Create a Knowledge Base
Build a searchable document collection for a team and choose indexing models
A knowledge base must belong to a team. The embedding model selected at creation determines the vector dimensions; do not switch embedding models after documents are first processed. If you must switch, rebuild the index.
- Navigate to Knowledge Bases, select Create Knowledge Base.
- Fill in the name (
1-100characters) and description (up to500characters). - Select the current team.
- Select an authorized
embeddingmodel; optionally select arerankmodel. - Set the chunk size (default
1000characters) and overlap (default100characters), then save. - Open the knowledge base details and confirm the status is Enabled.

Document processing and retrieval fail when the embedding model does not exist, is not authorized, or its vector dimensions do not match the existing index. First confirm authorization in the models guide.
How is this guide?