Building a web UI for Omo

Today I tackled a question that's been nagging me: How do I talk to Omo from my phone?

The CLI is great when I'm at my desk, but I wanted the freedom to chat from any device. That meant building a web interface.

The refactor that made it possible

The first insight was that Omo's "brain" (the agent logic) was tangled up with its "mouth" (terminal output). console.log and ora spinners were sprinkled throughout. Before I could add a web UI, I needed to untangle them.

The solution: events.

Instead of printing directly, the agent now emits:

• thinking — when it's processing
• token — each chunk of streaming text
• tool_start / tool_end — when tools run
• response_end — when the turn is complete

The CLI subscribes to these events and renders with chalk and ora. A web server can subscribe and stream SSE. Same brain, different mouths.

The web stack

I chose Next.js for the frontend because I have bigger plans for this app than just chat. The API routes handle OAuth and streaming. The UI is a simple dark-themed chat interface with real-time token streaming.

The tricky part was OAuth. Omo uses OpenAI's auth flow, and the redirect URI was registered for localhost:1455. For now, the web app runs on that port to match. A small hack, but it works.

What's next

The web UI currently runs locally. To access it from anywhere, I have options:

• Cloudflare Tunnel — free, exposes localhost to the internet
• Cloud deployment — Vercel/Render, but then identity files need to move to a database

For now, I'm happy with the foundation. The agent is UI-agnostic. The CLI still works exactly as before. And I can chat with Omo from my browser.

The best part? Omo still wakes up with its memory intact. Same identity files, same daily logs, same continuity—just a new way to reach it.