Unable to create a Knowledge Assistant in an enterprise tier workspace

[jnelson]

I am getting this error when I try to create a Knowledge Assistant agent in a workspace that is on an Enterprise tier

Errors observed:
1. INVALID_PARAMETER_VALUE:
Vector index type DIRECT_ACCESS is not supported for this workspace.
Request ID: bcabe324-fa90-4704-8a43-ebf85841e48d
Trace ID: 16985c5099a1239c425c972aa48f3754

2. RESOURCE_EXHAUSTED:
Maximum number of AI Search endpoints per workspace exceeded quota of 1.

[Ashwin_DSA]

Hi @jnelson,

Being on the Enterprise tier does not by itself guarantee this will work if the workspace still has an AI Search endpoint quota of 1 or does not support the Direct Access index type for this path.

Those two errors usually point to AI Search/Vector Search limits in that workspace, not just the workspace tier itself.

For the Knowledge Assistant, there are two things to check:

• Vector index type DIRECT_ACCESS is not supported for this workspace, which typically means the workspace does not support using a Direct Access vector index for this flow. If you're selecting an existing AI Search index, make sure it's configured as a supported index rather than Direct Access. The public docs for Knowledge Assistant are here: Use Knowledge Assistant to create a high-quality chatbot from your documents, and the public community guidance specifically states that Knowledge Assistant does not support self-managed or Direct Access indexes for this scenario.
• Maximum number of AI Search endpoints per workspace exceeded quota of 1 means the workspace has already hit its AI Search endpoint quota. In another community thread, it is mentioned that Agent Bricks currently creates a new vector search endpoint as part of this flow rather than reusing an existing one, which is why this quota can block creation even if you already have AI Search set up.

A few practical things to try:

• If you're using an existing AI Search index, recreate or switch it to a supported setup and avoid Direct Access for this Knowledge Assistant workflow.
• Check whether there is already an AI Search endpoint in the workspace consuming the only allowed slot.
• Verify the rest of the documented prerequisites as well... serverless compute, Unity Catalog, Model Serving, a nonzero serverless budget policy, a supported region, and supported knowledge source types / embedding models.

Hope this helps.

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.

[jnelson]

When you create a Knowledge Assistant type Agent, you don't to specify the type of index. The link you referred to (https://docs.databricks.com/aws/en/generative-ai/agent-bricks/knowledge-assistant) doesn't make any mention of any sort of restrictions. In my case, the files are in a volume and I have a non-zero budget specified.  There is no way to tell the Agent wizard to use the existing endpoint (to avoid quota limit issues)

[Ashwin_DSA]

Hi @jnelson,

Apologies. Having checked this internally, I don’t think this is something you can fix in the wizard itself.

If you create a Knowledge Assistant from Files in a Volume, the public docs only tell you to choose the volume path and use supported file types. They do not expose any setting to choose an index type, and they do not document a way to force the wizard to reuse an existing AI Search/Vector Search endpoint.

That’s why the two errors you’re seeing are confusing from a user perspective. On the surface, you are not being asked to configure any vector index details. But the public community thread indicates that Knowledge Assistant may still create search infrastructure under the hood, including a new Vector Search endpoint rather than reusing an existing one.

If you have already checked that the prerequisites (serverless compute, Unity Catalog, Model Serving, non-zero serverless budget, supported region, and supported file types in the volume) are all satisfied, you can check whether the workspace already has an AI Search / Vector Search endpoint using up the only allowed slot. If there is an unused endpoint, delete it and retry. If it still fails, this looks more like a workspace/product limitation than a wizard misconfiguration, because the volume-backed flow does not give the user control over the underlying index type or endpoint reuse. You may also want to raise a support ticket so the team can investigate it.