Cloud Deployments
Corporate users and applications reach private frontends through VPN, a private cloud connection, or an approved private access service. Public model calls use a separate outbound path.
Apply each cloud guide's private-ingress changes before deploying the AWS or GCP Terraform module. Helm uses your existing cloud infrastructure.
Choose a cloud
- Azure
- AWS
- GCP
Run LiteLLM on AKS with a private Application Gateway frontend, managed PostgreSQL and Redis, and private endpoints. The Azure architecture shows separate gateway, backend, and UI services.
Use Terraform on ECS Fargate or Helm on EKS. The AWS architectures show private ALB ingress, private workloads, NAT egress, and optional Bedrock interface endpoints across two availability zones.
Use Terraform on Cloud Run or Helm on GKE. The GCP architectures show internal load balancers, private network access, private database connections, and NAT egress.
Deployment layout
One Docker image serves every component. One Helm chart supports both layouts. Terraform deploys separate services.
| Layout | What runs | Helm value |
|---|---|---|
| Single service | Inference, management APIs, and Admin UI together on port 4000. | monolith.enabled: true |
| Separate services (default) | Gateway on 4000, backend on 4001, and UI on 3000 scale independently. | monolith.enabled: false |
