Skip to main content
Version: User Guides (Cloud)

Integrate with Model Providers

A text embedding or reranking model hosted by an external provider cannot be called from Zilliz Cloud until the provider can authenticate requests from your project. A model provider integration stores the provider-issued credential at the project level and gives Zilliz Cloud an integration ID that text embedding and reranking features can reference. This avoids placing credentials in individual Function or Ranker configurations.

📘Notes

Creating a model provider integration does not incur charges. The external provider may charge for model inference, and sending data to the provider may incur data transfer costs.

Supported model providers​

The following model providers can be integrated with Zilliz Cloud:

Model provider

Supported Zilliz Cloud features

Required credential

OpenAI

Text Embedding Function

API key. To obtain one, see the OpenAI API quickstart.

Cohere

Text Embedding Function and model-based Ranker

API key. To obtain one, see API Keys and Rate Limits.

Voyage AI

Text Embedding Function and model-based Ranker

API key. To obtain one, see API Key and Python Client.

Hugging Face

Text Embedding Function and Hugging Face Ranker

User Access Token with Make calls to Inference Providers permission. To obtain one, see User Access Tokens.

📘Notes

When selecting a model from an external provider, verify that the provider currently serves the model for the required task. Model availability, task support, stability, latency, and output quality depend on the provider and selected model. Evaluate these properties for your workload before using the model in production.

Before you start​

Before creating a model provider integration, make sure that:

  • You have Organization Owner or Project Admin permissions for the target Zilliz Cloud project. If you do not have sufficient permissions, contact your Zilliz Cloud Organization Owner.

  • You have the credential required by the selected model provider. See Supported model providers.

Create an integration in the Zilliz Cloud console​

To create a model provider integration:

1

Log in to the Zilliz Cloud console.

2

On your project page, navigate to Integrations from the left-side navigation pane.

3

Under the Model Providers section, click + Integration.

4

In the dialog box that appears, configure Basic Settings:

  • Model Provider: Select the model provider to integrate with.

  • Integration Name: A unique name for this integration (e.g., test).

  • Integration Description*(optional)*: A description for this integration (e.g., for model provider).

  • Provider (Hugging Face only): Keep the default value, hf-inference. Hugging Face Text Embedding and Hugging Face Ranker currently support only this Inference Provider.

5

Click Next. You'll be redirected to the Credential Information step:

  1. Enter the credential required by the selected model provider. For Hugging Face, enter your User Access Token in the Hugging Face Access Token field.

  2. Click Validate Integration to check the connection. Once its status changes to Successful, proceed to the next step.

6

Click Add.

Once created, the integration becomes available for use by model-based functions and rankers.

Manage integrations​

After an integration is created, you can manage it from the Integrations page:

  • Obtain your integration ID

    The integration ID is required when a Text Embedding Function or model-based Ranker uses the integration.

  • View integration details

  • Edit the integration name or description

  • Remove the integration when it is no longer needed

📘Notes

If an integration is removed or becomes invalid, collections or rankers that reference it may fail during insert or search operations until the integration is updated or replaced.

Next steps​

After creating a model provider integration, you can:

  • Use it with a Text Embedding Function to convert text into dense vectors.

  • Use a model-based Ranker to rerank search results.

For detailed instructions, refer to: