Model providers | Faustine Docs

A model provider is the account Faustine uses to call AI models for your workspace. Every workspace connects its own, in Settings > Model Provider. The provider bills you for the tokens, and Faustine does not charge for AI usage.

How it works

Analysis, automatic updates, the Advanced editor and chat all run on the workspace's provider. Until one is connected, the portal shows the banner Connect your AI provider to start. on every page, and the buttons that start AI work are disabled.

You connect a provider in a wizard with these steps:

Step	What you enter
Provider	One of the providers enabled for your workspace.
Credentials	The key or keys Faustine uses to call the provider.
Models	What each model Faustine uses is called on your provider. Shown for Amazon Bedrock, Azure AI Foundry, LiteLLM and Vercel AI Gateway.
Embeddings	The embeddings model that knowledge runs on. See Choose an embeddings provider.
Review & save	A recap, Test connection, and Save & activate.

Saving activates the provider immediately. Faustine stores the credentials encrypted and never shows them again. The portal shows a mask and a Configured badge in their place.

The providers you can choose are set for your workspace. When only one is enabled, the Provider step is already filled in and marked Set by your Faustine admin. To change them, contact Faustine.

Providers

Provider	What you need	Models step
Anthropic	An Anthropic API key.	No
Amazon Bedrock	A region and a Bedrock API key or an IAM keypair.	Required
Azure AI Foundry	A Foundry resource name and its API key.	Required
LiteLLM	The base URL of your LiteLLM gateway and an API key.	Required
Vercel AI Gateway	An AI Gateway API key from your Vercel team.	Optional overrides

Model mapping

Faustine refers to models by their first-party name, such as claude-opus-5. On Amazon Bedrock, Azure AI Foundry and LiteLLM the same model has a different ID, so the Models step asks for it.

The step has two groups:

Required for analysis & code generation lists the models Faustine needs for analysis, updates and the Advanced editor. Each one needs an ID on your provider.
Chat models is empty at first. The required models are already offered for chat. Select Add a chat model… to offer another one. A chat model that is not added is not offered to agents.

Test connection

Test connection checks the configuration you are editing without saving it. It runs in this order:

It verifies the credentials with the provider.
On Amazon Bedrock, Azure AI Foundry and LiteLLM, it sends one short request to every required model and every chat model you added. On Vercel AI Gateway, it sends one to every model in the list.
If embeddings are configured, it sends one embeddings request.
If a voice key is set, it checks the key with OpenAI.

The result has one line per check: Credentials verified, the number of models reachable, Embeddings working and, with a voice key, Voice key working. A model that fails shows the reason next to its row in the Models step.

Saving runs the credentials check again. Faustine refuses to save credentials the provider rejects. If the provider cannot be reached, the save goes through with a warning.

The model provider and an agent's chat model

The model provider is one setting for the whole workspace. It decides which account pays and which models can be reached.

The chat model is a setting of each Web Agent, in the agent's AI Model section. It picks which of the reachable models answers that agent's conversations. On Amazon Bedrock, Azure AI Foundry and LiteLLM, the models offered there are the required models plus the chat models you added in the Models step. See Choose the chat model.

Status and changes

Once a provider is saved, the tab shows a card with the provider, a status badge and one row per part of the configuration.

Badge	Meaning
Active	The credentials are stored and the provider is in use.
Credentials invalid	The provider started rejecting the stored credentials. AI features stop until you update them.
Credentials missing	The provider is saved without a credential.

Each row has an Edit button that opens the wizard at that step. In the Credentials step, an empty field keeps the stored value and a new value replaces it. A stored key is kept for the same destination only: when you change the Foundry resource or the host of the LiteLLM base URL, the form asks for the key of the new one. Re-validate runs the test on the stored configuration. Switch provider appears when more than one provider is enabled for the workspace. See Switch model provider.

The card has a Voice row. It holds the OpenAI API key that voice runs on, which is separate from the model provider and is kept when you switch provider. See Turn on voice.

Troubleshooting

Symptom	Cause	Fix
The banner Your AI provider credentials were rejected.	The provider refused the stored key during a run.	Select Fix credentials, enter a valid key and save.
The provider rejected the credentials when you save	The key is wrong, revoked or belongs to another account.	Create a new key on the provider and enter it again.
A model row shows Not found	The ID does not name a model, deployment or route on your provider.	Enter the ID your provider uses. The message under the row says which kind of ID it expects.
A model row shows Refused	The provider recognizes the ID and your account cannot use it.	Request access to the model on the provider's side.
An agent's AI Model section shows `No OpenAI models mapped`	The provider has a Models step and no OpenAI model was added under Chat models.	Add the model in the Models step and save.

Next steps

Connect your model provider: the first setup, start to finish.
Choose an embeddings provider: turn on knowledge for your agents.
Switch model provider: move the workspace to another provider.
Choose the chat model: pick the model an agent answers with.
Usage: see the tokens each feature consumed.