Orchestrator Sub Agent Config

1 min read

Detailed Evaluation Config

When Evaluation Config is active, the following nested settings are available:

| Parameter | Default | Description | | Assessment Type | HALLUCINATION | Defines which assessment type is used for the displayed sub-agent evaluation. | | Summarization Model | Default GPT-4o summarization model | Controls which model summarizes the collected sub-agent assessments. The exact model name shown in the UI can vary by deployment. | | Summarization System Message | Built-in summarization prompt | System prompt used by the summarization model. It instructs the model to summarize the sub-agent assessments precisely, focus on failures first, mention agent names, and avoid markdown formatting. | | Summarization User Message Template | Built-in sub-agent assessment template | User message template used to pass the sub-agent assessments into the summarization model. It renders the collected assessment data for one or more sub-agents. |

These settings matter when you want the orchestrator not only to collect sub-agent outputs, but also to generate a structured assessment summary of how those sub-agent responses performed.

Detailed Referencing Config

When Referencing Config is active, the following nested settings are available:

| Parameter | Default | Description | | Referencing Instructions For System Prompt | Built-in sub-agent referencing instructions | Instructions added to the main agent's system prompt so the orchestrator knows how to reference facts coming from sub-agent responses. | | Referencing Instructions For User Prompt | Built-in short referencing reminder | Instructions added to the main agent's user prompt as a shorter reminder to keep referencing behavior consistent in the final answer. |

These settings are useful when you want strong control over how facts from sub-agents are cited by the orchestrator, especially in spaces that rely heavily on grounded answers and inline references.