# AI Gateway — Govern LLM Spend Across OpenAI, Anthropic, OpenRouter and Bedrock

> Connect your LLM providers and put budgets on AI spend: live-verified keys, per-provider budgets with warning and critical alerts, virtual-key caps.

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

---

## Common questions

### Do my LLM prompts pass through ZopNight?

No. The gateway runs on hosted LiteLLM, which ZopNight configures, meters and budgets; prompt and completion bodies never reach ZopNight infrastructure. That is why per-request cost attribution works without anyone reading request bodies.

### Do I need a new API key to connect Bedrock?

No. Bedrock rides the AWS account you already connected and mints temporary credentials on demand. OpenAI, Anthropic and OpenRouter each take an API key, which ZopNight verifies live before saving so a revoked key cannot sit green.

### What happens when a virtual key reaches its budget?

It stops working. A warning event fires at 80% and the gateway enforces the hard cap, so the number is a limit rather than an alert.

AI spend is the fastest-growing line on the cloud bill. ZopNight connects to your LLM providers (OpenAI, Anthropic, OpenRouter, and AWS Bedrock) and brings model usage under the same budgets, alerts and dashboards as the rest of your infrastructure, including hard caps through gateway virtual keys.

## Live key checks, and Bedrock riding your AWS account

OpenAI, Anthropic and OpenRouter connect with an API key that ZopNight verifies live before saving (a revoked key can never sit green). Bedrock needs no key at all: it rides the AWS account you already connected, minting temporary credentials on demand. Optionally connect your LiteLLM-compatible AI gateway for per-team virtual keys with spend limits. Warnings fire at 80% and the gateway enforces the hard cap.

## Paste a key, or point at a connected AWS account

1. Settings > Integrations > AI Providers.
2. Pick the provider and paste its API key (or, for Bedrock, choose the connected AWS account and region).
3. Create budgets against providers or virtual keys and pick alert channels.

## Budgets, alerts and the optional LiteLLM gateway

Live-verified provider connections. Per-provider and per-virtual-key budgets with warning/critical events to any notification channel. Bedrock via existing AWS credentials (no new secrets). Optional LiteLLM gateway integration.

## Choosing between them

Pick by what you already have. If your spend is on one vendor, connect that vendor directly.
OpenRouter is the multi-vendor path: a single credential reaching several model families,
useful for comparing routing postures without provisioning accounts everywhere. Bedrock is worth
connecting when AI spend should land on the AWS bill you already govern.

They are not exclusive: connect several and let the routing posture decide which serves each
request.

## What is shared across all four

Virtual keys with hard budgets, per-request cost attribution, three routing postures and the
semantic cache behave identically regardless of provider. The provider choice affects which
models are available and where the bill lands, not how governance works.