I have a working system: Telegram bot saves customer data to Google Sheets (email, payment type, zelle) and context/rules to Qdrant. Email Assistant receives emails, finds customer in Sheets, searches Qdrant for relevant knowledge, and creates Gmail draft replies.
I have fixed scripts for different situations (new order prepay/postpay, tracking questions, payment questions).
Should I: store scripts in a separate Google Sheets table and substitute without LLM generation, or is there a better approach?
Can share n8n workflow JSON if helpful.
Describe the problem/error/question
What is the error message (if any)?
Please share your workflow
Share the output returned by the last node
Information on your n8n setup
- n8n version:
- Database (default: SQLite):
- n8n EXECUTIONS_PROCESS setting (default: own, main):
- Running n8n via (Docker, npm, n8n cloud, desktop app):
- Operating system:
Hi @Aleksei_Chprynin
Hope you’re doing well.
The “most robust approach for production” is almost always hybrid:
Scripts/templates as the source of truth for content, and the LLM only for rewriting in the right tone, adapting to the email context, carefully filling in gaps, avoiding robotic language, and stitching together the final response.
Why? Because in e-commerce, consistency and compliance (payment policy, timelines, tracking, returns) matter way more than “creativity.”
1 Like
Hi @Aleksei_Chprynin if your scripts are going to be static and you would not be changing them using a vector store would be faster and reliable and you can also feed your context if you have any specific set of rules you want the AI agent to handle the request according to, and if you are looking to test vector DB just start with Pinecone node that is free and would get you a lot of features, hope this helps.
1 Like