Ollama (Mac · Host)

Local LLM runner (Ollama) for obacht Macs. Unlike the Raspberry Pi bundle, Ollama here runs DIRECTLY ON THE MAC HOST, not inside the obacht VM, so it gets full Apple-Silicon GPU (Metal) acceleration and the machine's whole RAM. obacht downloads the pinned, checksum-verified Ollama release into its own managed directory and runs it as a user LaunchAgent (dev.obacht.hostsvc.*); no Homebrew, no admin rights, and nothing system-wide.

Pick a model at install. obacht downloads it into this Ollama right after the service starts and reports the app as ready once the model is on disk. Models live in obacht's own data directory for this service, so removing the template leaves nothing behind in your home folder. Pull and switch further models inside Open WebUI.

Pair it with the "Open WebUI (Mac)" template: that runs the web UI as a container in the obacht VM and talks to this host Ollama over the VM's gateway. Install Ollama first, Open WebUI second.

Mac-only: this template is not offered on Raspberry Pi devices (the Pi has its own ollama-openwebui bundle that runs everything in Docker).

Security note: Ollama listens on 0.0.0.0:11434 so the VM container can reach it via the host gateway. That also exposes it on the Mac's LAN. Keep this device on a trusted network (a future version will bind to the VM gateway interface only).