Guides Knowledge AI Documentation
Domains / Orchestrator domain

Orchestrator domain

An orchestrator is a special domain type that does not index any content of its own. Instead, it fans an incoming question out to multiple standard source domains , then merges and ranks their answers into a single response. Use an orchestrator when knowledge is split across several product or team domains and you want one chatbot that can answer across all of them.

Because orchestrators hold no content, they require no embedding configuration — only an LLM configuration for merging and ranking source answers. Retrieval, attribution, and content quality are all owned by the underlying source domains.

  1. From the Domains sidebar, click + Add.
  2. Choose the domain type Orchestrator.
  3. Enter a name and an optional description.
  4. Choose the default LLM configuration or supply a custom one. If you provide a custom configuration, click Validate to verify connectivity before continuing. Note that no embedding configuration is requested for orchestrators.
  5. Click Create. A banner appears directing you to configure source domains from the Sources page — the orchestrator cannot answer anything until at least one source is linked.

Open Settings → Sources on the orchestrator. You will see a checklist of every standard domain in your organization (other orchestrators are excluded). Each row shows the source domain's agent card description and its topic chips so you can judge what that domain covers at a glance.

  1. Select the source domains the orchestrator should route questions to.
  2. Click Save.

If a previously linked source is no longer eligible (for example it was deleted or converted), a warning is shown so you can update the selection.

Tip
The quality of orchestrator routing depends directly on each source domain's agent card. Make sure every source you link has a complete agent card — see the Agent card topic.

Analytics for an orchestrator are limited to App usage, Feedbacks, and Unanswered questions. The Attributions, Compare feedback, Human feedback, and LLM as a judge tabs are hidden, because document-level attribution and quality are owned by the source domains, not the orchestrator. To review those metrics, open the analytics of the individual source domain.

FAQ

What is an orchestrator domain and when should I use one?
An orchestrator is a special domain type that does not index any content itself. It fans an incoming question out to multiple standard source domains, then merges and ranks their answers into a single response. Use an orchestrator when knowledge is split across several product or team domains and you want one chatbot that can answer across all of them.
How do I create an orchestrator domain?
From the Domains sidebar, click + Add and choose the Orchestrator domain type. Enter a name and optional description, then choose the default LLM configuration or provide a custom one and click Validate to verify connectivity. Click Create; you’ll be directed to configure source domains, and the orchestrator cannot answer anything until at least one source is linked.
How do I configure source domains for an orchestrator?
Open Settings → Sources on the orchestrator to see a checklist of every standard domain in your organization (other orchestrators are excluded). Select the source domains the orchestrator should route questions to, then click Save. If a previously linked source is no longer eligible (for example, deleted or converted), a warning is shown so you can update the selection.
Why doesn’t an orchestrator ask for embedding configuration?
Orchestrators hold no content of their own, so they require no embedding configuration. They only need an LLM configuration for merging and ranking answers from source domains. Retrieval, attribution, and content quality are owned by the underlying source domains.
What analytics are available for an orchestrator domain?
Orchestrator analytics are limited to App usage, Feedbacks, and Unanswered questions. Tabs like Attributions, Compare feedback, Human feedback, and LLM as a judge are hidden because document-level attribution and quality are owned by the source domains. To review those metrics, open the analytics of the individual source domain.