Get Started
Back to Templates

Conversational VK Bot with GPT-4 Mini & Whisper Voice Transcription

Created by

Created by: Konstantin || zzsnowballzz

Konstantin

Last update

Last update 11 hours ago

Categories

Share

How it works

This workflow powers an intelligent, conversational AI bot for VK that can understand and respond to both text and voice messages. The bot uses an AI agent with built-in memory, allowing it to remember the conversation history for each unique user (or in each chat) and answer follow-up questions. It's a complete solution for creating an engaging, automated assistant within your VK group.

Step-by-step

  • VK Webhook (Trigger): The workflow starts when the Webhook node receives a new message from your VK group.
  • Duplicate Filtering: The data immediately passes through the Filter Dubles node, which checks for the x-retry-counter header. This is a crucial step to prevent processing duplicate retry requests sent by the VK API.
  • Voice or Text Routing: A Voice/Text (Switch) node checks if the message contains text (message.text) or a voice attachment (audio_message.link_mp3).
  • Voice Transcription: If it's a voice note, the Get URL (HTTP Request) node downloads the audio file. The file is then passed to the Transcribe (OpenAI) node, which uses the Whisper model to convert the audio to text.
  • Input Unification: Both the original text (from the 'Text' path) and the newly transcribed text (from the 'Voice' path) are routed to the Set Prompt node. This node standardizes the input into a single prompt variable.
  • AI Agent Processing: The prompt variable is passed to the AI Agent. This agent is powered by an OpenAI Chat Model and uses Simple Memory to retain conversation history, using the VK peer_id as the sessionKey. This allows it to maintain a separate history for both private messages and group chats.
  • Response Generation: The successful AI response is passed to the Send to VK (HTTP Request) node, which sends the message back to the user.
  • Error Handling: The AI Agent node has error handling enabled (onError). If it fails, the flow is redirected to the Error (HTTP Request) node, which sends a fallback message to the user.

Set up steps

Estimated set up time: 10 minutes

  1. Add your OpenAI credentials to the OpenAI Chat Model and Transcribe nodes.
  2. Add your VK group's API Bearer Token credentials to the two HTTP Request nodes named Send to VK and Error.
  3. Webhook Setup (Important!): This is a two-stage process: confirmation and operation.
    • Copy the Production Webhook URL from the Webhook node.
    • Stage A: Confirm Address (One-time)
      • In the Webhook node settings, set Response Mode to On Received.
      • In Options -> Response Data, temporarily paste the confirmation string that VK provides.
      • Activate the workflow (toggle "Active" in the top-right).
      • Paste the URL into your VK group's Callback API settings (Management -> API -> Callback API) and click "Confirm".
    • Stage B: Operational Mode (Permanent)
      • Return to the Webhook node.
      • Set Response Mode to Immediate.
      • In Options -> Response Data, type the word ok (lowercase).
      • Save and reactivate the workflow. The bot is now live.
  4. (Optional) Customize the system prompt in the AI Agent node to define your bot's name and personality.