Agent Query returns Internal Server Error everytime

Support
,
1

Description

My endpoint is srywmqkltlc61tqxcu1dqw.c0.us-east1.gcp.weaviate.cloud

Server Setup Information

  • Weaviate Server Version: 1.33.1
  • Deployment Method: None
  • Multi Node? Number of Running Nodes: No
  • Client Language and Version: ?
  • Multitenancy?: No

Any additional Information

I downloaded the 1k example csv dataset. I load it into my collection to play. I have an API key set up. I have admin role.

I go to the new Query Agent ( Weaviate Cloud ) and select my website collection and I ask a question and I get an Internal Server Error – every time….

image
image903Ă—931 53.2 KB

What am I doing wrong?

2

Hey there, welcome to the Weaviate Community!

It appears there something isn’t quite right when the Query Agent is configured with a collection using binary quantization - I was able to replicate your error and will get this back to the team for investigation.

Meanwhile, you can get around this error by using the other compression methods available: uncompressed / RQ1 / RQ8 are all working correctly with the Query Agent.

3

@linyaru Got it – thanks. What is the recommended compression when dealing with 100s of legal docs?

4

@djasnowski our current recommended compression is 8-bit rotation quantization (RQ-8), which provides a great balance between recall and performance - you can find more information here

5

Got it. Thanks. Question… if I have a real estate legal documents… what would be the best method… only vectorize the extracted values I KNOW people will search for? And secondly . If I don’t vectorize for those values.. can users still search and query for those in the vector db?

6

and one more question @linyaru Using the RESTful API endpoints, is it possible to insert data into my collection? and what API endpoint is the query agent? (or to query)

7

if I have a real estate legal documents… what would be the best method… only vectorize the extracted values I KNOW people will search for? And secondly . If I don’t vectorize for those values.. can users still search and query for those in the vector db?

Hi @djasnowski, you’ll generally get the best results by vectorizing all of the unstructured text, not just the extracted fields you expect people to search for. That allows the Query Agent (and vector search in general) to understand meaning, not just keywords, so it can retrieve relevant documents even if search phrases don’t appear verbatim. That being said, non-vectorized text is still searchable through BM25 keyword search.

Additionally, if there are shared structured attributes in your legal documents (ex. date, jurisdiction, address), these are best stored as metadata filters rather than embeddings.

I also recommend reading up our guidance on chunking strategies here.

Using the RESTful API endpoints, is it possible to insert data into my collection? and what API endpoint is the query agent? (or to query)

It certainly is. You can find manuals on CRUD with the core database here, and Query Agent usage docs here. We also have an Ask AI feature there that can answer more specific product questions based on docs.

1 Like