Purpose

Leverage a streamlined Voice Interface for you and your organization to engage with an AI Supervisor—a personal AI Assistant connected to your custom workflows and existing Agents.

Demo & Explanation

How it works

   
• Record a message in the Vagent App, which is transcribed and sent with a session ID to the registered webhook.
   
• The Main Agent acts as a router, interpreting the message with stored chat history (linked to the session ID) and choosing the appropriate tool to perform the required action. The tools, which are workflows, may contain subordinated Agents. Raw input is passed as a separate parameter.
   
• Sub-Agents handle specific tasks, each with separate chat memory (suffix added to session ID) for clear responsibility separation.
   
• An HTTP Request Tool is called as needed. The result is formatted in Markdown and returned to the Main Agent with a prompt to avoid re-interpretation.
   
• Drafts are formatted with indentation (angle brackets). If information is missing, no tool is called, and a message is returned to the user.
   
• The Main Agent then outputs the result from the Sub-Agent. If a draft is included, it is separated from spoken output.
   
• The formatted output is sent back to the webhook, split into spoken and text versions, allowing the App to read only the necessary information aloud.

For full documentation, visit: https://vagent.io/docs

Setup

   
• Import this workflow into your rantir instance.
   
• Follow the instructions in the sticky notes on the canvas.
   
• Configure credentials. OpenAI can be swapped for another LLM, but is required for App functionality. Google Calendar and Notion are needed for all features.
   
• Copy the Webhook URL from the Webhook node in the main workflow.
   
• Download the Vagent App from https://vagent.io.
   
• In the App settings, enter your OpenAI API Token, Webhook URL, and the password defined for Header Auth.
   
• Tap the Mic symbol in the App to record messages and interact with the Multi-Agent using your voice.

Tip: For testing the chat trigger, temporarily disable the nodes after the Tools Agent.