Switch edge runtime to Node.js to avoid Vercel 5-minute timeout

This patch moves the Big-AGI edge router from Vercel Edge runtime to
Node.js runtime to avoid the 5-minute hard timeout affecting slower
models like GPT-5 Pro.

Changes:
- Set runtime='nodejs' with maxDuration=600 (10 minutes)
- Added enhanced error handling matching cloud router
- Updated router documentation

Trade-offs:
- Slower cold starts (~250ms vs ~50ms)
- Single region instead of global edge distribution
- Higher costs on Vercel
- But: Reliable for slow models, 10min vs 5min timeout

For deployments needing >10 minutes, consider:
- Vercel Enterprise (15min max)
- Alternative hosting (Cloudflare, Railway, self-hosted)
- OpenAI Responses API with reconnection

Fixes #857

Co-authored-by: Enrico Ros <enricoros@users.noreply.github.com>

app/api/edge/[trpc]/route.ts

@@ -2,17 +2,40 @@ import { fetchRequestHandler } from '@trpc/server/adapters/fetch';
 
 import { appRouterEdge } from '~/server/trpc/trpc.router-edge';
 import { createTRPCFetchContext } from '~/server/trpc/trpc.server';
+import { posthogCaptureServerException } from '~/server/posthog/posthog.server';
 
 const handlerEdgeRoutes = (req: Request) => fetchRequestHandler({
   endpoint: '/api/edge',
   router: appRouterEdge,
   req,
   createContext: createTRPCFetchContext,
-  onError:
-    process.env.NODE_ENV === 'development'
-      ? ({ path, error }) => console.error(`❌ tRPC-edge failed on ${path ?? 'unk-path'}: ${error.message}`)
-      : undefined,
+  onError: async function({ path, error, type, ctx }) {
+
+    // -> DEV error logging
+    if (process.env.NODE_ENV === 'development')
+      console.error(`❌ tRPC-edge failed on ${path ?? 'unk-path'}: ${error.message}`);
+
+    // -> Capture node errors
+    await posthogCaptureServerException(error, {
+      domain: 'trpc-onerror',
+      runtime: 'nodejs',
+      endpoint: path ?? 'unknown',
+      method: req.method,
+      url: req.url,
+      additionalProperties: {
+        errorCode: error.code,
+        errorType: type,
+      },
+    });
+  },
 });
 
-export const runtime = 'edge';
+// PATCH: Switch from edge to nodejs runtime to avoid Vercel's 5-minute timeout
+// This sacrifices edge performance (speed, global distribution) for reliability
+// with slower models like GPT-5 Pro that can take >5 minutes to respond.
+// maxDuration: 600 seconds (10 minutes) - maximum for Vercel Pro plans
+// For longer durations, consider Vercel Enterprise or alternative hosting.
+export const maxDuration = 600;
+export const runtime = 'nodejs';
+export const dynamic = 'force-dynamic';
 export { handlerEdgeRoutes as GET, handlerEdgeRoutes as POST };

src/server/trpc/trpc.router-edge.ts

@@ -11,7 +11,9 @@ import { llmOpenAIRouter } from '~/modules/llms/server/openai/openai.router';
 import { youtubeRouter } from '~/modules/youtube/youtube.router';
 
 /**
- * Primary rooter, and will be sitting on an Edge Runtime.
+ * Primary router, originally designed for Edge Runtime.
+ * NOTE: Currently configured to run on Node.js runtime (see app/api/edge/[trpc]/route.ts)
+ * to avoid Vercel's 5-minute Edge timeout affecting slower models like GPT-5 Pro.
  */
 export const appRouterEdge = createTRPCRouter({
   aix: aixRouter,