Supabase vector store problem - or is it?

Hi All,
I’m using the hosted n8n and I’ve linked it to supabase.
I have a google gemini embedding and a default data loader (type of data binary), mode(load all input data), data format(text)
I have a recursive character text splitter connected.

I’ve run the required sql, so my tables are fine in supabase.

I start a chat that includes a file (feedinging into the supabase vector store) but all I get is this error message:
Error inserting: vector must have at least 1 dimension 400 Bad Request

when I go to the supabase vector store I see this input:

{
  "data0": {
    "data": "VGhp......"
"mimeType": "text/plain",
    "fileName": "filename.txt",
    "fileSize": "12335 bytes",
    "fileExtension": "txt",
    "fileType": "text"
  }
}

the output is then formatted as json (it looks correct)

The one thing that does look strange is the embeddings - the input is correct but the output is :

{
  "response": [
    [],
    [],
    [],
    [],
    [],
    [],
    [],
    [],
    [],
    [],
    [],
    [],
    []
  ]
}


any ideas what would cause this?
I'm in europe so I cant use the chat gpt embedding

thanks,
Paul.

The most likely cause of the issue is that the embedding (vector) output is empty — that is, it returns something like [[], [], [], ...]. As a result, Supabase is not receiving a valid embedding to insert into the vector table, which leads to the following error. Supabase Vector Store node documentation | n8n Docs

Hi,
that is what I suspected - but given that this is a no code solution, how can I debug/correct this?

Another possible issue could be the model’s production limitations. The Gemini model you’re using may have certain restrictions—especially in the free tier, where the number of requests per minute is limited. If these limits are exceeded, the model may silently fail and return empty embedding data. If I’ve answered your question and provided a solution, I’d appreciate it if you could mark it as resolved. Thank you, and best of luck!

1 Like