Model providers
Connect a model provider to your workspace by storing its credential once. Forgebench injects it into each call at the gateway, so your code only ever holds a Forgebench API key.
Some models run only on your own provider account. Without a credential for that provider, the call is refused with a message naming the missing key.
Add a credential
Adding, replacing or removing a provider credential requires the admin role.
Open the Credentials tab
In the sidebar, expand Admin, choose API keys, then open the Credentials tab.
Click Add credential
The Add a provider key dialog opens. Pick the provider from the Provider list.
Fill in what the provider needs
Most providers take a single API key. Azure and Google Vertex AI ask for more; see What each provider needs.
Click Save provider key
The value is stored encrypted and never shown again. To change it, save a new one over it.
Single-key providers can also be added from Admin > Secrets. Use the Credentials tab for Azure, which needs an endpoint and a deployment list. To point at your own OpenAI-compatible server instead, click OpenAI-compatible endpoint… in the same dialog.
See Secrets for how credentials are stored, audited and removed.
What each provider needs
| Provider | What you enter | Model names |
|---|---|---|
| OpenAI, Anthropic, Gemini | An API key from the provider. | The ids in the model catalog |
| xAI, Mistral, DeepSeek, Groq, Cohere | An API key from the lab. These models run only on your key. | Prefixed with the lab, for example xai-grok-4.6 |
| OpenRouter | An OpenRouter API key. | openrouter-<model> |
| Vultr Inference | A Vultr Inference API key. These models run only on your key. | vultr-<model> |
| Azure OpenAI | Endpoint, API version, deployment list, and an API key or a Microsoft Entra ID service principal. See Azure OpenAI & AI Foundry. | azure-<deployment> |
| Azure AI Foundry | The same fields as Azure OpenAI, for non-OpenAI models deployed in Azure AI Foundry. | azureai-<deployment> |
| Google Vertex AI | An Express API key, a service-account JSON, or a Workload Identity Federation configuration. See Google Vertex AI. | vertex-<model> |
| Amazon Bedrock | A Bedrock API key from your own AWS account. See Amazon Bedrock. | bedrock-<model>, mantle-<model> |
Choose which models developers can use
On the Credentials tab, open a credential card's menu and choose Manage models. It lists every model Forgebench serves for that provider, grouped by vendor, with a checkbox each.
- Check or uncheck models, or use Enable all and Disable all on the filtered list, then click Save.
- With every model checked, the dialog reads "All N enabled — models ForgeBench adds later are included too." Any narrower selection is saved as an exact list, so new models stay off until you turn them on.
- With nothing checked, developers can't call any model through that credential.
The selection applies to the model list, to key creation, and to every call, including from keys created earlier. Only admins can change it.
For Google Vertex AI, the list is Google's Model Garden, and each model links to its Google Cloud page. Turn on in Google asks Google to enable that model in the project from your service-account credential. Google only does this when the model's terms are already accepted, and Google bills its usage.