# AWS Bedrock

> ZopNight registers 4 AWS Bedrock models through the AI Gateway: Nova Pro and Nova Lite, plus Claude Sonnet and Haiku served via Bedrock. Region-specific inference profiles are rewritten automatically, and spend is attributed per team alongside the rest of your AWS bill.

Source: https://zop.dev/integrations/ai-gateway/aws-bedrock
Updated: 2026-08-20

---

## Common questions

### Will a Bedrock model ID from us-east-1 work in eu-west-1?

Yes. Bedrock addresses many models through region-prefixed inference profiles (us., eu., apac.), and the gateway rewrites the prefix for your region so the same model ID does not silently fail elsewhere.

### Do models I enabled in my own Bedrock account show up?

Yes. Live discovery calls ListFoundationModels and ListInferenceProfiles, so anything you have enabled appears without waiting for a ZopNight release. Amazon, Anthropic, Meta, Mistral and Cohere are all reachable this way.

### Bedrock spend is already on my AWS bill. Why meter it again?

Because the invoice cannot tell you which team spent it. Metering at the gateway attributes each request to a team and a virtual key, which is what a large AWS account otherwise loses.

## Amazon Nova and Anthropic models on Bedrock

Four, spanning Amazon's own family and Anthropic served through Bedrock:

- `nova-pro`, `nova-lite`: Amazon Nova
- `bedrock-sonnet` → claude-sonnet-4-5, `bedrock-haiku` → claude-haiku-4-5

## Inference profiles are rewritten for you

Bedrock addresses many models through region-prefixed inference profiles: `us.`, `eu.`,
`apac.`. The gateway rewrites the profile for your region, so a model ID that works in
`us-east-1` does not silently fail in `eu-west-1`.

Live discovery runs `ListFoundationModels` and `ListInferenceProfiles`, so models you have
enabled in your own Bedrock account appear without a ZopNight release. Vendors reachable this
way include Amazon, Anthropic, Meta, Mistral and Cohere.

## Why Bedrock spend is worth attributing separately

Bedrock usage lands on your AWS bill, where it is easy to lose inside a large account. Metering
it at the gateway gives you per-team and per-key attribution that the AWS invoice alone cannot
provide.

## Bedrock requests stay between you and AWS

ZopNight never sits in your request path. The gateway runs on hosted LiteLLM; ZopNight
configures it, meters it and enforces the budget, but your prompts and completions do not
traverse our infrastructure. That distinction is why per-request cost attribution is possible
without us reading request bodies.

## Virtual keys capping Bedrock spend per team

Each team gets a virtual key with a hard USD budget and its own model allow-list. When a key
hits its cap it stops working. This is a limit, not an alert. Keys rotate and revoke
independently, so a leaked key is contained to one team's budget rather than the org's.

## Posture routing and semantic cache for Bedrock

Three postures decide which model serves a request: **Cost-saver**, **Balanced** and
**Quality-first**. Requests are classified into four semantic categories (coding, analysis and
reasoning, extraction and classification, chit-chat) and routed accordingly, so cheap models
take the traffic that does not need an expensive one.

A semantic cache sits in front, with strictness settings from strict (0.98 similarity) to loose
(0.90). On repetitive workloads the cache is usually a larger saving than the routing itself.