Run one command on your Mac (a Mac mini is perfect). It serves its Ollama models over your Wi-Fi and prints a QR code. Scan it, and your phone gets a chat app that streams answers from the Mac. The phone is the screen, the Mac runs the model. Free and open source.
npx github:VedSoni-dev/berth serve
Install Ollama, pull a model (ollama pull llama3.2), then run npx github:VedSoni-dev/berth serve in Terminal. It prints a QR code.
Same Wi-Fi, the camera app. The chat opens with the pairing code in its address; add it to your Home Screen.
The Mac runs the model and streams the answer back, with tokens per second and "running on" your Mac above it.
Served by the Mac. Streaming answers, a model picker with the Mac's Ollama models, Stop, light and dark, one-handed at 390 px, installable to the Home Screen.
In the QR code, checked on every request; nothing answers without it. --new-token unpairs every phone.
One small Node server, a QR encoder and a static app. Nothing goes through a server of ours.
Shows the command and whether a phone is paired. Also runs a small model in a tab on WebGPU, or connects Ollama for a local OpenAI-compatible endpoint.
Ollama by default; --ollama <url> points it at LM Studio or llama.cpp instead.
Put the Mac and the phone on Tailscale and open the Mac's Tailscale address. berth has no relay of its own.
.local names: run berth serve --ip.git clone https://github.com/VedSoni-dev/berth cd berth && npm run serve # the Mac server and phone app npm install && npm run dev # the site and dashboard
berth is coming to publik, a free software marketplace: the listing goes up after 19 November 2026. Until then the dashboard opens with no sign-in.