feat(console): route migrated Go inference (#48123)

packages/console/app/src/lib/inference-proxy.ts

@@ -8,11 +8,17 @@ const paths: Record<string, string | undefined> = {
   "POST /zen/v1/chat/completions": "/openai/v1/chat/completions",
   "POST /zen/v1/responses": "/openai/v1/responses",
   "POST /zen/v1/messages": "/anthropic/v1/messages",
+  "POST /zen/go/v1/chat/completions": "/go/openai/v1/chat/completions",
+  "POST /zen/go/v1/responses": "/go/openai/v1/responses",
+  "POST /zen/go/v1/messages": "/go/anthropic/v1/messages",
+  "GET /zen/v1/models": "/v1/models",
+  "GET /zen/go/v1/models": "/go/v1/models",
+  "GET /zen/go/v1/usage": "/go/v1/usage",
 }
 
 export async function proxyInference(
   request: Request,
-  generation: {
+  generation?: {
     provider?: "openai" | "anthropic" | "google"
     /** The provider's native model ID, not the public Zen alias. */
     model?: string
@@ -28,7 +34,8 @@ export async function proxyInference(
       : undefined)
   if (!path) return undefined
 
-  const key = path.startsWith("/anthropic/")
+  const go = url.pathname.startsWith("/zen/go/")
+  const key = url.pathname.endsWith("/messages")
     ? request.headers.get("x-api-key")
     : path.startsWith("/google/")
       ? request.headers.get("x-goog-api-key")
@@ -47,7 +54,7 @@ export async function proxyInference(
       .innerJoin(WorkspaceTable, eq(WorkspaceTable.id, KeyTable.workspaceID))
       .leftJoin(
         ProviderTable,
-        generation.provider
+        !go && generation?.provider
           ? and(
               eq(ProviderTable.workspaceID, KeyTable.workspaceID),
               eq(ProviderTable.provider, generation.provider),
@@ -61,7 +68,7 @@ export async function proxyInference(
       .then((rows) => rows[0]),
   )
   if (!workspace?.migratedAt) return undefined
-  const model = workspace.provider ? generation.model : undefined
+  const model = workspace.provider ? generation?.model : undefined
   if (workspace.provider && !model) throw new Error("Legacy BYOK model mapping is unavailable")
 
   const destination = new URL(Resource.ConsoleMigration.inferenceUrl)
@@ -80,8 +87,19 @@ export async function proxyInference(
   // Model extraction has already read part of the body; forward its replay stream.
   const forwarded = new Request(
     destination,
-    new Request(request, { method: request.method, body: generation.body(model) }),
+    generation ? new Request(request, { method: request.method, body: generation.body(model) }) : request,
   )
+  // Migrated requests use ordinary destination authentication and accounting.
+  for (const name of [
+    "x-zen",
+    "x-zen-model",
+    "x-zen-ip",
+    "cf-access-client-id",
+    "cf-access-client-secret",
+    "host",
+    "content-length",
+  ])
+    forwarded.headers.delete(name)
   forwarded.headers.set("authorization", `Bearer ${key}`)
   const ip = request.headers.get("cf-connecting-ip")
   if (ip) forwarded.headers.set("x-real-ip", ip)
@@ -90,3 +108,10 @@ export async function proxyInference(
 
   return fetch(forwarded, { redirect: "manual" })
 }
+
+export function inferenceUnavailable() {
+  return Response.json(
+    { error: { type: "api_error", message: "Inference routing is unavailable. Please retry later." } },
+    { status: 503, headers: { "Cache-Control": "no-store" } },
+  )
+}

packages/console/app/src/routes/zen/go/v1/models.ts

@@ -1,12 +1,15 @@
 import type { APIEvent } from "@solidjs/start/server"
 import { ZenData } from "@opencode-ai/console-core/model.js"
 import { buildModelsResponse, buildOptionsResponse } from "../../util/modelsHandler"
+import { inferenceUnavailable, proxyInference } from "~/lib/inference-proxy"
 
 export async function OPTIONS(_input: APIEvent) {
   return buildOptionsResponse()
 }
 
-export async function GET(_input: APIEvent) {
+export async function GET(input: APIEvent) {
+  const response = await proxyInference(input.request).catch(inferenceUnavailable)
+  if (response) return response
   const models = Object.keys(ZenData.list("lite").models)
   return buildModelsResponse(models)
 }

packages/console/app/src/routes/zen/go/v1/usage.ts

@@ -6,8 +6,11 @@ import { UserTable } from "@opencode-ai/console-core/schema/user.sql.js"
 import { WorkspaceTable } from "@opencode-ai/console-core/schema/workspace.sql.js"
 import { LiteData } from "@opencode-ai/console-core/lite.js"
 import { Subscription } from "@opencode-ai/console-core/subscription.js"
+import { inferenceUnavailable, proxyInference } from "~/lib/inference-proxy"
 
