Configuring embedding settings
Knowledge Hub is an Early Access release. Refer to the Early Access disclaimer for details.
Knowledge Hub uses a single, Boomi-managed embedding model, Cohere Embed v4 on Amazon Bedrock, to convert your chunked text into vectors for storage and retrieval. You do not choose between models at Early Access.
About the embedding model
An embedding model transforms text chunks into numerical vector representations. When an AI agent queries a Knowledge Base, Knowledge Hub uses the same model to embed the query and finds the closest matching chunks in the vector store.
Because the query and the indexed content must use the same vector space, changes to the embedding configuration require reingesting all content; existing knowledge is not updated automatically.
For model details, including supported input types and token limits, refer to the Cohere Embed v4 documentation on Amazon Bedrock.
Model access
The Embedding tab's Model Access setting controls who hosts the embedding model:
| Option | Description |
|---|---|
| Managed | Fully managed by the platform. No configuration or maintenance required. Selected by default and the only option available at Early Access. |
| Custom | Connect your own model provider for complete control over configuration. Not yet available (Coming Soon). |
Viewing the embedding model
The embedding model is set automatically when you create a Knowledge Base. To review it on an existing KB:
- Open Knowledge Hub, select the repository, then select the Knowledge Base.
- Select the Embedding tab.
- Under Embedding Configuration, review the Model Provider and Embedding Model fields.
The Embedding Model field is read-only if at least one ingestion run has completed for this Knowledge Base.
