Internal Error and Indexing Error When Uploading JSON Files (~5MB) to Azure AI Knowledge

[Youssef-El-Garmit]

Hi everyone,

I’m facing an issue with Azure AI Foundry when trying to upload local files to Azure Knowledge.

Everything works fine for small files, but when I try to upload a .json file around 5MB, the upload seems to complete successfully — however, I immediately get an “Internal Error” and an “Indexing Error” without any detailed message.

I also tried splitting the JSON into multiple smaller files. Strangely, some of them upload and index successfully, while others randomly fail with the same indexing error, even though they’re similar in structure and all go to the same vector base.

I’m not using an Azure AI hub, just a simple Azure AI Foundry project, and I’m located in the Switzerland North region.
The model I’m using is GPT-4.

What’s confusing is that the official upload limit is 200MB, so a 5MB file should be fine. It really looks like something inconsistent is happening during the indexing phase.

For context: I have an agent that I send documents to — it’s supposed to find clients linked to these documents from a knowledge base that I synchronize daily with Azure AI SDK.

Has anyone experienced similar behavior or found a workaround for this kind of issue?

Thanks in advance!

[RichardReukemaOptimus]

You state the use of a json payload, but where is the text you are indexing?

In my scenario, I put files into Azure blob storage as my data source (docx, text, etc) and then update the index.

Im not familiar how to load text into AI Search thats a json payload.. more learning as always

You may want to also look at content understanding for a wider array of file types

Hth
R

[Youssef-El-Garmit]

@RichardReukemaOptimus

Thanks for your reply!

In my case, I’m using the recommended Azure AI Foundry project (not the Hub project). When I upload files through the interface, they are automatically stored and indexed by Azure Foundry — I don’t handle the blob storage or index directly.

It’s quite abstract and simplified compared to setting everything up manually with Azure AI Search or Blob Storage. That’s why I was surprised to see these internal and indexing errors happening, especially since small files work fine while slightly larger ones (like 5MB JSON) fail inconsistently.

[RichardReukemaOptimus]

With the amount of development occurring in this area, im surprised its working at all!