in my case im using a Cloud Run server in GCP that gets the messages as an external API (depending on the workflow and if its needed to do some semantic search) and the custom code already has the knowledge base vectorized so i just do a semantic search and return the result.