Gmail + GPT-4o-mini Auto Reply Drafts (n8n) w/ Telegram

Draft knowledge-grounded Gmail replies automatically—then notify your team on Telegram (with an escalation option)

This n8n workflow rebuilds your Google Docs knowledge base into an in-memory vector store and, when new unread Gmail messages arrive, uses OpenAI GPT-4o-mini to draft a reply grounded in your documentation—saving it as a Gmail draft and sending a preview via Telegram.

What this workflow does

• Builds a knowledge base: On manual test or a daily 6:00 schedule, it downloads a specified Google Docs file from Google Drive, splits it into chunks, embeds the content, and stores it in an in-memory vector store.
• Monitors Gmail: Every minute, it polls Gmail for new unread customer emails.
• Filters automation noise: It excludes messages where the sender/subject contains “no-reply”, “noreply”, or “unsubscribe”.
• Drafts a reply with knowledge retrieval: Using GPT-4o-mini with a vector-search tool, it looks up relevant passages and generates a plain-text reply—or outputs an [ESCALATE] line when the knowledge base doesn’t cover the question.
• Saves as a Gmail draft: The generated text is added as a draft reply in the original email thread.
• Notifies + logs status: It sends a Telegram message with a preview of the draft and appends a status row (e.g., draft ready vs. needs human) to a Google Sheets support log.

Use cases

• SaaS support teams drafting consistent replies based on internal documentation
• Automation engineers setting up an n8n “email-to-draft” agent with safe escalation ([ESCALATE])
• Founders validating faster response workflows without fully sending automated emails

Technical details

• Integrations/nodes: Gmail, filter, Telegram, Sticky Note, Google Drive, and a Gmail trigger/polling loop
• AI: OpenAI GPT-4o-mini for drafting and OpenAI embeddings for knowledge base indexing
• Logging: Google Sheets “Support Log” sheet