Set up local AI

A computer with local AI is the heart of your network. Olace handles the setup: it detects your hardware, installs a runtime, and recommends a starter model that actually fits your GPU.

From the app

Choose Start local AI during onboarding, or Settings › Local AI any time. Olace:

1. Detects your GPU and available memory.
2. Recommends a runtime: LM Studio's llmster on Apple Silicon Macs (for MLX), llama.cpp on discrete AMD GPUs, Ollama on everything else. You can pick a different one if you prefer, or run several.
3. Installs it with your consent, showing the exact install command before running it.
4. Suggests a starter model sized to your hardware.

Once set up, the background service keeps local AI and Remote Access available even when the app is closed, and starts your runtime on demand.

From the terminal

olace setup

olace setup does the same detection and consent-gated install. Force a choice with --provider ollama, --provider lms, or --provider llamacpp. Nothing installs without your yes.

Three runtimes, one library

Olace supports Ollama, LM Studio, and llama.cpp side by side. Models from all three appear together in the Model hub and in every model selector, and the Olace Bridge serves all of them. You never have to standardize on one.

Models you downloaded before Olace are picked up too, so an existing llama.cpp or Jan library shows up without being copied or moved.

What you need

Any reasonably modern computer works; more GPU memory lets you run bigger models with longer context. Olace reads your hardware and only recommends models that fit, and the Model hub labels anything that needs more GPU memory than you have. See Hardware & context for how sizing works.