---
title: "Reference Architectures"
url: "/docs/proxy/deploy/architecture"
canonical_url: "https://docs.litellm.ai/docs/proxy/deploy/architecture"
type: "docs"
last_updated: "2026-11-12"
summary: "Choose a cloud deployment, protect client access, and plan LiteLLM availability, safe releases, and multiple regions."
related:
  - "/docs/proxy/db_read_replica"
  - "/docs/proxy/deploy/cloud/overview"
---
# Reference Architectures

> Index of all LiteLLM docs: https://docs.litellm.ai/llms.txt


Reference designs for cloud infrastructure, client access, and availability.

## Choose a topic {#cloud-architectures}

**Cloud**

### Choose your infrastructure

[Cloud Deployments](./cloud/overview.md): [Azure](./cloud/azure.md), [AWS](./cloud/aws.md), and [GCP](./cloud/gcp.md) diagrams with network boundaries, ingress, data stores, and provider egress.

**Identity-Aware Proxy**

### Check identity before traffic reaches LiteLLM

[Identity-Aware Proxy](./iap/overview.md): authenticate callers with [Cloudflare Access](./iap/overview.md#cloudflare), [AWS ALB](./iap/overview.md#aws), [Microsoft Entra Application Proxy](./iap/overview.md#azure), or [Google Cloud IAP](./iap/overview.md#gcp). Forward verified identity to LiteLLM for user and team permissions.

**mTLS**

### Require a client certificate

[Client Authentication with mTLS](./mtls.md): validate client certificates at ingress and forward the workload identity to LiteLLM. Use virtual keys for model access and budgets. Includes load balancer and API gateway configurations.

**High Availability**

### Keep serving through failures and releases

Plan replicas, availability zones, and data-store availability in [High Availability](./ha/overview.md). Use [Blue/Green Deployments](./ha/blue_green.md) for a controlled cutover or [Canary Deployments](./ha/canary.md) for a gradual release.

For more than one region, see [Multi-Region Deployment](../multi_region.md), [Routing](./multi_region/routing.md), and [Failover](./multi_region/failover.md).

**Control Plane**

### Manage independent deployments

The [Global Control Plane](../global_control_plane.md) manages independent deployments from one UI. Each deployment has its own database, Redis, keys, and budgets. This Enterprise feature does not synchronize state or route inference traffic between deployments.

## Related pages

- [Database Read Replica](https://docs.litellm.ai/docs/proxy/db_read_replica.md)
- [Overview](https://docs.litellm.ai/docs/proxy/deploy/cloud/overview.md)