Spent the last few months on this and I think it's finally ready to share here.

CloseAI is an iOS app that takes a fresh Ubuntu machine and turns it into a private chatbot you can use from your phone. You enter the IP and SSH credentials, the app uploads an install script, and a few minutes later you're chatting with an open-source model over your own HTTPS endpoint.

The whole point is that the user never opens a terminal. No commands typed, no nginx config, no Let's Encrypt dance, no manually editing a systemd unit. The app handles all of it over SSH (TLS uses self-signed certs with TOFU pinning, like SSH host keys).

It works fine on a $10/month VPS, but the part I'm most excited about is that it works just as well on hardware you already own. I've been running it on an old Ubuntu desktop in my closet with Llama and Gemma runing snappy, Qwen 2.5 Coder and DeepSeek R1 usable. From my phone it's the same UI and the same model, except nothing is going to a third party. No relay, no account, no analytics, no telemetry. The model is yours and the data is yours.

The five models it ships with (more to come):

• Llama 3.2:3b
• Gemma 3:4b
• Phi 4 Mini
• Qwen 2.5 Coder:7b
• DeepSeek R1:7b

Custom model support is what I'm working on next. Port 11434 stays open on the server, so if you want to ollama pull other models manually they'll work over the API — they just won't show up in the in-app picker yet.

Real limitations (so nobody feels ambushed in the comments):

• iOS only. Android would be a separate codebase and isn't on the roadmap yet.
• No system prompt UI yet (planned)
• No thinking-mode toggle for R1. <think> blocks get stripped from the chat view (planned)
• The app installs its own FastAPI/TLS layer, so if you already have Ollama running on the box it'll coexist alongside, not replace it
• No temperature/top_p/quant controls in the UI yet

Free on the App Store: https://apps.apple.com/us/app/closeai/id6760688649

P.S. I know most of you don't need an installer for any of this. After all, you're the people who taught the rest of us how to do it. But if you've got a friend, partner, or coworker who keeps bouncing off the standard instructions, this might be the thing that finally gets them running something local. Genuinely happy to take feedback from people who live in this stack: what's missing, what's wrong, what feels off.