n8n Multimodal Telegram Bot: Voice, Image & Video Analysis

Input: Telegram message (voice/picture/video/text) → Process: n8n workflow routes by type, extracts media, transcribes or analyzes, merges into unified input, AI Agent generates reply → Output: Telegram text message

Who

Users who want to build a Telegram AI agent in n8n

You get

Bot receives and understands voice, image, video, and text messages, returning AI-generated text replies

You need

n8n · telegram · google-gemini

Steps

1. Create Telegram bot and configure credentials

   In Telegram, search for @BotFather, send /newbot, follow the prompts to create a bot, and copy the access token. In n8n, open the Telegram Trigger node, create a telegramApi credential, paste the token, and save.

2. Add LLM credentials

   In n8n's LLM nodes, add the required credentials: OpenAI (voice transcription), Google Gemini (image and video analysis), and Anthropic (AI Agent chat model). Select the models and ensure each account has credits.

3. Import workflow and verify message routing

   Import the template workflow. Confirm the Telegram Trigger node listens for message updates. The Switch node branches into five paths: /start command goes to No Operation, while voice, picture, video, and text each enter their respective processing branches.

4. Configure media processing branches

   Voice branch: Get a Audio File node fetches the voice file by file_id, then Transcribe a recording (OpenAI) converts it to text. Picture branch: Get a Photo node fetches the image, then Analyze an image (Gemini) processes it. Video branch: Get a Video File node fetches the video, then Analyze video (Gemini) processes it. All branch outputs flow into the Merge node.

   Execution prompt · Source text。Analyze video node text parameter (expression)

   Copy prompt

   ={{ $('Telegram Trigger').item.json.message.caption || $('Telegram Trigger').item.json.message.text }}

5. Set AI Agent system prompt and model

   The AI Agent node uses define prompt mode. Set systemMessage to the source-provided system prompt to ensure plain UTF-8 string output. In the Anthropic Chat Model node, select claude-sonnet-4-20250514 with maxTokensToSample set to 10000.

   Execution prompt · Source text。AI Agent node systemMessage field

   Copy prompt

   You are a helpful, intelligent assistant.
   
   Output ONLY a plain UTF-8 string for Telegram. 
   No JSON. No code fences. No prefaces. 
   If content is empty, output "…".

6. Configure memory and optional tools

   In the Simple Memory node, set sessionKey to the Telegram chat ID ({{ $('Telegram Trigger').item.json.message.chat.id }}) and contextWindowLength to 50, so the AI Agent retains conversation context per chat. The Date & Time tool can be attached as an optional tool for the AI Agent.

7. Activate and test

   Save and activate the workflow. Send voice, image, video, or text messages to your bot in Telegram and verify each branch processes correctly and returns an AI reply. If no reply is received, check the n8n execution log for credential errors or node failures.

Please build a multimodal Telegram AI bot using the following n8n workflow template. First confirm you have ready: a Telegram bot token, OpenAI API key, Google Gemini API key, Anthropic API key, and an available n8n instance. Then follow the steps to configure credentials, import the workflow, verify Switch routing, set the AI Agent system prompt, and finally activate and test.