 export async function GET(input: APIEvent) {
+  const response = await proxyInference(input.request).catch(inferenceUnavailable)
+  if (response) return response
   const apiKey = input.request.headers.get("authorization")?.match(/^Bearer (\S+)$/)?.[1]
 
   if (!apiKey) {

packages/console/app/src/routes/zen/util/handler.ts

@@ -50,7 +50,7 @@ import { countryFromRequest, isModelCountryRestricted } from "~/lib/request-coun
 import { isPeakPricing } from "./pricing"
 import { prepareRequestBody } from "./requestBody"
 import { requiresGoTrainingConsent } from "./trainingConsent"
-import { proxyInference } from "~/lib/inference-proxy"
+import { inferenceUnavailable, proxyInference } from "~/lib/inference-proxy"
 
 type ZenData = Awaited<ReturnType<typeof ZenData.list>>
 type PreparedBody = Awaited<ReturnType<typeof prepareRequestBody>>
@@ -102,22 +102,22 @@ export async function handler(
     const rawZenApiKey = opts.parseApiKey(input.request.headers)
     const zenApiKey = rawZenApiKey === "public" ? undefined : rawZenApiKey
     const zenData = ZenData.list(opts.modelList)
-    if (opts.modelList === "full" && model) {
+    if (model) {
       // Read routing metadata without running legacy model, auth, or balance checks.
       const configured = zenData.models[model]
       const entry = Array.isArray(configured)
         ? configured.find((entry) => entry.formatFilter === opts.format)
         : configured
       const response = await proxyInference(input.request, {
-        provider: entry?.byokProvider,
-        model: entry?.providers.find((provider) => provider.id === entry.byokProvider)?.model,
+        provider: opts.modelList === "full" ? entry?.byokProvider : undefined,
+        model:
+          opts.modelList === "full"
+            ? entry?.providers.find((provider) => provider.id === entry.byokProvider)?.model
+            : undefined,
         body: (providerModel) => requestBody?.stream(providerModel ?? model, false) ?? body,
       }).catch(() => {
         void (requestBody ? requestBody.cancel() : body.cancel()).catch(() => {})
-        return Response.json(
-          { error: { type: "api_error", message: "Inference routing is unavailable. Please retry later." } },
-          { status: 503, headers: { "Cache-Control": "no-store" } },
-        )
+        return inferenceUnavailable()
       })
       if (response) return response
     }

packages/console/app/src/routes/zen/v1/models.ts

@@ -5,7 +5,7 @@ import { KeyTable } from "@opencode-ai/console-core/schema/key.sql.js"
 import { WorkspaceTable } from "@opencode-ai/console-core/schema/workspace.sql.js"
 import { ModelTable } from "@opencode-ai/console-core/schema/model.sql.js"
 import { buildOptionsResponse, buildModelsResponse } from "~/routes/zen/util/modelsHandler"
-import { Resource } from "@opencode-ai/console-resource"
+import { inferenceUnavailable, proxyInference } from "~/lib/inference-proxy"
 
 export async function OPTIONS(_input: APIEvent) {
   return buildOptionsResponse()
@@ -14,12 +14,7 @@ export async function OPTIONS(_input: APIEvent) {
 export async function GET(input: APIEvent) {
   const apiKey = input.request.headers.get("authorization")?.split(" ")[1]
   if (apiKey && apiKey !== "public") {
-    const response = await proxyModels(input, apiKey).catch(() =>
-      Response.json(
-        { error: { type: "api_error", message: "Inference routing is unavailable. Please retry later." } },
-        { status: 503, headers: { "Cache-Control": "no-store" } },
-      ),
-    )
+    const response = await proxyInference(input.request).catch(inferenceUnavailable)
     if (response) return response
   }
 
@@ -45,26 +40,3 @@ export async function GET(input: APIEvent) {
 
   return buildModelsResponse(models)
 }
-
-async function proxyModels(input: APIEvent, apiKey: string) {
-  // No legacy revocation or model-policy checks before destination authentication.
-  const workspace = await Database.use((tx) =>
-    tx
-      .select({ migratedAt: WorkspaceTable.migrated_at })
-      .from(KeyTable)
-      .innerJoin(WorkspaceTable, eq(WorkspaceTable.id, KeyTable.workspaceID))
-      .where(eq(KeyTable.key, apiKey))
-      .limit(1)
-      .then((rows) => rows[0]),
-  )
-  if (!workspace?.migratedAt) return undefined
-
-  const destination = new URL(Resource.ConsoleMigration.inferenceUrl)
-  destination.pathname = `${destination.pathname.replace(/\/$/, "")}/v1/models`
-  destination.search = new URL(input.request.url).search
-  destination.hash = ""
-  const headers = new Headers({ authorization: `Bearer ${apiKey}` })
-  const ip = input.request.headers.get("cf-connecting-ip")
-  if (ip) headers.set("x-real-ip", ip)
-  return fetch(destination, { headers, signal: input.request.signal, redirect: "manual" })
-}