@piotrsofts ,
Short answer: yes, you can build and use a Databricks Knowledge Assistant today. Exposing it directly inside the Databricks One chat UI is currently limited to specific agent types.
What the Knowledge Assistant is
The Knowledge Assistant is an Agent Bricks pattern for building a domain-specific RAG assistant over your own documents. It answers questions with citations and gets better over time through expert feedback (ALHF). Think โyour orgโs knowledge, grounded, auditable, and improvable.โ
Can you use it from Databricks One?
Right now, the Databricks One chat experience supports Genie agents. There isnโt a first-class, generic Knowledge Assistant UI in One yet. That said, this is a known gap and is being considered as a future enhancement.
What you can do today
You can build a Knowledge Assistant using Agent Bricks (Beta) and test it end-to-end in AI Playground. From there, you can generate production-ready code for API access (curl or Python) or apply it directly to data.
For end-user access, the recommended path is Databricks Apps. Thereโs a Marketplace app template specifically designed to front a Knowledge Assistant endpoint, which gives you a clean, turnkey UI with minimal wiring.
If youโre aiming for a more unified experience across structured and unstructured data, you can also use a Multi-Agent Supervisor. This lets you coordinate a Genie Space (for structured data) alongside your Knowledge Assistant (for unstructured content) behind a single conversational entry point.
Whatโs required to build one
Youโll need Mosaic AI Agent Bricks enabled (Preview/Beta), Unity Catalog, serverless compute, and access to foundation models under system.ai. On AWS, supported regions today are us-east-1 and us-west-2.
Your knowledge sources can live in Unity Catalog volumes (txt, pdf, md, ppt/pptx, docx) or in a Vector Search index built with the databricks-gte-large-en embedding model.
Quick setup flow
Go to Agents โ Knowledge Assistant โ Build. Configure your knowledge sources, add optional system instructions, and create the agent.
Test everything in AI Playground โ review answers, citations, and traces โ and use expert feedback to improve quality via ALHF.
When youโre ready to share, grant end users CAN QUERY on the agent endpoint and surface it through a Databricks App or your own custom UI.
Hope this helps, Louis.
