Agent hierarchy

You can create complex multi-layered orchestration topologies by specifying sub-agents for sub-agents.

For example, you can specify that main agent A has sub-agents B and C. And then, you can specify for agent C, that it has it’s own sub-agents D and E. When conversation with agent A starts, all above-mentioned sub-agents are initialized and are part of the conversation.

Agents are allowed to communicate (via pass_question or send_message tools) with:

• their parent - if such exists

• their siblings - i.e. sub-agents that share the same parent

• their sub-agents - i.e. any agent explicitly configured as sub-agent

In the above example:

• A may communicate with B and C

• B may communicate with A and C

• C may communicate with A, B, D and E

• D may communicate with C and E

• E may communicate with C and D

If, for example, you want to allow D communicate with B, you may add B as D’s sub-agent.