zen: new inference

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

@@ -210,7 +210,10 @@ export async function handler(
       )
       logger.debug("REQUEST URL: " + reqUrl)
       logger.debug("REQUEST: " + reqBody.substring(0, 300) + "...")
-      const res = await fetchWith429Retry(reqUrl, {
+      const isNewInference = providerInfo.id.startsWith("console.") || providerInfo.id.startsWith("console-go.")
+      const res = await fetchWith429Retry(
+        reqUrl,
+        {
           method: "POST",
           headers: (() => {
             const headers = new Headers(input.request.headers)
@@ -240,9 +243,11 @@ export async function handler(
           // Propagate caller disconnects to the upstream provider request so
           // abandoned Console requests do not leave orphaned inference work open.
           signal: input.request.signal,
-      })
+        },
+        { count: isNewInference ? MAX_429_RETRIES : 0 },
+      )
 
-      if (providerInfo.id.startsWith("console.") || providerInfo.id.startsWith("console-go.")) {
+      if (isNewInference) {
         const resEndpointId = res.headers.get("x-opencode-endpoint-id")
         const resEndpointModelId = res.headers.get("x-opencode-upstream-model-id")
         if (resEndpointId && resEndpointModelId)