diff --git a/ui-tui/src/entry.tsx b/ui-tui/src/entry.tsx
index 3ab4be96ba..abbcf7a4c8 100644
--- a/ui-tui/src/entry.tsx
+++ b/ui-tui/src/entry.tsx
@@ -1,7 +1,8 @@
 #!/usr/bin/env node
-import { render } from '@hermes/ink'
-
-import { App } from './app.js'
+// Import order matters for cold start: `GatewayClient` has only node-builtin
+// deps (<20ms), so spawning the python gateway before loading @hermes/ink
+// + App (~200ms combined) gives python ~200ms of free parallel time to run
+// its own module imports instead of starting those after node is done.
 import { GatewayClient } from './gatewayClient.js'
 
 if (!process.stdin.isTTY) {
@@ -11,6 +12,7 @@ if (!process.stdin.isTTY) {
 
 const gw = new GatewayClient()
 gw.start()
-render(<App gw={gw} />, {
-  exitOnCtrlC: false
-})
+
+const [{ render }, { App }] = await Promise.all([import('@hermes/ink'), import('./app.js')])
+
+render(<App gw={gw} />, { exitOnCtrlC: false })