Generate GPU-aware cloud deployment bundles for running Refua campaigns across public and private cloud providers with Kubernetes and Compose support.
Refua Deploy generates validated deployment bundles for running Refua campaigns across 12 public cloud and 6 private cloud providers. It supports Kubernetes and Docker Compose as runtime targets with automatic network inference, GPU scheduling, and cluster bootstrap.
The tool auto-detects local ClawCures and refua-mcp versions for matching image tags, generates plan files for CI/CD review, and produces ready-to-apply manifests.
Namespace, ConfigMap, Secrets, PVCs, Deployments, Services, CronJobs, Ingress, NetworkPolicy, and Kustomization in one rendered bundle.
Docker Compose files with refua-mcp and campaign runner services plus environment templates for quick local or private cloud deployment.
Three GPU modes: auto (default), required, or off. Supports NVIDIA, AMD, and Intel GPUs with vendor-specific resource requests.
Automatic ingress host inference from config, environment variables, or cloud metadata endpoints. Allowed hosts and origins are derived when omitted.
Auto-generated bootstrap scripts, metadata, and network config files to spin up a fresh cluster and apply Refua manifests in one step.
A plan.json output summarizes the deployment for CI/CD review and
approval workflows before applying changes.
pip install refua-deploy# Generate config with sensible defaults
refua-deploy init --output deploy.yaml \
--name refua-prod --visibility public --provider aws
# Render deployment artifacts
refua-deploy render --config deploy.yaml --output-dir dist
# Bootstrap the cluster
bash dist/bootstrap/cluster-bootstrap.shRefua Deploy takes the complexity out of cloud deployment. From a single config file to a running cluster in three commands, across any major cloud provider.