Use Your GPU (native Ollama)

GCTRL ships with a bundled Ollama so it works out of the box - but that one runs inside Docker, which is CPU-only (Docker can't reach your GPU: Apple Metal, or NVIDIA/AMD on Windows/Linux). For GPU speed, run Ollama natively on the host and point GCTRL at it.

It's two short steps.

1. Expose native Ollama to GCTRL

Native Ollama listens on 127.0.0.1 (localhost) by default - only the host can reach it, not the GCTRL containers. Tell Ollama to listen on all interfaces by setting OLLAMA_HOST=0.0.0.0:11434, then restart it. This is Ollama's own setting

one environment variable, no GCTRL-specific change.

Linux (systemd)

sudo mkdir -p /etc/systemd/system/ollama.service.d
printf '[Service]\nEnvironment="OLLAMA_HOST=0.0.0.0:11434"\n' | \
  sudo tee /etc/systemd/system/ollama.service.d/override.conf
sudo systemctl daemon-reload && sudo systemctl restart ollama

macOS

launchctl setenv OLLAMA_HOST "0.0.0.0:11434"
# then quit and reopen the Ollama app (menu bar)

Windows

setx OLLAMA_HOST "0.0.0.0:11434"
# then quit Ollama from the system tray and reopen it

Security note: 0.0.0.0 makes Ollama reachable from your network, not just localhost. On a trusted/private machine that's fine; on an exposed host, firewall port 11434 to the Docker bridge only.

2. Point GCTRL at it

In the dashboard, go to Settings → AI Models (or Infrastructure) and set the Ollama base URL to:

http://localhost:11434

GCTRL automatically routes localhost to your host (host.docker.internal) from inside the containers - so the natural value just works. Click Test connection; once it's reachable, your installed models (and GPU) appear.

What GCTRL does for you

Auto-routing: you can paste localhost:11434 - GCTRL rewrites it to reach your host, so you don't need to know host.docker.internal.
Safe fallback: if the installer finds a native Ollama that's localhost-only, it keeps the bundled CPU Ollama running so the platform still works out of the box - you switch to native (GPU) whenever you're ready, with no downtime.
Guided UI: when Ollama isn't reachable, Settings → AI Models shows these exact commands for your OS with a one-click Test again.

Models and VRAM

The GPU only helps for models that fit in VRAM. As a rough guide: embeddings (nomic-embed-text) and small models (llama3.2:3b, qwen2.5:3b) fit in ~3-4 GB; qwen2.5:7b needs ~6 GB; larger models need proportionally more. A model that doesn't fit falls back to CPU (slow) or fails to load - pick a model sized for your GPU, or use a cloud model for the big ones.

→ Related: LLM Providers · Infrastructure & Ollama · Installation