n8n Telegram AI Support Agent with pgvector & OpenRouter

Turn Your Telegram Bot Into an AI Support Agent Powered by pgvector + OpenRouter

This n8n workflow automatically handles incoming customer messages in Telegram, retrieves answers from your Postgres pgvector knowledge base, and uses an OpenRouter chat model to respond—while cleanly escalating to a human support chat when confidence is low or retrieval fails.

What this workflow does

• Listens for new Telegram messages sent to your bot.
• Searches your Postgres pgvector knowledge base using OpenAI embeddings and selects the best match using a minimum similarity threshold.
• If retrieval is confident, calls an OpenRouter chat model using Postgres chat memory plus a pgvector search tool to answer strictly from your documents.
• Verifies response quality: if the agent reply is not empty and not “HANDOFF,” it sends the answer back to the customer in Telegram.
• Hands off to humans when needed: if retrieval or the agent is not confident, the workflow informs the customer and forwards the question to your Telegram support chat with a “Respond” button.
• Captures human resolution: if the support team replies within 24 hours, forwards the human answer to the customer and saves the Q&A as a new embedded FAQ entry in pgvector.
• Logs results: stores answered/unanswered Q&A in Postgres and records workflow errors in a Postgres error table with an alert to the support chat.

Use cases

• Customer support for SaaS FAQs where answers must come from your knowledge base.
• Reducing support load by automating first-response handling in Telegram.
• Continuous improvement: turning resolved human answers into new pgvector entries.

Technical details

• Integrations/Nodes: Telegram, Postgres (with pgvector), OpenRouter (via chat model), plus n8n nodes such as if, set, code, and sticky note.
• Core features: pgvector retrieval with similarity threshold, OpenAI embeddings, Postgres chat memory, confidence-based “HANDOFF,” Q&A logging, and error recording to a Postgres error table.

Setup: create and connect a Telegram bot, add it to your support group, and configure supportChatId. Ensure Postgres is set up with pgvector and the required tables/structures.