VertexAI Conversational Agent Slack integration and file creation problems

[Teppo]

I have created a Conversational agent connected to a BigQuery database with cloud run functions using VertexAI builder and connected it to Slack with a ready-made integration.

It seems that not all of the functional queries given in the preview window work in Slack, for example, if I ask for the total number of rows in a database table, it works, but if I ask to return the first 10 rows, it doesn't work, but says "I'm sorry, I can't fulfill this request. The tool I use to access data is not working properly"?

Another problem is related to file creation. The code_interpreter tool connected to the agent does generate a .csv file, for example, but it only does so according to the agent's cache, and the file cannot be loaded from anywhere. How could this problem be solved and could the agent be asked to create a file from the records to be returned?

[MJane]

Hi @Teppo,

Welcome to the Google Cloud Community!

I understand that you're facing a couple of issues with your conversational agent integrated with BigQuery, Cloud Run functions, and Slack. Let's address the challenges you're encountering and discuss some potential solutions : 

1. Issue with Query Execution (First 10 Rows Not Working) - It seems like the agent works fine for returning row counts but encounters an issue when fetching the first 10 rows. The issue could be with how the query is written, permissions, or timeouts. Try testing and debugging the query logic for fetching the first 10 rows specifically. Also, check the BigQuery connection settings, timeouts, and read permissions for the database.
2.  File Creation & Access Problem - You mentioned that while the agent is generating .csv files, they are stored in the cache and cannot be loaded or accessed. This is likely because the file is being created in temporary storage within the agent's environment.You may try to set up your agent to upload the .csv file to a Google Cloud Storage bucket instead. This way, the file will be stored permanently and can be easily accessed or shared with a signed URL.

Was this helpful? If so, please accept this answer as “Solution”. If you need additional assistance, reply here within 2 business days and I’ll be happy to help.

[Teppo]

1. I explained it unclearly, sorry. BigQuery does return the data just fine and I can see it in the agent preview. This error "I'm sorry, I can't fulfill this request..." comes up when I try to perform the same task via Slack. It seems like something mysterious is happening with the Slack integration.

2. Creating Cloud Storage Bucket would be a possible solution. I will look at it. I assume I would need to make a separate tool for that and call it in instructions.