This n8n AI Voice Agent Will Call Anyone For You (100% automatic)

Key Concepts

Voice AI agent, nadn, Telegram, ChatGPT, web scraping, Google Maps, HTTP requests, Appify, vAPI, AI assistant, calendar integration, fallback mechanisms, prompt engineering, voice cloning.

Voice AI Agent Overview

The video demonstrates the creation of a voice AI agent using nadn capable of making phone calls to businesses or individuals. The agent automates tasks like finding restaurants, booking appointments, and adding events to a calendar. The creator emphasizes the experimental nature of the project and the importance of thorough testing before deployment.

Workflow Breakdown

1. Trigger (Telegram): The process begins with a message received via Telegram. Other platforms like Twilio (SMS) or WhatsApp could also be used.
2. Message Processing:
   • Audio Transcription: If the message is an audio file, it's downloaded and transcribed using an OpenAI module.
   • Text Merging: Regardless of whether the input is audio or text, the message content is extracted and stored in a json.text field.
3. AI Agent (Reasoning):
   • The core of the system is an AI agent powered by a chat model (e.g., ChatGPT).
   • The agent uses a prompt to understand the user's intent (e.g., finding restaurants, booking appointments).
   • The prompt defines the agent's role, available tools, and rules for operation.
   • The agent has a memory (window buffer memory) to remember past conversations (default: 5 messages).
4. Tools: The AI agent utilizes various tools to fulfill requests:
   • Personal Contact Finder: Searches Google Contacts for phone numbers and email addresses.
   • Google Maps Web Scraper: Uses Appify to extract business listings from Google Maps.
   • vAPI Integration: Uses vAPI to make phone calls and manage voice interactions.
   • Calendar Integration: Adds confirmed bookings to the user's calendar.

Example: Restaurant Booking

1. The user sends a message like, "Find me five Indian restaurants in Montreal."
2. The AI agent uses ChatGPT to understand the request.
3. The agent uses the Google Maps web scraper (via Appify) to find five Indian restaurants in Montreal.
4. The results are sent back to the user via Telegram.
5. The user selects a restaurant (e.g., "Pick number two").
6. The user provides booking details (e.g., "Book a table for 2 tonight between 5:00 to 9:00 p.m.").
7. The AI agent uses vAPI to call the restaurant.
8. The AI agent interacts with the restaurant employee, providing booking details and handling potential issues (e.g., no window table available).
9. The call transcript is analyzed, and if the booking is successful, the AI agent adds the reservation to the user's calendar and sends a confirmation message.

Example: Hair Appointment Booking

1. The user sends a message like, "I'd like to book a hair appointment."
2. The AI agent prompts for more details (date, time, service).
3. The user provides details and specifies a preferred barber shop.
4. The AI agent uses vAPI to call the barber shop.
5. If the barber shop is unavailable, the AI agent follows a fallback plan (e.g., scheduling another call).

Google Maps Web Scraper (Appify)

This section details how the AI agent uses Appify to scrape data from Google Maps.

1. Trigger: The workflow is triggered by the main AI agent.
2. HTTP Request (Appify):
   • A POST request is sent to Appify's API.
   • The request includes the API key, actor ID (Google Maps extractor), and search parameters (e.g., "Indian restaurants in Montreal").
   • The waitForFinish parameter ensures the workflow waits for the web scraping to complete (up to 300 seconds).
3. HTTP Request (Get Results):
   • A GET request retrieves the scraped data from Appify.
   • The API key is included for authentication.
4. Aggregator: The results are aggregated into a single list for easier processing.

Key Details:

• Appify Actors: Pre-built web scraping solutions for various tasks.
• API Key: Required for authenticating with Appify.
• Credits: Appify offers a free tier with limited credits.

Voice AI Agent (vAPI)

This section explains how the AI agent uses vAPI to make phone calls.

1. Prompt Creation:
   • OpenAI is used to generate a call prompt based on the user's request and context.
   • The prompt includes an opening line, call details, and a fallback plan.
2. HTTP Request (vAPI):
   • A POST request is sent to vAPI's API to initiate a call.
   • The request includes the API key, phone number ID, assistant ID, and call parameters.
3. Call Parameters:
   • Phone Number ID: The phone number used to make the call (obtained from vAPI or Twilio).
   • Assistant ID: The ID of the AI assistant configured in vAPI.
   • Voice: The voice used for the AI agent (e.g., "will" from PlayHT).
   • First Message: The opening message generated by OpenAI.
   • Context: The main prompt containing call details and fallback information.
4. Call Monitoring:
   • A loop repeatedly checks the call status using vAPI's API.
   • The loop waits for 5 seconds between checks.
5. Transcript Retrieval:
   • Once the call is completed, the transcript is retrieved from vAPI.
6. Calendar Booking:
   • The transcript is analyzed, and if the booking is successful, the AI agent adds the reservation to the user's calendar.

Key Details:

• vAPI Assistants: Configurable AI agents for voice interactions.
• Voice Cloning: The ability to use different voices for the AI agent.
• Fallback Mechanisms: Handling scenarios where the call doesn't go as planned.

Limitations and Considerations

The creator emphasizes that the voice AI agent is not perfect and has limitations:

• Incomplete Information: The agent may not have all the necessary information to complete a task (e.g., number of people for a restaurant booking).
• Testing Required: Thorough testing is crucial to ensure the agent works reliably for different scenarios.
• Fallback Plans: Implementing robust fallback plans is essential to handle unexpected situations.

Quote: "If you guys are going to use something like this just make sure to um really test it out make sure to have a fallback line make sure to have every everything in place um and add anything in there that you need so that you're comfortable before using it."

Conclusion

The video provides a detailed overview of building a voice AI agent using nadn, Appify, and vAPI. While the technology is promising, it requires careful planning, thorough testing, and robust fallback mechanisms to ensure reliable performance. The creator encourages viewers to experiment with the provided blueprints and adapt them to their specific needs. The community mentioned at school.com provides additional resources and support for those interested in AI agents and automation.