v2.5

A new isolate_vector_search option on the Knowledge class lets multiple agents or teams share the same vector database while keeping their search results isolated. Each agent only retrieves documents relevant to its own knowledge scope, even when the underlying storage is shared.

This simplifies infrastructure by reducing the number of vector databases you need to manage, while preserving the accuracy and separation required for multi-agent or multi-tenant deployments.

Details

• Enable with isolate_vector_search=True on any Knowledge instance
• Shared vector database, isolated query results per agent or team
• Reduces infrastructure cost and operational overhead for multi-agent systems
• No changes required to existing vector database configuration

View the Vector Search docs.