Using agents in Orchestrate Chat
In IBM watsonx Orchestrate, you can use AI agents within the chat interface to automate tasks, manage workflows, and coordinate actions across multiple agents. Agents either respond to you directly or work in the background, sharing data and coordinating actions.
Each agent is identified by a name and a color-coded avatar, based on a predefined sequence (for example, blue, gray, green) to help users recognize which agent is active. To manage chat threads, use the overflow menu (⋮) to rename
or delete them. For multi-line input, press "Alt + Enter" or "Shift + Enter" (for Windows) or "Shift + return" (for Mac). This shortcut lets you format your message across
multiple lines before you send it.
Learn how to use watsonx Orchestrate Chat to quickly explore its features.
Note: This video has no narration.
Transcript
The video shows the user selecting the Employee Support agent and starting a conversation in the Orchestrate Chat, by asking to update their personal address. Then, the user changes to another agent that is named Compensation, by asking for their compensation benefits.Prebuilt agent behavior
Prebuilt agents provide essential AI capabilities using large language models (LLMs). You can rely on them for tasks such as:
- Question answering
- Summarization
- Content generation
- Text classification
- Translation
- Sentiment analysis
What to expect:
- A prebuilt agent appears only when no other agents are active in your tenant.
- It is hidden from the agent selection menu and becomes inaccessible once a custom agent is deployed.
- The system hides recent chat threads until a custom agent is added.
- Chat history is not saved, refreshing the screen clears the conversation.
- Conversational search is not supported.
Connecting an agent to a channel
To make an agent available for real-time interaction, you must connect it to a channel. This makes the agent appear in the chat interface and perform tasks based on its configured workflows.
Once connected, the agent shows up in the Agents list within the chat view. If you don’t see a specific agent, you must request the builder to enable it.
For more information, see Connecting to channels.
Connecting to a live agent
In some situations, a conversation needs to be handed off to a live agent for personal assistance. Before you can enable live agent transfers, you must integrate a supported custom service desk system.
The type of service desk you use depends on how users connect with your live agents. When integrated with a supported service desk, you can escalate conversations to a live agent as needed. This process is known as escalation and ensures users receive human assistance when required. For more information, see Connecting to a live agent.
Agent collaboration in orchestration
Agents can collaborate behind the scenes even if they’re not directly connected to the chat interface. They support orchestration by:
- Supporting tasks initiated by other agents
- Sharing relevant data
- Coordinating actions across workflows
This background collaboration enables a multi-agent system where specialized agents combine their strengths to deliver smooth, intelligent automation.
Showing or hiding reasoning traces in Orchestrate chat
Reasoning traces show how an agent generates a response during a conversation. Use this visibility to debug or validate the agent’s logic. Users can view reasoning traces to understand how an agent forms a response, while some domains might choose to hide them to keep the chat interface simple.
Set the Hide reasoning property in the agent configuration to control whether users can see reasoning traces in the Orchestrate chat. This property affects only the user experience. The Agent Builder preview always shows reasoning traces, regardless of the property value. For more information, see Authoring agents.
When you add a collaborator agent with Hide reasoning set to true under a supervisor agent with it set to false, the supervisor agent’s configuration overrides the collaborator’s setting.