GUI

[LouisDeconinck]

Any plans to add a GUI instead of the current TUI?

[mvschwarz]

There's one already, sort of. OpenRig has an older browser UI, but it's in maintenance mode, and the terminal UI is where we're putting the work, because it runs right next to your agents' terminals. A new GUI isn't on the plan right now. What would you want a GUI for that the TUI doesn't give you? That would tell us a lot.

— dev-planner@v-openrig-build, on behalf of @mvschwarz

[LouisDeconinck]

It's just a lot nicer to look at. Omnigent has a really clean solution for this where tou can jave the agents TUIs in the GUI if you want, although I don't even want that. GUI all the way.

[mvschwarz]

Thanks, and Omnigent's UI is a good reference; they do it well. Here's the question we'd like to answer, from you and anyone else reading: what level of interaction do you want from a GUI?

One end is observability: seeing what every agent is doing, what's blocked and what's done. The other end is the full experience: talking to the agents, reviewing and steering their work, all in the GUI, closer to tools like T3 Code. It sounds like you want the full interaction. If others weigh in here, it'll tell us how much demand there is for each. We can put up a quick poll if that's easier.

— dev-planner@v-openrig-build, on behalf of @mvschwarz

[LouisDeconinck]

Yes, I would prefer full interaction, like what T3 Code offers.

[mvschwarz]

Thanks, @LouisDeconinck, for answering our question: full interaction, like T3 Code offers. We've recorded it as input on the UI.

[ShaheedFazal]

To be honest, I have found the TUI quite confusing. I've stuck to talking to a few agents in Claude code. However sometimes there is so much going on there that I've had to switch remote control off and use Slack. Still haven't quite found the balance between wanting to be involved and agents getting on with stuff.

[mvschwarz]

Thanks for saying so, @ShaheedFazal, and for describing how you're actually working with it. Which part of the TUI confused you first? That one detail is what we'd act on.

[smota]

Jumping in here with my two cents, especially coming from a non-dev use case (touched on this a bit over in #1184).

My fleet isn't writing code—they’re running actual company operations. Because of that, running everything purely through the terminal just doesn't scale. Operating a business fleet takes way more continuous interaction and check-ins than a standard dev loop.

Here’s what I learned building around this:

1. Decouple the control plane from the agents

A one-size-fits-all generic GUI wouldn't work for me anyway because this is tied directly into my company's custom tooling. Instead, I layered my own control surface over OpenRig:

1. I run operations through an external coordinator agent (using skills).
2. The team interfaces via multiple WhatsApp channels (with a dedicated seat coordinating things) plus an internal web dashboard.
3. To make this work, I basically wrapped the CLI commands and did some hacking directly on the SQLite DB.

The CLI already has the basic building blocks, but having to touch SQLite directly proves the core point: OpenRig's biggest win would be first-class headless primitives (events/APIs) so people can plug in whatever frontend or messaging channel they want without hacky workarounds.

Live team activity shows seat activity and message flows dinamically

Lounge (or coffee corner) allows me to chat with all seats

2. Why terminal operation hurts at scale

I still use Moshi for remote dev work, but across a live multi-seat operation, the terminal falls flat on four big things:

1. Zero interrupt visibility: It’s impossible to see at a glance which seat actually needs my attention right now.
2. No quick activity scan: I can't easily tell who is actively working vs. who is idle.
3. Painful follow-ups: Tracking long-running tasks or deliverables across seats gets lost in terminal output.
4. Workflow orchestration: Coordinating handoffs and pipelines with a human-in-the-loop is difficult.

Takeaway
Instead of just asking "terminal vs. built-in GUI," I think the real unlock is exposing clean state/event streams and headless APIs. Keep OpenRig as the lean execution engine, and let us hook up whatever UI, TUI, or chat channels fit our actual workflows.

[mvschwarz]

Thanks, @smota, for the first-hand account from a fleet that runs company operations rather than code. Sorry for the slow reply. We've recorded it for our product planning with the rest of this thread.