Guides Knowledge AI
Domains / Creating domains

Creating domains

A domain is the top-level container for your content, chatbot, and analytics. Each domain has its own model configuration, indexed documents, and chat history.

From the Domains sidebar, click + Add.

Choose a domain type — Standard or Orchestrator — before filling in the name.

Standard is the default and is what you want for any content-holding domain that indexes its own documents. An Orchestrator instead fans questions out to other standard domains; it is covered in the Orchestrator domain topic.

Enter a Domain name (required) and an optional Description (up to 500 characters).

Decide whether to use the platform's default model configuration or bring your own — see Use default configuration below.

If using custom configuration, click Validate to verify your credentials and model connectivity, then click Create. If using defaults, click Create directly.

The domain appears in the sidebar and is ready for content upload.

When creating a domain the Use default configuration checkbox is enabled by default. With this option checked, the domain inherits the organization's platform-managed LLM and embedding models — no credentials or model details are required. Simply provide a name and click Create.

Uncheck Use default configuration to supply your own model credentials. Three configuration cards appear: LLM configuration, Embedding configuration, and an optional Evaluation LLM configuration. Each card lets you choose a provider and fill in the required fields for that provider.

Guides Knowledge AI's BYOM (Bring Your Own Model) feature lets you connect your own LLM and embedding endpoints instead of relying on platform defaults. This is useful when you need to use privately hosted models, fine-tuned deployments, or a provider not covered by the default configuration.

To use BYOM, uncheck Use default configuration during domain creation (or navigate to Settings → Model configuration for an existing domain). You will see three configuration sections described below.

Supported providers

Guides Knowledge AI currently supports the following LLM and embedding providers. The available models and fields are loaded dynamically from the platform, so new providers or models may appear without a UI update.

  • Azure OpenAI — requires an API key, endpoint URL, API version, and a model/deployment selection.
  • WatsonX — requires an API key, project ID, and model selection. Optionally accepts a custom URL (defaults to us-south.ml.cloud.ibm.com) and a space ID.
  • Gemini — requires an API key. Model selection is optional (defaults to gemini-1.5-flash for LLM and text-embedding-004 for embeddings).

LLM configuration

Select a Provider from the dropdown. The form dynamically updates to show the fields required by that provider. Common fields include:

  • API key (required) — your provider's secret key. The value is masked after entry.
  • Model (required) — choose from the provider's supported model list, or type a model name if the provider allows free-form input.
  • Deployment ID (optional) — for Azure OpenAI deployments that use a deployment name distinct from the model name.
  • Endpoint (required for Azure OpenAI, WatsonX) — the base URL of your model deployment.
  • API version (required for Azure OpenAI) — the API version string, e.g. 2024-02-01.

Behavioral parameters — these optional settings fine-tune how the LLM generates responses:

  • Temperature — controls randomness (default 0.7). Lower values produce more deterministic output; higher values increase creativity.
  • Max tokens — the maximum number of tokens in the generated response.
  • Top P — nucleus sampling threshold. The model considers tokens whose cumulative probability reaches this value.
  • Seed — when supported by the provider, fixes the random seed for reproducible outputs.

Embedding configuration

The embedding model converts your documents and queries into vector representations for retrieval. Select a Provider and fill in the required credential and model fields — the same providers listed above are available for embeddings. Typical required fields are API key, Model, Endpoint, and API version (provider-dependent).

Evaluation LLM configuration

This section is disabled by default and can be enabled with the toggle switch. When enabled, it configures a separate LLM used for answer quality evaluation (LLM-as-a-Judge) and evaluation experiments. The fields are identical to the main LLM configuration — pick a provider, supply credentials, and select a model. Using a dedicated evaluation LLM keeps judging costs and rate limits separate from your production answer-generation model.

Validation flow

After filling in all required fields, click Validate. Guides Knowledge AI sends a test request to each configured model (LLM, embedding, and evaluation LLM if enabled) and reports per-model results:

  • Valid — the model responded successfully. Response latency is displayed.
  • Invalid — the connection failed. An error message describes the issue (e.g. invalid API key, unreachable endpoint).

The Create button is only enabled after all models pass validation. You can re-validate at any time after changing fields.

FAQ

What is a domain?
A domain is the top-level container for your content, chatbot, and analytics. Each domain has its own model configuration, indexed documents, and chat history.
How do I create a new domain?
From the Domains sidebar, click + Add, choose a domain type (Standard or Orchestrator), and enter a required domain name plus an optional description (up to 500 characters). Decide whether to use the default model configuration or provide custom credentials, then click Create (or Validate first if using custom configuration). After creation, the domain appears in the sidebar and is ready for content upload.
What’s the difference between a Standard domain and an Orchestrator domain?
Standard is the default and is intended for content-holding domains that index their own documents. An Orchestrator domain instead fans questions out to other standard domains.
What does “Use default configuration” do when creating a domain?
When enabled (it’s checked by default), the domain inherits the organization’s platform-managed LLM and embedding models, so no credentials or model details are required. You only need to provide a name and click Create. If you uncheck it, you must supply your own model credentials and configure LLM, embedding, and optionally an evaluation LLM.
How does BYOM (Bring Your Own Model) work for domain model configuration, and what providers are supported?
BYOM lets you connect your own LLM and embedding endpoints instead of using platform defaults, which is useful for privately hosted models, fine-tuned deployments, or providers not covered by defaults. To use it, uncheck Use default configuration during domain creation (or go to Settings → Model configuration for an existing domain) and fill in the LLM, embedding, and optional evaluation LLM sections. Supported providers include Azure OpenAI, WatsonX, and Gemini, and you can click Validate to test connectivity; Create is enabled only after all configured models pass validation.