Vertex AI Search / Conversational Agent - Website DataStore

Google Cloud Build with AI Custom ML & MLOps
,
1

Hello!

I am trying to experiment on creating a conversational agent or utilize Gemini along with Vertex AI Search to get back information regarding my website.

I have created a datastore with the website (indexing applied - yet most likely not finished even though the status is set as " Initial index completed ").

When I preview the app or when i run API calls, I get empty responses back. Additionally, the conversational agent seems to be incompatible with the indexed website thus I cannot create it.

I do not understand where the problem lies. It should be noted that the website is powered by MediaWiki. Does that play any role in my configuration or the indexing happening behind the scenes?

I would really appreciate the help.

Type Website (Advanced)
Serving state Enabled
Region global

Language N/A
Data store size -
Number of data store documents -
Project documents usage/project quota limit 0 / 1,000,000
Initial Index Enabled
Automatic refresh Enabled
Sitemap mode Discovery
Exclude from gen AI features False

2

Hi @danae,

Welcome to the Google Cloud Community!

It looks like you are encountering an issue where no documents have been successfully indexed into your Vertex AI Search data store from your MediaWiki website.

Here are the potential ways that might help with your use case:

  • Look for the number of data store documents: If your Vertex AI Search data store shows a size of “-” or “0” and has no documents after a reasonable period (a few hours, or up to 24 hours for larger sites), indexing has failed to process the content. Checking these fields helps determine whether the ingestion was successful.
  • Incorrect Domain/URLs: Verify the domain and any specified URL patterns for inclusion or exclusion in your data store settings.
  • Permissions: Ensure that your service account or user account executing your API calls has the necessary IAM permissions, such as Vertex AI Search Service User or more specific roles like dialogflow.sessions.detectIntent.
  • Verify Public Accessibility: Open an incognito browser window and try to access the URLs you expect to be indexed. Ensure that you don’t need to log in with your credentials.

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.