C3 AI Documentation Home

Set Up LLMs and Embedders Via UI

Before you can create agents from C3 AI Studio, you need to set up an LLM client. This guide shows you how to configure clients for Azure OpenAI, Azure AI, AWS Bedrock, or Google Vertex AI using the Model Hub user interface.

About Model Clients

A model client connects your C3 Agentic AI Platform agents to external large language model and embedder services. The client handles three key components:

  • Authentication: Credentials that verify your access to the model service.
  • Model: The specific language or embedding model you want to use (for example, GPT-5, Claude, or Gemini).
  • Client: The wrapper that combines authentication and model configuration into a reusable interface.

You complete three main steps to set up a model client:

  1. Configure authentication.
  2. Create and configure the model client.
  3. Test the newly created client.

For instructions on setting up model clients via the UI, follow the steps in this guide. If you prefer to create and manage clients programmatically, see Set Up LLMs and Embedders Programmatically for instructions on setting up LLM and Embedding Clients via code.

Model Hub UI

The Model Hub UI allows you to manage model clients within an application. Two different sections exist:

  • Clients: Configure and test different models and clients.
  • Authentication: Examine existing authentication configurations or create new ones.

Step 1: Configure Model authentication

Before creating a model client, you must set up authentication credentials for your chosen provider. The Model Hub authentication section stores the credentials and connection details needed to access external model services.

Create an authentication object

  1. Open Model Hub and select Authentication.

  2. Select Create Authentication.

  3. Select a Provider type (Azure OpenAI, Azure AI, AWS Bedrock, Google Vertex AI, or OpenAI).

    If there are any custom providers defined in the app or package, those will show up here. You can mix in the Genaicore.Llm.Auth Type to define a custom provider.

  4. Enter field values:

    • Provider-specific fields (required): API keys, endpoints, credentials, and other fields required by your chosen provider.
    • Supported Model Names (optional): Comma-separated list of model names compatible with this authentication. Example: gpt-4, gpt-4-turbo, text-embedding-ada-002. When you create an LLM client, only models in this list can be selected.
    • Supported Model Aliases (optional): Custom names or aliases for your models. You specify pairs of actual model names and their aliases. For example, map the alias gpt-5 to the actual model name gpt-5-2025-06-05. When you select a model in a client, the system uses the alias to derive the actual model in API calls.
    • Config Override: The scope level where this authentication is stored. When configured via the UI, the value is always set to APP.

  5. Select Create.

Step 2: Create and configure the client

After you configure authentication, create the model client in the Model Hub UI.

Create a model client

  1. Open Model Hub and select Models.
  2. Select Create Model Client. A modal opens up.
  3. In Create Model Client, enter a Name, select a Type, and optionally enter a Description. Supported client types include:
  • Completion Client: Uses traditional Chat Completions-compliant APIs for message-based generation.
  • Response Client: Uses OpenAI Responses-compliant APIs for response-oriented generation workflows.
  • Embedding Client: Generates text embeddings for similarity search and retrieval workflows.
  1. Select Next.
  2. Under Configure and Test Client, select a Provider (Azure Open AI, AWS Bedrock, Google Vertex AI, or OpenAI), an Authentication object (created in the previous step), and a Model from the list (if Supported Model Names or Supported Model Aliases were populated for the selected authentication), or enter a model name manually.
  3. Select Create.

The model parameter must match an officially supported model name from your chosen provider.

Model name for Azure needs to match the deployment name from the Azure AI Foundry.

Model Hub Models Page

Choose Models under Model Hub to see a list of configured clients. From here, you can filter the list of clients by their Type, Provider, Authentication, Model, and Config Override. You can also search for clients by name.

Edit a Model Client

After creating a model client the UI shows the model client configuration page. You can also select a specific client displayed in the model clients grid. To change the client configuration, select the edit button at the top right of the client configuration page. In addition to the fields entered during creation, you can configure:

  • Temperature: Controls response randomness. Lower values produce more deterministic and focused responses, while higher values produce more varied and creative responses.

Some models have a max temperature of 1 while others have a max temperature of 2. Choose your value based on the specific model you use.

  • Max Completion Tokens: Sets a strict upper bound on generated tokens for a response (including internal reasoning tokens when applicable).

Some providers use max tokens and some use max completion tokens. The C3 AI Platform will handle using the correct parameter for different supported models.

  • Custom Parameters: Allows users to specify options not exposed as dedicated fields in the form. These are saved as part of the client's default options. Make sure the custom parameters entered are supported by the model being configured.

Saved values act as default settings for that client. Request-level options can still override these defaults at runtime.

Step 3: Test the model client

After creating your model client, verify that it can successfully connect to the language model service.

Test client connection

  1. Open the model client details page.
  2. Select Test Connection.
  3. Review the result.

If validation fails, verify configurations on the selected authentication object and model client are correct. The UI also exposes the specific error logs in the failed response for further introspection by users.

After successfully creating the model client, see Create Agents for how to use an LLM client when creating an agent.

Additional operations

Duplicate a model client

  1. Open the model client details page.
  2. Select the ellipsis menu icon (⋮) on the right and select Duplicate.
  3. Enter a new Name and optional Description.
  4. Select Duplicate.

The duplicate starts with the source client configuration and can be edited independently.

Delete a model client

  1. Open Model Hub > Models.
  2. Select the model client.
  3. Select the ellipsis menu icon (⋮) on the right and select Delete.
  4. Confirm deletion.

If deletion is blocked by permissions or dependencies, resolve the warning shown in the confirmation dialog.

Generate Python and JavaScript usage snippets

  1. Open the model client details page.
  2. Select the <> icon.
  3. Choose Python or JavaScript.
  4. Copy the generated snippet for your integration.

Handle common error states

Error stateWhat to check
Missing authenticationSelect a valid authentication object for the chosen provider
Invalid authenticationRecheck API key, endpoint, region, and provider-specific fields
Model not availableSelect a supported model for the selected provider and auth
Duplicate nameUse a unique name for the client or authentication object
Permission deniedVerify role permissions

See also

  • Configure Models (genAiSearch: genai-ug-configure-models.c3doc.md)
  • Manage Users (genAiSearch: genai-ug-manage-users.c3doc.md)
  • Configure and use Large Language Models (genAiSearch: genai-llm-creds-setting.c3doc.md)
Was this page helpful?