Skip to main content

Providers and Models

A provider is an upstream account the gateway calls with your organization's key. A model is an entry users can call by name, pointing at one provider. Only models listed on the Models tab can be called.

Both tabs are under Control > LLM Gateway. Changing them needs the Organization > Edit Settings permission.

Providers​

The Providers tab lists each provider with its type, the last characters of its key, its base URL if you set one, and how many models use it.

Add a provider​

  1. Go to Control > LLM Gateway > Providers.

  2. Select Add Provider.

  3. Choose the Type:

    TypeUse it for
    AnthropicClaude models, with an sk-ant-… key
    OpenAIOpenAI models, with an sk-… key
    OpenAI-compatibleAny service that speaks the OpenAI Chat Completions API with a bearer key, such as Groq, Together, OpenRouter, or a self-hosted vLLM

  4. Enter a Name. It defaults to the type.

  5. Enter a Base URL if you need one. It is optional for Anthropic (default https://api.anthropic.com) and OpenAI (default https://api.openai.com/v1), and required for OpenAI-compatible, for example https://api.groq.com/openai/v1.

  6. Paste the API key.

  7. Leave Enabled on, and select Save.

The key is encrypted when you save it and is never shown again. The provider card shows only a hint such as sk-ant-…a1b2.

Change or remove a provider​

  • Rotate the key. Select the pencil, paste a new API key, and save. Leave the field blank to keep the stored key.
  • Pause it. Turn off the provider's switch. A disabled provider stops serving all of its models until you turn it back on.
  • Delete it. Select the trash icon. You have to remove or move the provider's models first.

The Type cannot be changed after the provider is created. Add a new provider instead.

Base URLs on hybrid

On a hybrid deployment the gateway calls providers from your own cluster, so a base URL can point at a model server on your internal network, such as a vLLM service.

Models​

The Models tab lists each model with its alias, display name, provider and provider model id, prices, and the groups that may use it. The tab asks you to add a provider first if there are none.

Add a model​

  1. Go to Control > LLM Gateway > Models.

  2. Select Add Model.

  3. Choose the Provider.

  4. Select Browse models under Catalog to search the models that provider offers. Picking one fills in the provider model id, the display name, and the prices. You can also type the fields yourself for a model that is not in the catalog.

  5. Check the fields:

    FieldWhat it is
    Provider model idThe id the gateway sends to the provider, for example claude-sonnet-4-5 or gpt-5
    AliasThe name users and clients ask for. It defaults to the provider model id. Letters, digits, and . _ : / @ -, up to 100 characters, unique in the organization
    Display nameA readable label, for example Claude Sonnet
    Price per 1M tokens (USD)Input, Output, and optionally Cached input. Set input and output together, or leave all three empty
    Who can use itThe groups allowed to call the model. Leave it on Everyone to allow every user

  6. Leave Enabled on, and select Save.

Prices are what budgets and usage count. A model with no price records every call at $0, so it never uses up a budget.

How a request finds its model​

The gateway matches the model a request names in this order:

  1. A model whose Alias is that name.
  2. Otherwise, a model whose Provider model id is that name. This lets clients that send vendor model ids, such as Claude Code and Codex, work with no extra client configuration.
  3. If the request names no model, the default model.

Only enabled models on an enabled provider that has a key count. If a match exists but the caller is not in any of its groups, the call is refused with You don't have access to model …. If nothing matches, it is refused with Model … is not available in your organization's LLM gateway. Both messages name the model the client asked for, which tells you what to add. See Troubleshooting.

Offer the models your clients ask for

Claude Code sends the model ids it was built with, and also calls a small Haiku model for background tasks. If you only list one Sonnet model, Claude Code's other calls are refused. Add each model id that appears in a refusal, or point an alias at the model you want to serve instead.

The default model​

Select the star on a model to make it the default. The gateway uses it when a request does not name a model. The first model you add becomes the default automatically.

Change or remove a model​

  • Rename the alias. Edit the model. Budgets and the default that use the old alias are updated with it. Clients that ask for the old alias are refused until they use the new one.
  • Pause it. Turn off the model's switch.
  • Delete it. Select the trash icon. If a budget covers the model or falls back to it, update that budget first.

Organizations on the original Claude Code setup​

Organizations that used the gateway before the model list existed see a notice that any Claude model is allowed. Until you add a model, the gateway passes any Claude model through to the original Anthropic key, unpriced. Adding the first model switches the organization to the model list: from then on, only listed models can be called, and calls for any Claude model you have not listed yet are refused. Make a list of the models your users rely on first, then add them one after the other.

On this page