Azure AI Search SKU's & Vector Index

[jamlab]

Hello community,
i have an Azure AI Project and trying to index data within the "Data + index" Section in AI Foundry.
I go to the "Indexes" Section, add a new index, choose my blob store (which contains pdfs), AI Search Service, Choose "Add vector search to this search resource" and choose a embedding model (large).

My Problem:
If i use a search service with Basic SKU no vector index is created even the documentation says there's a vector quota of 5 GB.
Only if i use a search service with S1 SKU or above the vector index will be created.

I verified the behaviour in the playground where i can use the data indexes as data source.

Only with the S1 created Index i can choose Hybrid (vector & keyword) as search type.
The Seach Index created with the Basic SKU gives me only the option of keyword and semantic as search type.

Is this the expected behaviour or do i miss something?
Could you provide me a link to an explanation where the difference between basic and S1 regarding vector search is documented?
I can only find different quota sizes, but not functionality.

[amynic]

Hi @jamlab - thanks for detailing your findings above - I am wondering if this documentation page supports your needs - understanding better what pricing tier offers but not just from a size of storage but also from a feature perspective: https://learn.microsoft.com/en-us/azure/search/search-sku-tier#feature-availability-by-tier

[farzad528]

Hi @jamlab — thanks for raising this. The Data + Indexes option you're referring to is outdated and no longer recommended for index creation. Apologies for the confusion around the experience.

Assuming your goal is to create an Agent grounded with knowledge stored in Azure Blob Storage, here’s the recommended approach:

1. Navigate to https://ai.azure.com

2. Click + Create New

3. Select Azure AI Foundry Resource

4. In the left navigation pane, select Agents

5. Click + New Agent

6. During setup, choose your deployed LLM

7. Click + Add Knowledge

8. Select the radio button "Indexes that are not part of this project", and choose your Azure AI Search resource.
   – If your resource isn’t connected yet, click "Connect other Azure AI Search resource" and link your Basic SKU.

9. Click "Create a new index"

   A new browser tab will open with a 3-step index creation wizard:

   • Step 1: Name your index (or use the default)
   • Step 2: Connect to your Blob Storage container
   • Step 3: Link your model deployment for vectorization
     Note: You must deploy an embedding model in your Foundry project (e.g. text-embedding-3-small) beforehand.

   ✔️ Don’t forget to check the box acknowledging the cost of vectorizing content.

10. Return to your Foundry project and select the index from the dropdown
    – Tip: If it doesn’t appear, click the refresh icon

11. Give your index a Display Name

12. Choose a Search Type:

    • simple
    • semantic
    • vector
    • hybrid (vector + keyword)
    • hybrid + semantic

[jamlab]

Hi @farzad528,
sorry for being late on this issue.
I followed your post and it worked in one Azure Subscription (Dev).
I tried this also on another Azure Subscription (Production) and at Step 9 i've got this error.

"The renderComponentIntoRoot component encountered an error while loading."

No matter what i try i (changed browsers, used different search service, ...) i always get this error in this subscription (production environment). The same steps in our Dev Subscription works.

Do you have any idea?

Just for reference:
Meanwhile i tried to create a vector index with the basic SKU again (as described in post 1) and it seems to work now.

[jamlab]

There's also some sort of inconsistency:
If i create a index with the method you described a index with vector data is maybe created (according to the size display of the index).
Unfortunatly i can't use the vector index - at least in the "normal" Playground (without Agent). I can only select keyword or semantic, no hybrid (vector & keyword)

If i use an agent and configure the datasource i can reuse my created index and can select "Hybrid (vector + keyword).
Unfortunatly it's not saved. If i open the dialog again the fields are empty again. Maybe only some ui issue.

If i try the "Normal" Playground and the "Agent" Playground i have different results now although they are shareing the same system prompt and same index.

I'm a bit clueless, i have to say.