CLIO + ALICE Integration

Use CLIO's remote execution and multi-agent capabilities to deploy, manage, and orchestrate ALICE across GPU fleets.

Overview

CLIO excels at infrastructure automation. ALICE provides local GPU-accelerated generation. Together, they enable programmatic deployment and management of generation infrastructure - from a single Steam Deck to a fleet of GPU servers.

Use Cases

Remote ALICE Deployment

Deploy ALICE to any machine via SSH:

# From your local machine
clio --new
> Deploy ALICE to gpu-server-1 via SSH, install Docker, pull ROCm image, start service

CLIO will:

  1. SSH into the target machine
  2. Check GPU hardware (NVIDIA/AMD/Apple Silicon)
  3. Install Docker and pull the appropriate ALICE image
  4. Configure config.yaml for the detected GPU backend
  5. Start the systemd/Docker service
  6. Verify health endpoint responds

GPU Fleet Orchestration

Deploy ALICE across multiple machines in parallel:

# Parallel deployment to fleet
clio --new
> Deploy ALICE to all machines in gpu-cluster: [gpu-1, gpu-2, gpu-3, gpu-4] using Puppeteer mode

CLIO's Puppeteer mode coordinates the deployment across all targets with rate limiting and progress tracking.

Automated Model Management

Use CLIO to manage models across ALICE instances:

# Download and distribute models
clio --new
> On all ALICE servers: download Juggernaut XL and DreamShaper XL from CivitAI, verify checksums, update model index

Automated Benchmarking

Run generation benchmarks across your fleet:

# Benchmark suite
clio --new
> Run ALICE benchmark on all GPUs: 10 generations each at 1024x1024, 30 steps, collect timing and VRAM usage, generate report

CLIO Tools for ALICE Operations

Tool ALICE Use Case
remote_execution SSH into GPU servers, run deployment scripts
terminal_operations Run ALICE CLI commands, Docker commands
file_operations Edit config.yaml, manage model files
web_operations Download models from CivitAI/HuggingFace APIs
multi_agent Parallel deployment across fleet
remote_execution + terminal_operations Puppeteer mode for orchestration

Deployment Patterns

Docker Deployment (Recommended)

CLIO can deploy ALICE via Docker Compose profiles:

# CLIO generates and runs:
docker compose --profile cuda up -d      # NVIDIA
docker compose --profile rocm up -d      # AMD
docker compose --profile cpu up -d       # CPU only

systemd/launchd Deployment

For bare-metal or macOS:

# Linux systemd
sudo cp alice.service /etc/systemd/system/
sudo systemctl enable --now alice

# macOS launchd
./scripts/install_macos.sh

Steam Deck / SteamOS

CLIO can run the SteamOS installer remotely:

clio --new
> SSH into Steam Deck, run ALICE SteamOS installer, configure ROCm, start user service

Calling ALICE API from CLIO

CLIO can call ALICE's OpenAI-compatible API directly for testing:

# Test generation via API
clio --new
> POST to http://gpu-server:8080/v1/images/generations with prompt "test image", verify response, check image URL

See Also