Choose an embeddings provider

Embeddings are what Faustine uses to index the documents you upload and to search them from chat. They are part of the workspace's model provider configuration. While embeddings are Not configured, documents cannot be uploaded and agents cannot search them.

Before you begin

A model provider is saved for the workspace, or you are in its setup wizard at the Embeddings step.
You have a key for one of the options your provider offers.

Embeddings options

The options depend on the workspace's model provider.

Model provider	Options
Anthropic	OpenAI API key
Amazon Bedrock	OpenAI API key, Azure OpenAI (or OpenAI-compatible)
Azure AI Foundry	Azure OpenAI, OpenAI API key
LiteLLM	Use my LiteLLM gateway, Use a separate OpenAI key
Vercel AI Gateway	Use my Vercel AI Gateway, Use a separate OpenAI key

Every provider also has Not configured.

The LiteLLM and Vercel AI Gateway options reuse the credentials of the model provider, so there is no second key. They are selected by default on those providers.

Set the embeddings provider

To set or change embeddings on a saved provider:

In the portal, go to Settings > Model Provider.
In the Embeddings row, select Edit.
Select an option.

Fill in the fields for that option:

Field	Shown for	Value
Base URL	Azure OpenAI	Your Azure OpenAI endpoint, such as `https://<RESOURCE>.openai.azure.com/openai/v1/`. On Azure AI Foundry this is a different resource from the Foundry one.
Model	OpenAI, LiteLLM, Vercel AI Gateway	The embeddings model ID. The default is `text-embedding-3-small`, written `openai/text-embedding-3-small` on Vercel AI Gateway.
Deployment name	Azure OpenAI	The name of your embeddings deployment.
API key	OpenAI, Azure OpenAI	A key used only for embeddings.

Select Continue.
In Review & save, select Test connection. The result includes Embeddings working with the model name.
Select Save changes.

The Embeddings row now shows the provider and the model, with a Key configured badge when the option has its own key.

Faustine stores vectors at 1536 dimensions. Use a model that produces vectors of that size. Base URL has to be an https URL on a public hostname.

When you edit embeddings later, an empty API key keeps the stored key. If you change the option or the Base URL, enter the key again.

Turn embeddings off

Select Not configured in the Embeddings step and save. This removes the embeddings configuration and its key, and knowledge stops working for every agent in the workspace.

Troubleshooting

Symptom	Cause	Fix
`Embeddings failed — The embeddings provider rejected the credentials.`	The key is wrong or revoked.	Enter a valid key for the option you selected.
`The embeddings endpoint does not serve "<model>".`	The model or deployment name does not exist on that endpoint.	Enter the exact model ID or deployment name.
`embeddings.baseUrl must be a public hostname` when you save	The URL points to `localhost`, an IP address or an internal name.	Use the public `https` address of the endpoint.
The Embeddings row shows Key missing	The embeddings option was saved without a key.	Select Edit and enter the key.

Next steps

Model providers: the rest of the provider configuration.
Add knowledge from files: upload documents once embeddings work.
Switch model provider: the embeddings options change with the provider.