Get Started
Back to Templates

Create a Voice & Text Telegram Assistant with Lookio RAG and GPT-4.1

Created by

Created by: Guillaume Duvernay || duv

Guillaume Duvernay

Last update

Last update 2 days ago

Categories

Share

Create a Telegram bot that answers questions using Retrieval-Augmented Generation (RAG) powered by Lookio and an LLM agent (GPT-4.1).

This template handles both text and voice messages (voice transcribed via a Mistral model by default), routes queries through an agent that can call a Lookio tool to fetch knowledge from your uploaded documents, and returns concise, Telegram-friendly replies. A security switch lets you restrict use to a single Telegram username for private testing, or remove the filter to make the bot public.

Who is this for?

  • Internal teams & knowledge workers: Turn your internal docs into an interactive Telegram assistant for quick knowledge lookups.
  • Support & ops: Provide on-demand answers from your internal knowledge base without exposing full documentation.
  • Developers & automation engineers: Use this as a reference for integrating agents, transcription, and RAG inside n8n.
  • No-code builders: Quickly deploy a chat interface that uses Lookio for accurate, source-backed answers.

What it does / What problem does this solve?

  • Provides accurate, source-backed answers: Routes queries to Lookio so replies are grounded in your documents instead of generic web knowledge.
  • Handles voice & text transparently: Accepts Telegram voice messages, transcribes them (via the Mistral API node by default), and treats transcripts the same as typed text.
  • Simple agent + tool architecture: Uses a LangChain AI Agent with a Query knowledge base tool to separate reasoning from retrieval.
  • Privacy control: Includes a Myself? filter to restrict access to a specific Telegram username for safe testing.

How it works

  1. Trigger: Telegram Trigger receives incoming messages (text or voice).
  2. Route: Message Router detects voice vs text. Voice files are fetched with Get Audio File.
  3. Transcribe: Mistral transcribe receives the audio file and returns a transcript; the transcript or text is normalized into preset_user_message and consolidated in Consolidate user message.
  4. Agent: AI Agent (GPT-4.1-mini configured) runs with a system prompt that instructs it to call the Query knowledge base tool when domain knowledge is required.
  5. Respond: The agent output is sent back to the user via Telegram answer.

How to set up

  1. Create a Lookio assistant: Sign up at https://www.lookio.app/, upload documents, and create an assistant.
  2. Add credentials in n8n: Configure Telegram API, OpenAI (or your LLM provider), and Mistral Cloud credentials in n8n.
  3. Configure Lookio tool: In the Query knowledge base node, replace <your-lookio-api-key> and <your-assistant-id> placeholders with your Lookio API Key and Assistant ID.
  4. Set Telegram privacy (optional): Edit the Myself? If node and replace <Replace with your Telegram username> with your username to restrict access. Remove the node to allow public use.
  5. Adjust transcription (optional): Swap the Mistral transcribe HTTP node for another provider (OpenAI, Whisper, etc.) and update its prompt to include your jargon list.
  6. Connect LLM: In OpenAI Chat Model node, add your OpenAI API key (or configure another LLM node) and ensure the AI Agent node references this model.
  7. Activate workflow: Activate the workflow and test by messaging your bot in Telegram.

Requirements

  • An n8n instance (cloud or self-hosted)
  • A Telegram Bot token added in n8n credentials
  • A Lookio account, API Key, and Assistant ID
  • An LLM provider account (OpenAI or equivalent) for the OpenAI Chat Model node
  • A Mistral API key (or other transcription provider) for voice transcription

How to take it further

  • Add provenance & sources: Parse Lookio responses and include short citations or source links in the agent replies.
  • Rich replies: Use Telegram media (images, files) or inline keyboards to create follow-up actions (open docs, request feedback, escalate to humans).
  • Multi-user access control: Replace the single-username filter with a list or role-based access system (Airtable or Google Sheets lookup) to allow multiple trusted users.
  • Logging & analytics: Save queries and agent responses to Airtable or Google Sheets for monitoring, quality checks, and prompt improvement.