Infrastructure Upgrade: Utari Moves to AWS Auto-Scaling

We've completed a major infrastructure upgrade that makes Utari faster, more reliable, and ready to scale.

What Changed

Our backend API has moved from a single-server setup on Railway to an AWS Application Load Balancer (ALB) backed by auto-scaling Fargate containers.

This is the same infrastructure used by companies like Stripe and
Notion to handle millions of requests.

What This Means For You

Faster responses. Instead of one server handling all requests, your API calls are now distributed across multiple servers. This means lower latency, especially during peak usage.

Zero downtime. We now run a minimum of 2 API servers across different AWS availability zones. If one server encounters an issue, the other continues serving requests seamlessly. You won't notice a thing.

Auto-scaling. When traffic spikes — say, when an agent is handling multiple concurrent tasks — our infrastructure automatically spins up additional servers within 60 seconds. It scales back down when things quiet down, so we're not wasting resources.

Better streaming. Agent conversations use Server-Sent Events (SSE) for real-time streaming. Our new setup has a 10-minute idle timeout, which means long-running agent tasks won't get disconnected mid-conversation.

Webhook URL Update

If you use webhook triggers, the base URL has changed:

- Old: https://utari.up.railway.app/v1/triggers/{id}/webhook
- New: https://api.utari.ai/v1/triggers/{id}/webhook

New triggers automatically use the new URL. Existing webhooks will continue to work at the old URL for the next 30 days. We recommend updating your external services (Shopify, Stripe, Zapier, etc.) to use the new URL at your earliest convenience.

No Action Required

Unless you use webhook triggers, there's nothing you need to do. The dashboard, agent creation, chat, and all other features work exactly the same. You might just notice things feel a bit snappier.

What's Next

This infrastructure upgrade is the foundation for some exciting features we're rolling out soon, including personalized AI instances with custom subdomains.

Stay tuned.