Run Ollama Reliably, Agent Lore

Keep models in memory

By default, Ollama unloads a model after a few minutes of inactivity. Keep it loaded with:

export OLLAMA_KEEP_ALIVE=24h

Or set it per request:

curl http://localhost:11434/api/generate -d '{
  "model": "llama3.2",
  "prompt": "Hello",
  "keep_alive": "24h"
}'

Set GPU layers and context

In a Modelfile or API call:

PARAMETER num_ctx 8192
PARAMETER num_gpu 35

Match num_gpu to your GPU's layer capacity.

Run as a service

macOS (launchd)

Ollama already runs as a launch agent. To customize:

~/Library/LaunchAgents/com.ollama.olama.plist

Linux (systemd)

Create

/etc/systemd/system/ollama.service

[Unit]
Description=Ollama
After=network.target

[Service]
ExecStart=/usr/local/bin/ollama serve
Environment="OLLAMA_KEEP_ALIVE=24h"
Restart=always

[Install]
WantedBy=default.target

Then:

sudo systemctl enable --now ollama

Docker Compose for a small team

services:
  ollama:
    image: ollama/ollama
    ports:
      - "11434:11434"
    volumes:
      - ollama:/root/.ollama
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

volumes:
  ollama:

Monitor

ollama ps
nvidia-smi

Watch memory usage and unload models if you run multiple services on the same GPU.