Docker deployment

Containerised Soriku + Ollama, ready to bring up with one command.

Bring up the stack

docker compose up -d
docker compose exec ollama ollama pull qwen2.5-coder:7b
docker compose logs -f soriku

Volumes

The compose file mounts ./data for Soriku's persistent state and a named volume ollama-models for pulled models. Back up ./data and you back up agents, capability map, routing logs and keystore.

GPU acceleration

On Linux with NVIDIA, pass through the GPU via the NVIDIA Container Toolkit:

# docker-compose.override.yml
services:
  ollama:
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

macOS Docker can't reach the Metal GPU. For Apple Silicon, native Python is much faster than Docker.

Updating

docker compose pull
docker compose up -d