Getting Started

Minikube

For testing Anex with minimal impact on your existing infrastructure, Minikube is the easiest option. Anex’s gateway needs a publicly reachable endpoint for remote machines to connect to, so running on a laptop is usually not feasible — but running Minikube on an EC2 (or any public-IP) instance works fine.

The gateway is exposed via a NodePort. Map the ports when starting Minikube:

minikube start --driver=docker --listen-address=0.0.0.0 \
  --ports=31000:31000/udp,31001:31001/tcp

If using AWS EC2, allow UDP 31000 and TCP 31001 in the Security Group. The TCP port is the wstunnel endpoint used by RunPod machines (Vast.AI uses UDP).

Configure the gateway class to node-port and set domain to the machine’s public IP in values.yaml:

deployment:
  gateway:
    class: "node-port"
    domain: "3.44.111.222"
    nodePortUDP: 31000
    nodePortTCP: 31001

The rest of the setup is identical to AWS.

AWS

Prerequisites

Before you begin, ensure you have:

• AWS Load Balancer Controller — Anex needs to expose a UDP endpoint for the gateway
• (optional) ExternalDNS — for stable DNS names so reinstalls don’t require re-handshaking peers

Installation

1. Create an API key in your Vast.AI account.

2. Create an account on ANEX.sh and get a download token for agent binary

3. Create minimal quickstart.values.yaml:

   deployment:
     gateway:
       class: "aws-eks"
   
   appConfig:
     cluster:
       # clusterUUID must be unique across clusters under the same Vast.AI account
       clusterUUID: "mycluster"
   
     cloudProvider:
       # one of: vastai, runpod, mock
       active: "vastai"
       vastAI:
         apiKey: ""    # API key from step 1
   
     cdn:
       anexAuthToken: ""  # download token from step 2"
   
     virtualNode:
       # Node name must be unique in your Kubernetes cluster
       nodeName: "virtual-node"
       # Labels picked up by pod nodeSelector / affinity
       labels:
         node-provider: vastai
   

4. Install the Anex Helm chart:

   helm upgrade --install anex \
     --namespace anex --create-namespace \
     -f quickstart.values.yaml \
     oci://public.ecr.aws/m4v1f8q5/gpu-provider/helm \
     --version 1.0.0
   

5. Once the provider is up, you’ll see a new node in your cluster.

Deploying GPU Workloads

A minimal pod that schedules onto the virtual node — the default scheduler places it there, and the virtual kubelet provisions a machine in Vast.AI to run the container.

apiVersion: v1
kind: Pod
metadata:
  name: dummy
  annotations:
    # Use annotations to specify machine requirements
    anex.sh/region: "europe"
    anex.sh/gpu-names: "RTX 4090,RTX 3090"
    anex.sh/price-max: "0.5"
spec:
  containers:
    - name: dummy
      image: ubuntu:22.04
      # Command must be explicitly set; image default is not used
      command: ['sleep', '3600']
  nodeSelector:
    node-provider: vastai
  # Default taints on the virtual node
  tolerations:
    - key: "virtual-kubelet.io/provider"
      operator: "Equal"
      value: "vastai"
      effect: "NoSchedule"
    - key: "ignore-taint.cluster-autoscaler.kubernetes.io/manual-ignore"
      operator: "Equal"
      value: "true"
      effect: "NoSchedule"

See Running GPU Workloads for full details on machine selection, and Examples for the VirtualService CRD that exposes virtual pods through a Service.