How Agent Hub Processes Requests
Agent Hub uses an orchestrator to evaluate your request and determine how to handle it. The orchestrator helps connect your request with an appropriate AI capability without requiring you to know which agent or capability to use.
Note:
Agent Hub orchestration uses AI models, and the tokens it consumes are billed in AIDP Units.When you enter a prompt without first selecting an agent, the orchestrator evaluates your request and the AI resources available to you. It continues to evaluate each new request as the conversation progresses. Depending on the request, Agent Hub can handle it in the following ways:
- Route the request to an agent: The orchestrator identifies one agent whose capabilities match your request and routes the request to that agent.
- Present agents for selection: If Agent Hub recommends multiple agents, you can use the suggested agent or choose another recommended agent.
- Invoke a tool: If the orchestrator identifies a tool that can handle your request, Agent Hub invokes the tool to perform the appropriate action.
- Provide a direct response: If your request doesn't require an additional AI capability, the orchestrator can provide a response directly using the configured large language model (LLM).
When you start a chat without selecting or tagging a particular agent, different agents can handle successive requests within the same conversation. For example, one agent might investigate a business issue, while another agent handles a follow-up request about how to address it.
When routing a request to a different agent, the orchestrator can provide relevant information from earlier in the conversation. The information available to the agent depends on what the orchestrator provides and how the agent processes it.
Depending on your approval setting, Agent Hub might ask you to select an agent before continuing or automatically use the agent it identifies as the best match.
To see which agent or tool Agent Hub invoked while processing your request, click the processing status indicator (for example, "Completed in 18s") in the chat to open the Thinking Activity panel.
To identify relevant agents, the orchestrator uses descriptive information from registered Agent Cards, including an agent's name, description, and skills. Clear and complete Agent Card information helps Agent Hub match requests with appropriate agents.
If you select or tag an agent when starting a chat, that agent remains associated with the conversation. You can't replace the selected agent within the same chat, although relevant MCP tools can still be invoked.
If you start the chat from a topic, the topic narrows the resources that Agent Hub considers when determining how to handle your request. See Understand How Topics Affect a Chat.
Why this matters
You don't need to know which agent is best suited to every task before you start a conversation. You can describe what you want to accomplish and let Agent Hub help connect your request with an appropriate AI capability.