> ## Documentation Index
> Fetch the complete documentation index at: https://docs.opper.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Gateway

> One endpoint in front of 300+ models. Drop in your SDK, route by goal, keep data in the EU, and see one bill across every provider.

The AI Gateway is the request path. Your app sends a call, the gateway picks a model, runs it, and returns the result. One endpoint, one API key, [300+ models](/capabilities/models) behind it, all hosted in the EU.

It replaces wiring up each provider yourself, so there are no per-provider SDKs, keys, billing, or rate limits to manage. To swap a model, change a string.

## What the gateway gives you

<CardGroup cols={2}>
  <Card title="One API, every model" icon="plug">
    Reach OpenAI, Anthropic, Google, Mistral, and 300+ more through a single endpoint and key.
  </Card>

  <Card title="Every modality" icon="shapes" href="/build/multimodal/overview">
    Not just text — generate images, speech, and video, transcribe audio, and run realtime voice.
  </Card>

  <Card title="Drop-in SDKs" icon="code">
    Keep the OpenAI, Anthropic, or Google AI SDK you already use. Change the base URL, nothing else.
  </Card>

  <Card title="Switch models without code" icon="signs-post">
    Set a default model per project with a Route rule. Change it from the platform, no deploy.
  </Card>

  <Card title="Bring your own keys" icon="key">
    Register your own provider deployments and API keys. They work like any other model.
  </Card>

  <Card title="EU by default" icon="shield-check">
    Restrict calls to EU-only or zero-retention providers. Enforced at the gateway.
  </Card>

  <Card title="One bill, every provider" icon="receipt">
    Every response carries its cost. Spend across all providers in one place.
  </Card>

  <Card title="Observability built in" icon="eye">
    Every call is metered: cost, latency, and tokens. Turn on retention for full traces.
  </Card>
</CardGroup>

## One API, every model

Address any model as `provider/model`. The same call works whether you point it at OpenAI, Anthropic, or a model hosted in the EU.

```bash theme={null}
openai/gpt-5            # OpenAI
anthropic/claude-opus-4-7   # Anthropic
mistral/mistral-large   # Mistral, EU-hosted
```

Some models are served by more than one provider (for example the same Claude model via Anthropic, Azure, or Bedrock). Opper balances across them for you, so a single name keeps working even when one provider is busy.

To call a model, point your SDK at `https://api.opper.ai/v3/compat` — see [Drop-in SDKs](/build/gateway/drop-in-sdks). The same endpoint covers text, [structured output](/build/gateway/structured-output), [tool calling](/build/gateway/tools), and [multimodal input](/build/multimodal/vision-pdfs), with the same routing, governance, and tracing on every call.

The gateway isn't text-only. Dedicated endpoints generate [images](/build/multimodal/images) (`POST /v3/images`), [speech and transcripts](/build/multimodal/audio) (`POST /v3/audio/*`), and [video](/build/multimodal/video) (`POST /v3/videos`), and [realtime voice](/build/realtime/quickstart) runs over a WebSocket — all behind the same key and governance. See [Multimodality](/build/multimodal/overview).

## Pin and switch models without touching code

Keep model names out of your application. Set a default once and change it from the platform:

* **[Route](/control-plane/route)** pins a default model per organization or project. Edit one rule to move a project to a different model.
* **[Custom models and aliases](/capabilities/custom-models)** let you register your own deployments and point a stable name (like `production/main`) at whatever model you choose.

Callers can still pass an explicit `model` for per-call trade-offs, as long as [Comply](/control-plane/comply) allows it.

## Bring your own keys

You can go beyond Opper's hosted models. Register your own provider deployments and API keys (for example a private Azure OpenAI deployment), and they show up in the catalog alongside built-in models with the same routing, governance, and tracing.

```bash theme={null}
curl -X POST https://api.opper.ai/v2/custom-models \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPPER_API_KEY" \
  -d '{
    "name": "example/my-gpt",
    "provider": "azure",
    "model_id": "gpt-production",
    "api_key": "your-provider-key",
    "params": {"api_base": "https://your-deployment.openai.azure.com/", "api_version": "2024-06-01"}
  }'
```

Call it like any other model: `example/my-gpt`. See [Custom models](/capabilities/custom-models) for the full setup.

## Keep data in the EU

The gateway is hosted exclusively in the EU. The only thing that leaves is the model call itself, and you control where that goes.

With a [Comply](/control-plane/comply) rule you can restrict every call to:

* EU-only providers (Mistral, Azure EU, and others)
* Specific regions or countries
* Zero-retention providers for the strictest workloads

If you need to store nothing at all, a Zero Data Retention rule keeps request and response content off disk entirely. Only the usage counters needed for billing are kept.

Anything that breaks a rule is rejected at the gateway before it reaches a provider. See [Security](/overview/security) for more.

## One bill, full visibility

Every response includes its `cost`, computed the same way across all 300+ models. The gateway aggregates spend and usage across providers, so you get one bill and one place to watch it.

Every call is metered (cost, latency, tokens) and shows up in **Analytics** for spend over time. Turn on a [retention rule](/control-plane/comply) and you also get the full [trace](/overview/concepts) for each call, with inputs, outputs, and every step.

## How it fits with the control plane

The gateway runs your requests. The [Control Plane](/control-plane/overview) sets the rules it follows.

|                   | AI Gateway                                  | Control Plane                        |
| ----------------- | ------------------------------------------- | ------------------------------------ |
| **Role**          | Runs every request                          | Sets the rules                       |
| **You use it by** | Making calls                                | Writing rules in the platform        |
| **Examples**      | Model access, routing, EU enforcement, cost | Observe, Route, Guard, Comply, Steer |

In short, the Control Plane is where you configure, and the Gateway enforces those rules on every call.

## Security

The gateway also enforces data residency. The next page covers how Opper is hosted and protected.

<CardGroup cols={2}>
  <Card title="Security" icon="shield-check" href="/overview/security">
    EU hosting, sub-processors, encryption, and deletion.
  </Card>

  <Card title="Models" icon="brain" href="https://opper.ai/models">
    The full catalog. EU-hosted models marked.
  </Card>

  <Card title="Build on Opper" icon="signs-post" href="/build/overview">
    Ready to build? Text, multimodal, and voice.
  </Card>

  <Card title="Drop-in SDKs" icon="plug" href="/build/gateway/drop-in-sdks">
    Use the OpenAI, Anthropic, or Google AI SDK you already have.
  </Card>
</CardGroup>