Creating an Agent in Generative AI Agents

Create an agent with one or more tools, and set up an agent endpoint in the Generative AI Agents service.

Important

Before you start, ensure that you have permission to manage the service. To select the model that the agent uses, see User Access to Model Selection in Agents.

Navigating in the Console

In the navigation bar of the Console, select a region with Generative AI Agents, for example, US Midwest (Chicago) or UK South (London). See Regions with Generative AI Agents for the regions in which Generative AI Agents is hosted.
On the Agents list page, select Create agent. If you need help finding the list page, see Listing Agents.

Entering Basic Information

Enter the following information:
- Name: A name that starts with a letter or underscore, followed by letters, numbers, hyphens, or underscores. The length can be from 1 to 255 characters.
- Compartment: Select the compartment that you want to store the agent in.
- Description: An optional description.
- Welcome message: An optional message that greets users when they start the chat with the agent. If you leave this field blank, the agent starts the conversation with the default welcome message that's displayed as a hint text in the text box.
- Routing instructions: An optional prompt about routing the tools that can influence or affect the agent's behavior. For example, if an agent has several tools, a prompt could be "Always invoke the SQL tool first."
Model Selection: Select from the following options:
- Routing LLM type
  - Default: A model used by the Generative AI Agents service
  - Generative AI model: A model hosted in the Generative AI service
  - Generative AI endpoint: A model hosted on a dedicated AI cluster endpoint in the Generative AI service
Tip

For guidance and a list of supported models and regions, see the Model Selection in Generative AI Agents page.
If you selected Generative AI model select a model. Based on the region that you're creating the agent in, the list of models populates.
If you selected Generative AI endpoint, select the Generative AI endpoint compartment and the endpoint.
(Optional) Update the default hyperparameters.
Hyperparameters and their default values differ by model type.
(Optional) Select Show tagging options and add one or more tags to the agent. If you have permissions to create a resource, then you have permission to update its tags. If you need help, see Tags and Tag Namespace Concepts.

Adding a Tool

Important

This step doesn't create the tools. The tools are created after you create the agent.

Set up the creation of one of the following tools.
- RAG tool
- SQL tool
- Function calling tool
- API endpoint calling tool
- Agent tool (Agent as a tool)
(Optional) Repeat the previous step to set up the creation of another tool.

See Limits and Limitations for Generative AI Agents for the number of tools allowed for an agent.

Setting Up an Agent Endpoint

Important

This step sets up an endpoint with common endpoint features. To enable and disable other features not included in this step, such as multi-language support edit the endpoint.

(Optional) Select Automatically create an endpoint for this agent to create an endpoint when the agent is created.
If you selected to create an endpoint, then select whether to enable the following optional features for the endpoint:
- Enable human in the loop: Whether human input is required such confirming an agent's generated response before the response can be used.
- Content moderation for input (user prompt), output (generated response), or both:
  - Disable: Don't apply content moderation.
  - Block: Help identify and apply content moderation.
  - Inform: Don't apply content moderation, but aim to inform the user if the agent detects content that needs moderation.
- Prompt injection (PI) protection:
  - Disable: Don't apply PI protection.
  - Block: Help identify and protect against prompt injection.
  - Inform: Don't apply PI protection, but aim to inform the user if the agent detects content that needs PI protection.
- Personally identifiable information (PII) protection for input (user prompt), output (generated response), or both:
  - Disable: Don't apply PII protection.
  - Block: Help identify and protect PII.
  - Inform: Don't apply PII protection, but aim to inform the user if the agent detects content that needs PII protection.

Reviewing the Details

Review the details. If you need to change any of the options, select Previous, update the information, and navigate back to the review step.
Select Create agent.
Read the Llama 3 License Agreement and AUP and select the checkbox to agree the Llama 3 License Agreement and Acceptable Use Policy, and then select Submit.