

@@ -1,4 +1,4 @@
-import { MISTRAL_API_KEY } from '@/lib/config.server';
+import { MISTRAL_API_KEY, INCEPTION_API_KEY } from '@/lib/config.server';
 import type { NextRequest } from 'next/server';
 import { NextResponse } from 'next/server';
 import z from 'zod';
@@ -25,12 +25,45 @@ import { debugSaveProxyRequest } from '@/lib/debugUtils';
 import { sentryLogger } from '@/lib/utils.server';
 import { getBYOKforOrganization, getBYOKforUser } from '@/lib/byok';
 
-const MISTRAL_URL = 'https://api.mistral.ai/v1/fim/completions';
+const MISTRAL_FIM_URL = 'https://api.mistral.ai/v1/fim/completions';
+const INCEPTION_FIM_URL = 'https://api.inceptionlabs.ai/v1/fim/completions';
 const FIM_MAX_TOKENS_LIMIT = 1000;
 
+type FimProvider = 'mistral' | 'inception';
+
+function resolveFimProvider(model: string): {
+  provider: FimProvider;
+  upstreamModel: string;
+  upstreamUrl: string;
+} | null {
+  if (model.startsWith('mistralai/')) {
+    return {
+      provider: 'mistral',
+      upstreamModel: model.slice('mistralai/'.length),
+      upstreamUrl: MISTRAL_FIM_URL,
+    };
+  }
+  if (model === 'inception/mercury-edit') {
+    return {
+      provider: 'inception',
+      upstreamModel: 'mercury-edit',
+      upstreamUrl: INCEPTION_FIM_URL,
+    };
+  }
+  return null;
+}
+
+function getSystemApiKey(provider: FimProvider): string | null {
+  switch (provider) {
+    case 'mistral':
+      return MISTRAL_API_KEY || null;
+    case 'inception':
+      return INCEPTION_API_KEY || null;
+  }
+}
+
 const FIMRequestBody = z.object({
   //ref: https://docs.mistral.ai/api/endpoint/fim#operation-fim_completion_v1_fim_completions_post
-  provider: z.enum(['mistral', 'inceptionlabs']).optional(),
   model: z.string(),
   prompt: z.string(),
   suffix: z.string().optional(),
@@ -82,17 +115,15 @@ export async function POST(request: NextRequest) {
     return invalidRequestResponse();
   }
 
-  if ((requestBody.provider ?? 'mistral') !== 'mistral') {
+  // Resolve provider from model name
+  const resolved = resolveFimProvider(requestBody.model);
+  if (!resolved) {
     return NextResponse.json(
-      { error: requestBody.provider + ' provider not yet supported' },
+      { error: requestBody.model + ' is not a supported FIM model' },
       { status: 400 }
     );
-    //NOTE: mistral does not do data collection on paid org accounts like ours.
-    //If we ever support OTHER providers, we need to either ensure they don't
-    //either, or at least enforce the rules the org settings configure
-    //see getBalanceAndOrgSettings below and its usage in the openrouter proxy.
-    //ref: https://help.mistral.ai/en/articles/347617-do-you-use-my-user-data-to-train-your-artificial-intelligence-models
   }
+  const { provider: fimProvider, upstreamModel, upstreamUrl } = resolved;
 
   // Validate max_tokens
   if (!requestBody.max_tokens || requestBody.max_tokens > FIM_MAX_TOKENS_LIMIT) {
@@ -102,35 +133,24 @@ export async function POST(request: NextRequest) {
     return temporarilyUnavailableResponse();
   }
 
-  // Map FIM model to OpenRouter format for org settings compatibility
-  const fimModel_withOpenRouterStyleProviderPrefix = requestBody.model;
-
-  const requiredModelPrefix = 'mistralai/';
-  if (!fimModel_withOpenRouterStyleProviderPrefix.startsWith(requiredModelPrefix)) {
-    return NextResponse.json(
-      { error: fimModel_withOpenRouterStyleProviderPrefix + ' is not a mistralai model' },
-      { status: 400 }
-    );
-  }
-
-  const mistralModel = fimModel_withOpenRouterStyleProviderPrefix.slice(requiredModelPrefix.length);
-
   // Use new shared helper for fraud & project headers
   const { fraudHeaders, projectId } = extractFraudAndProjectHeaders(request);
   const taskId = extractHeaderAndLimitLength(request, 'x-kilocode-taskid') ?? undefined;
 
