Overview of the agent builder

The agent builder is a unified studio where you can customize, test, and refine your agents. It combines the power of AI, automation, and integration to handle tasks that might normally require human effort or complex software development.

Use the agent builder to build agents that can act autonomously to achieve goals, decide, and influence its environment based on internal models or reasoning processes. The agents that you build can:

  • Read, interpret, and respond to human input.
  • Automate workflows from sending emails and updating databases to approving requests and generating reports.
  • Connect systems to bridge the gap between tools like CRMs, ERPs, spreadsheets, APIs, and cloud services.
  • Choose the best course of action based on real-time data.
  • Improve over time to offer smarter responses and better outcomes.

How the agent operates

On watsonx Orchestrate, an agent usually operates in its environment in a form of perception-reasoning-decision-action loop, where the agent can continuously adapt and act intelligently. In this loop, the agent:

  • Perceives the environment.
  • Interprets the data and updates agents' internal state.
  • Reasons about what to do and decides on an action.
  • Runs the action.

Deploy the agents that you create as single-agent solutions that independently manage entire tasks or workflows, or have them collaborate within a coordinated multi-agent orchestration to handle more complex objectives.

As single-agent solutions, these agents can be ideal for linear processes due to their simplicity in design, testing and deployment. In a multi-agent orchestration, you put multiple specialized agents to work collaboratively to enable modular, scalable, and reusable solutions.

On watsonx Orchestrate, you can combine the following key components to develop your agents:

  1. Knowledge: Embeds domain-specific rules, data, and context to guide the agent’s decisions.
  2. Tools: Provide the agent with functional capabilities like API access, data processing, or AI services.
  3. Collaborator agents: Define the logic and behavior, handling tasks, decisions, and interactions.
  4. Channels: Connect to interfaces through which the agent communicates, such as a chat.

The key components that you can use to develop your agent.
Figure 1. The key components that you can use to develop your agent.

Creating and configuring an agent

Creating and configuring an agent involves defining its profile, knowledge, toolset, and behavior. These areas help ensure that the agent operates effectively and can collaborate with other agents in multi-agent workflows.

The key areas to configure your agent:

  • Profile: Defines the agent's identity and role.
  • Knowledge: Provides the information that is needed for the agent to respond to user queries.
  • Toolset: Integrates external tools and agents to extend functionality.
  • Behavior: Specifies how the agent responds to user requests and workflows.

The role of environments in agent building

The agent builder has two environments for agents: draft and live. An agent in draft is in the process of development and is not available for use in channels like chat. Deploying the agent to live makes it available for use with the current copy of the resources that you use to ground it.

The resources that you add to an agent can be:

  • Knowledge
  • Tools
  • Collaborator agents
  • Channels

Manage agents and their resources from the Manage agents page within the agent builder. The agents cards have the label "Live" for deployed agents, or none label for agents that are in draft.

The draft environment for building

An agent in the draft environment contains all your in-progress work while you build and test the agent with different resources such as collaborator agents, tools, and knowledge. Everything you do in the agent while it's in draft is not exposed to users.

The following table details the actions that you can do within an agent based on the resources that you can add to it while in the draft environment:

Resource Test in the Preview chat Use in the chat homepage Add as a resource to agents
Collaborator agents Green check icon Red cross icon Green check icon
Tools Green check icon Red cross icon Green check icon
Knowledge Green check icon Red cross icon Green check icon
Voice Green check icon Red cross icon Green check icon

The live environment for production

An agent in the live environment contains the resources that the agent must use to interact with users over the channels that you configured while you build the agent.
The version of the resources that the agent uses is defined at the time that you deploy the agent.

The following table details the actions that you can do within an agent based on the resources that you can add to it while in the live environment:

Resource Test in the Preview chat Use in the chat homepage Add as a resource to agents
Collaborator agents Green check icon Green check icon Green check icon
Tools Green check icon Green check icon Green check icon
Knowledge Green check icon Green check icon Green check icon
Voice Red cross icon Red cross icon Red cross icon

Versioning resources

When you deploy an agent, it brings every resource that you use during development. Each of these resources can evolve independently over time, and to track their changes, your agent uses the latest version available for the resource either in the draft or live environment.

A version of a resource is the copy of the resource at a specific point in time. When you deploy an agent, it automatically uses the latest version of each of its resources to help ensure that the deployment reflects the most up-to-date state of all its components.

Using resources within agents

What an agent brings from each resource might vary depending on the resource:

  • Knowledge

    You add knowledge within the agent, and the settings that you do for grounding the agent with knowledge are in the scope of that specific agent only. It is not reused across agents within your instance.
    When you deploy the agent, these settings are not available for reuse across agents within your watsonx Orchestrate instance.

  • Tools

    You create tools within the agent. When you deploy an agent, the tools that you create become available in their current state for use.
    Also, members of your watsonx Orchestrate instance can reuse these tools in their own agents.

    Note: Connection management for tools is done and maintained through the ADK. For more information, see Managing connections in the IBM watsonx Orchestrate Agent Development Kit documentation.

  • Collaborator agents

    Collaborator agents operate by using the copy of their current state at the time that you deploy the main agent.
    Using an agent as a collaborator doesn't depend if it's in the draft or live environment.

  • Channels

    The agent is available in a channel after you set up the channel and deploy the agent. You use the agent and every resource that you use on it through the channel that you set.

  • Voice configuration (Preview)

    Configure a voice for your agent and make possible a verbal interaction with users. This feature is only available for Agent builder preview for moment.