Manage and use AI models

This guide explains how to create models in the GenAI gateway, use them in a workflow and share them with other users and teams.

Before you start

See Team roles and permissions to understand who can create and use GenAI models.

Navigate to the GenAI models page

GenAI model scope determines how to access the model page:

• Team GenAI models — open the team's page, then select GenAI models from the left menu.
• Global GenAI models — open the global administration page, then select Global GenAI models from the left menu.

Creation, editing and deleting models works the same for all scopes and is described below.

Create a model

Only team admins can create team models. Only global admins can create global models.

1. Navigate to the GenAI models page.
2. Click New GenAI model.
3. Select the Platform.
4. Fill in the platform-specific fields. See Supported platforms.
5. Click Create.

The model appears in the models table.

Edit a model

Only users with can edit permissions can edit a model.

1. Navigate to the GenAI models page.
2. Click the three-dot menu on the model entry.
3. Select Edit.
4. Update the values and click Save changes.

Delete a model

Irreversible operation

All the workflows that are using the model will fail. This cannot be undone.

Only users with can edit permissions can delete a model.

1. Navigate to the GenAI models page.
2. Click the three-dot menu on the model row.
3. Select Delete.
4. Type the model name to confirm, then click Delete permanently.

Manage access to a model

Only team admins can manage access to a team model. Only global admin can manage access to a global model.

1. Navigate to the GenAI models page.
2. Click the three-dot menu on the model entry.
3. Select Manage access.
4. Type the team to share with.
5. Choose an access level: can use or can edit.
6. Click Apply.

Use a model in a workflow

1. Add the KNIME Authenticator node to the workflow.
2. Connect it to the Knime LLM Selector or the Knime Embedding Model Selector node.
3. Select the desired model in the configuration dialog.
4. Connect it to downstream nodes such as the LLM Prompter or the Agent Chat Widget node.

List all available models

1. Add the KNIME Authenticator node to the workflow.
2. Connect it to the Knime AI Model Lister.
3. Execute the node.

The output table shows all the models accessible by the authenticated user.

Use models in deployment

A deployed workflow, such as a data app, may run under a different identity than the one used to build it. For example, an external, unauthenticated user opening a data app. In that case, configure the KNIME Authenticator node to explicitly authenticate as a user with access to the model.

WARNING

If the Authenticator node isn't configured with explicit credentials, the deployment fails to access the model for users who aren't authenticated as someone with access.

1. Create an application password for a user with access to the model.
2. Add the KNIME Authenticator node to the workflow.
3. In the configuration dialog, enter the API URL of the Hub, then provide the application password from step 1.
4. Connect it to the Knime LLM Selector or the Knime Embedding Model Selector node.
5. Select the desired model in the configuration dialog.
6. Connect it to downstream nodes such as the LLM Prompter or the Agent Chat Widget node.

Select a model dynamically with a flow variable

By default, you select a model manually in the Knime LLM Selector or the Knime Embedding Model Selector node dialog. The name shown in the picker, for example a team name followed by the model name, is a display label and cannot be used directly in a flow variable.

Use the Knime AI Model Lister to List all available models and find the underlying identifier for a model. Pass this identifier into a string flow variable and use it to override the model selection control in the Flow Variables button.

Keep the following in mind:

• The identifier is tied to one specific model. It is not portable between users or instances just because two models share the same display name.
• Knowing a model's identifier does not grant access to it. The user or team executing the workflow still needs Use or Edit rights on the model, see Manage access to a model.