   // Extract properties for usage context
   const promptInfo = extractFimPromptInfo(requestBody);
 
+  const byokProviderKey = fimProvider === 'mistral' ? 'codestral' : 'inception';
+
   const userByok = organizationId
-    ? await getBYOKforOrganization(readDb, organizationId, ['codestral'])
-    : await getBYOKforUser(readDb, user.id, ['codestral']);
+    ? await getBYOKforOrganization(readDb, organizationId, [byokProviderKey])
+    : await getBYOKforUser(readDb, user.id, [byokProviderKey]);
 
   const usageContext: MicrodollarUsageContext = {
     api_kind: 'fim_completions',
     kiloUserId: user.id,
-    provider: 'mistral',
-    requested_model: fimModel_withOpenRouterStyleProviderPrefix,
+    provider: fimProvider,
+    requested_model: requestBody.model,
     promptInfo,
     max_tokens: requestBody.max_tokens ?? null,
     has_middle_out_transform: null, // N/A for FIM
@@ -151,44 +171,57 @@ export async function POST(request: NextRequest) {
     auto_model: null,
   };
 
-  setTag('ui.ai_model', fimModel_withOpenRouterStyleProviderPrefix);
+  setTag('ui.ai_model', requestBody.model);
   // Use read replica for balance check - this is a read-only operation that can tolerate
   // slight replication lag, and provides lower latency for US users
   const { balance, settings, plan } = await getBalanceAndOrgSettings(organizationId, user, readDb);
 
-  if (balance <= 0 && !isFreeModel(fimModel_withOpenRouterStyleProviderPrefix) && !userByok) {
+  if (balance <= 0 && !isFreeModel(requestBody.model) && !userByok) {
     return NextResponse.json({ error: { message: 'Insufficient credits' } }, { status: 402 });
   }
 
   // Use shared helper for organization model restrictions
   // Model allow list only applies to Enterprise plans
   // Provider allow list applies to Enterprise plans; data collection applies to all plans (but FIM doesn't use provider config)
   const { error: modelRestrictionError } = checkOrganizationModelRestrictions({
-    modelId: fimModel_withOpenRouterStyleProviderPrefix,
+    modelId: requestBody.model,
     settings,
     organizationPlan: plan,
   });
   if (modelRestrictionError) return modelRestrictionError;
 
+  const systemKey = getSystemApiKey(fimProvider);
+  const apiKey = userByok?.at(0)?.decryptedAPIKey ?? systemKey;
+
+  if (!apiKey) {
+    return NextResponse.json(
+      {
+        error: 'This model requires a BYOK API key. Please configure your API key in settings.',
+      },
+      { status: 400 }
+    );
+  }
+
   sentryRootSpan()?.setAttribute(
-    'mistral-fim.time_to_request_start_ms',
+    'fim.time_to_request_start_ms',
     performance.now() - requestStartedAt
   );
 
-  const mistralRequestSpan = startInactiveSpan({
-    name: 'mistral-fim-request-start',
+  const fimRequestSpan = startInactiveSpan({
+    name: 'fim-request-start',
     op: 'http.client',
   });
 
-  const bodyWithCorrectedModel = { ...requestBody, model: mistralModel };
-  // Make upstream request to Mistral
-  const proxyRes = await fetch(MISTRAL_URL, {
+  const bodyForUpstream = { ...requestBody, model: upstreamModel };
+
+  // Make upstream request to the resolved provider
+  const proxyRes = await fetch(upstreamUrl, {
     method: 'POST',
     headers: {
       'Content-Type': 'application/json',
-      Authorization: `Bearer ${userByok?.at(0)?.decryptedAPIKey ?? MISTRAL_API_KEY}`,
+      Authorization: `Bearer ${apiKey}`,
     },
-    body: JSON.stringify(bodyWithCorrectedModel),
+    body: JSON.stringify(bodyForUpstream),
   });
   usageContext.status_code = proxyRes.status;
 
