AI Models & Providers
How AI actions choose a provider and model, where credentials are stored, and which providers are available for each kind of AI work.
Overview
Every AI action runs against a provider (the AI vendor, such as OpenAI or Google) and a model (a specific model offered by that provider). You pick both directly on the action using the model selector, and the engine calls the matching provider at run time using the API key you configured for the current environment. See AI Actions for the actions that use these settings.
Supported providers
The selector only offers providers that support the action's capability. For example, a text-to-speech action will not let you choose a provider that has no speech voice. The matrix below shows which providers are offered for each capability.
| Provider | Chat / text | Image generation | Image recognition | Speech‑to‑text | Text‑to‑speech | Moderation |
|---|---|---|---|---|---|---|
| OpenAI | Yes | Yes | Yes | Yes | Yes | Yes |
| Azure OpenAI | Yes | Yes | Yes | Yes | Yes | Yes |
| Anthropic | Yes | – | Yes | – | – | – |
| Google (Gemini) | Yes | Yes | Yes | Yes | – | – |
| xAI (Grok) | Yes | Yes | Yes | – | – | – |
Selecting a provider and model
The model selector has two parts. The provider is normally chosen from a drop-down, but you can also resolve it at run time from a context key or a variable when a single rule set needs to switch providers dynamically. The model is a free text field where you enter the provider's model code (for example, the identifier you copy from the provider's model list); it too can be supplied as a constant or resolved at run time.
Model codes are provider-specific, so consult the provider's own documentation for the exact value. A help dialog inside the selector links out to each provider's model list. Be sure to choose a model whose capabilities match the action — a model trained for moderation for a moderation action, a vision-capable model for image recognition, and so on.
For Azure OpenAI, the model field expects the deployment name you created in your Azure OpenAI resource, not the underlying model name. The resource endpoint URL is configured separately on the provider settings page.
Configuring provider credentials
Provider API keys are managed on the Provider Settings page and are scoped to the current environment, so a development environment and a production environment can use different keys. Each provider has its own key entry: OpenAI, Anthropic, Gemini, xAI, and Azure OpenAI. Azure OpenAI additionally requires its resource endpoint URL.
- Secrets are write-only — once saved, a key cannot be read back in plain text, only replaced or deleted.
- Editing or deleting a saved key requires the administrator role; other members can view that a key is present.
- If an action references a provider with no configured key, the call will fail at run time.
A note on credits
Running your own AI actions consumes your provider account's quota under your API key. Some in-app AI features — such as the AI Template Generator and AI Project Generator — instead consume organization credits. See Credits & Billing for details.