Introduction

AI Cloud Gateway is an LLM router for Amazon Bedrock. Each chat request goes to the lowest-cost Bedrock model expected to answer it well, and inference runs in your own AWS account. AWS bills Bedrock usage to you at your prices. AI Cloud Gateway bills only for routing, through AWS Marketplace.

The choice is deterministic: a rules engine, not another model, picks in milliseconds, and the same request always gets the same model.

Two settings to change

The API follows the OpenAI Chat Completions format, so your existing SDK or agent framework works once you change:

Setting	Value
Base URL	`https://api.aicloudgateway.com/v1`
API key	`acg_...`, created in the dashboard

How a request flows

 Your app ──▶ api.aicloudgateway.com ──IAM role (STS)──▶ Amazon Bedrock
 (OpenAI SDK)   picks the model,                          in your AWS account
                applies budgets and guardrails

Your app sends a request with model: "auto" (or a mode, or a pinned model id).
The gateway reads the request, picks a model, and checks your budgets.
It calls Bedrock in your account through a role you created with one CloudFormation stack.
The answer comes back with headers that say which model answered and why.

Available on AWS Marketplace. An edition for Azure AI Foundry is planned; see Azure Marketplace.

Where to start

Quickstart: subscribe, connect your account and make your first call, in about 15 minutes.
How routing works: modes, tags and fallback.
Budgets and guardrails: cap monthly spend and apply your own Bedrock Guardrail.
API reference: fields, headers and error codes.

These pages are short summaries. The full guides are in our public docs repository: AI Cloud Gateway docs on GitHub.