@@ -200,19 +233,19 @@ export async function POST(request: NextRequest) {
   if (proxyRes.status >= 400) {
     await captureProxyError({
       user,
-      request: bodyWithCorrectedModel,
+      request: bodyForUpstream,
       response: proxyRes,
       organizationId,
-      model: fimModel_withOpenRouterStyleProviderPrefix,
-      errorMessage: `Mistral FIM returned error ${proxyRes.status}`,
+      model: requestBody.model,
+      errorMessage: `FIM provider returned error ${proxyRes.status}`,
       trackInSentry: proxyRes.status >= 500,
     });
   }
 
   const clonedResponse = proxyRes.clone(); // reading from body is side-effectful
 
   // Account for usage using FIM-specific parser
-  countAndStoreFimUsage(clonedResponse, usageContext, mistralRequestSpan);
+  countAndStoreFimUsage(clonedResponse, usageContext, fimRequestSpan);
 
   return wrapInSafeNextResponse(proxyRes);
 }

src/lib/config.server.ts

@@ -28,6 +28,7 @@ export const NEXTAUTH_SECRET = getEnvVariable('NEXTAUTH_SECRET');
 export const OPENROUTER_API_KEY = getEnvVariable('OPENROUTER_API_KEY');
 export const MISTRAL_API_KEY = getEnvVariable('MISTRAL_API_KEY');
 export const OPENAI_API_KEY = getEnvVariable('OPENAI_API_KEY');
+export const INCEPTION_API_KEY = getEnvVariable('INCEPTION_API_KEY');
 export const INTERNAL_API_SECRET = getEnvVariable('INTERNAL_API_SECRET');
 export const CODE_REVIEW_WORKER_AUTH_TOKEN = getEnvVariable('CODE_REVIEW_WORKER_AUTH_TOKEN');
 

src/lib/llm-proxy-helpers.ts

@@ -32,7 +32,7 @@ import type {
 } from '@/lib/processUsage.types';
 import { getMaxTokens } from '@/lib/providers/openrouter/request-helpers';
 import { KILO_AUTO_BALANCED_MODEL, KILO_AUTO_FREE_MODEL } from '@/lib/kilo-auto-model';
-import type { GatewayChatApiKind } from '@/lib/providers/types';
+import type { GatewayChatApiKind, ProviderId } from '@/lib/providers/types';
 
 // FIM suffix markers for tracking purposes - used to wrap suffix in a fake system prompt format
 // This allows FIM requests to be tracked consistently with chat requests
@@ -378,7 +378,7 @@ export function extractFimPromptInfo(body: { prompt: string; suffix?: string | n
 // FIM-Specific Code
 // ============================================================================
 
-export type MistralFimUsage = {
+export type FimUsage = {
   prompt_tokens: number;
   completion_tokens: number;
   total_tokens: number;
@@ -388,7 +388,7 @@ export type MistralFimCompletion = {
   id: string;
   object: 'fim.completion';
   model: string;
-  usage: MistralFimUsage;
+  usage: FimUsage;
   created: number;
   choices: Array<{
     index: number;
@@ -409,23 +409,38 @@ export type MistralFimStreamChunk = {
     };
     finish_reason: string | null;
   }>;
-  usage?: MistralFimUsage; // Only present in final chunk
+  usage?: FimUsage; // Only present in final chunk
 };
 
-function computeMistralFimMicrodollarCost(usage: MistralFimUsage): number {
-  return Math.round(usage.prompt_tokens * 0.3 + usage.completion_tokens * 0.9);
+function computeInceptionFimMicrodollarCost(usage: FimUsage): number {
+  return Math.round(usage.prompt_tokens * 0.25 + usage.completion_tokens * 0.75);
 }
 
