---
title: Models and Routing
description: Learn about provider-prefixed model names, wildcard policies, credential resolution, and managed routing.
---

Zorveus provides a unified OpenAI-compatible routing layer across OpenAI, Anthropic, Google Gemini, NVIDIA NIM, and OpenRouter.

---

## Model Naming Convention

Model identifiers in Zorveus use a standardized `provider/model_name` convention:

| Model ID | Target Provider | Description |
| :--- | :--- | :--- |
| `openai/gpt-4.1-mini` | OpenAI | Fast, cost-efficient reasoning model |
| `openai/gpt-4.1` | OpenAI | High-intelligence flagship reasoning model |
| `anthropic/claude-3-7-sonnet` | Anthropic | Hybrid reasoning and coding model |
| `gemini/gemini-2.5-flash` | Google Gemini | Ultra-high speed multimodal model |
| `openrouter/deepseek-r1` | OpenRouter | Open-weights deep reasoning model |

<Callout type="tip" title="Unprefixed Aliases">
  Standard aliases like `gpt-4o-mini` or `claude-3-5-sonnet` automatically resolve to their canonical provider-prefixed routes when allowed by your connection model policy.
</Callout>

---

## Model Policies and Wildcards

When configuring an App or App Connection, you specify allowed model patterns:
- `*`: Allows all current and future models available on the platform.
- `openai/*`: Allows any model hosted by OpenAI.
- `anthropic/claude-3-*`: Allows any Claude 3/3.5/3.7 variant.
- `openai/gpt-4.1-mini`: Restricts the key strictly to a single model.

---

## Credential Resolution Priority

When an inference request arrives:
1. **Connection-Specific Override**: Checks if the app connection has an explicitly bound provider key.
2. **Organization BYOK Credential**: Checks if the organization has an active BYOK credential matching the requested model policy (sorted by priority).
3. **Zorveus-Funded Credential**: Falls back to Zorveus-managed enterprise provider infrastructure (debiting the organization wallet).


Use `GET /v1/models` with the API key that will make the request. The response expands allowed policy patterns into concrete model IDs. Send one of those concrete IDs in the model request.