---
title: Provider integrations
description: Discover and configure upstream credentials for Bring Your Own Key routing.
---

Manage provider credentials in the Zorveus dashboard under **Provider Credentials**. Application code does not call a provider-credential API. Use a Zorveus API key for inference, not the upstream provider secret.

<CardGroup cols={2}>
  <Card title="OpenAI" icon="cpu" href="/providers/openai">
    Configure an OpenAI API key and an `openai/*` model policy.
  </Card>
  <Card title="Google Gemini" icon="zap" href="/providers/gemini">
    Configure a Google AI Studio key for `gemini/*` routes.
  </Card>
  <Card title="Anthropic" icon="layers" href="/providers/anthropic">
    Configure an Anthropic API key for `anthropic/*` routes.
  </Card>
  <Card title="OpenRouter & Open Weights" icon="terminal" href="/providers/openrouter">
    Select OpenRouter if it is available under Provider Credentials in the dashboard.
  </Card>
</CardGroup>

## How routing selects a credential

The gateway considers active credentials whose model policy matches the request. `auto_resolve` credentials can participate automatically. `manual_grant_only` credentials require an active grant to the app connection. Lower routing-priority numbers are considered before higher numbers.

If `auto_fallback_to_wallet` is enabled and the wallet can fund the request, the gateway can add wallet-funded routing after eligible provider credentials. Successful BYOK usage records the provider cost and can also record a configured Zorveus platform fee.

## Verify a provider

1. Open **Provider Credentials** in the dashboard.
2. Select the provider and save the required credentials.
3. Confirm the model policy and routing mode in the dashboard.
4. Configure app-connection access when the mode requires a grant.
5. Select an available model from [the model catalog](https://www.zorveus.com/models).
6. Send an inference request with your Zorveus API key and inspect usage in the dashboard.

A provider route or wildcard alone is not evidence that every upstream model is available.