-function parseMistralFimUsageFromString(response: string): MicrodollarUsageStats {
+function computeFimMicrodollarCost(usage: FimUsage, provider: ProviderId): number {
+  switch (provider) {
+    case 'mistral':
+      return Math.round(usage.prompt_tokens * 0.3 + usage.completion_tokens * 0.9);
+    case 'inception':
+      return computeInceptionFimMicrodollarCost(usage);
+    default:
+      console.error('Unknown provider for FIM cost calculation', provider);
+      return 0;
+  }
+}
+
+function parseMistralFimUsageFromString(
+  response: string,
+  provider: ProviderId
+): MicrodollarUsageStats {
   const json: MistralFimCompletion = JSON.parse(response);
-  const cost_mUsd = computeMistralFimMicrodollarCost(json.usage);
+  const cost_mUsd = computeFimMicrodollarCost(json.usage, provider);
 
   return {
     messageId: json.id,
     model: json.model,
     responseContent: json.choices[0]?.text || '',
     hasError: !json.model,
-    inference_provider: 'mistral',
+    inference_provider: provider,
     inputTokens: json.usage.prompt_tokens,
     outputTokens: json.usage.completion_tokens,
     cacheHitTokens: 0,
@@ -444,11 +459,12 @@ function parseMistralFimUsageFromString(response: string): MicrodollarUsageStats
 
 async function parseMistralFimUsageFromStream(
   stream: ReadableStream,
-  requestSpan: Span | undefined
+  requestSpan: Span | undefined,
+  provider: ProviderId
 ): Promise<MicrodollarUsageStats> {
   requestSpan?.end();
   const streamProcessingSpan = startInactiveSpan({
-    name: 'mistral-fim-stream-processing',
+    name: 'fim-stream-processing',
     op: 'performance',
   });
   const timeToFirstTokenSpan = startInactiveSpan({
@@ -462,18 +478,15 @@ async function parseMistralFimUsageFromStream(
   let reportedError = false;
   const startedAt = performance.now();
   let firstTokenReceived = false;
-  let usage: MistralFimUsage | undefined;
+  let usage: FimUsage | undefined;
 
   const reader = stream.getReader();
   const decoder = new TextDecoder();
 
   const sseStreamParser = createParser({
     onEvent(event: EventSourceMessage) {
       if (!firstTokenReceived) {
-        sentryRootSpan()?.setAttribute(
-          'mistral.time_to_first_token_ms',
-          performance.now() - startedAt
-        );
+        sentryRootSpan()?.setAttribute('fim.time_to_first_token_ms', performance.now() - startedAt);
         firstTokenReceived = true;
         timeToFirstTokenSpan.end();
       }
@@ -525,12 +538,12 @@ async function parseMistralFimUsageFromStream(
     model,
     responseContent,
     hasError: reportedError,
-    inference_provider: 'mistral',
+    inference_provider: provider,
     inputTokens: usage?.prompt_tokens ?? 0,
     outputTokens: usage?.completion_tokens ?? 0,
     cacheHitTokens: 0,
     cacheWriteTokens: 0,
-    cost_mUsd: usage ? computeMistralFimMicrodollarCost(usage) : 0,
+    cost_mUsd: usage ? computeFimMicrodollarCost(usage, provider) : 0,
     is_byok: null,
     upstream_id: null,
     finish_reason: null,
@@ -553,8 +566,10 @@ export function countAndStoreFimUsage(
   const usageStatsPromise = !clonedResponse.body
     ? Promise.resolve(null)
     : usageContext.isStreaming
-      ? parseMistralFimUsageFromStream(clonedResponse.body, requestSpan)
-      : clonedResponse.text().then(content => parseMistralFimUsageFromString(content));
+      ? parseMistralFimUsageFromStream(clonedResponse.body, requestSpan, usageContext.provider)
+      : clonedResponse
+          .text()
+          .then(content => parseMistralFimUsageFromString(content, usageContext.provider));
 
   after(
     usageStatsPromise.then(usageStats => {

src/lib/providers/types.ts

@@ -7,6 +7,7 @@ export type ProviderId =
   | 'bytedance'
   | 'coding-plan'
   | 'corethink'
+  | 'inception'
   | 'martian'
   | 'mistral'
   | 'morph'