Connecting Your Custom LLM to Vapi: A Comprehensive Guide

This guide provides a comprehensive walkthrough on integrating Vapi with OpenAI’s gpt-3.5-turbo-instruct model using a custom LLM configuration. We’ll leverage Ngrok to expose a local development environment for testing and demonstrate the communication flow between Vapi and your LLM.

​

Prerequisites

• Vapi Account: Access to the Vapi Dashboard for configuration.
• OpenAI API Key: With access to the gpt-3.5-turbo-instruct model.
• Python Environment: Set up with the OpenAI library (pip install openai).
• Ngrok: For exposing your local server to the internet.
• Code Reference: Familiarize yourself with the /openai-sse/chat/completions endpoint function in the provided Github repository: Server-Side Example Python Flask.

​

Step 1: Setting Up Your Local Development Environment

1. Create a Python Script (app.py):

from flask import Flask, request, jsonify
import openai

app = Flask(__name__)
openai.api_key = "YOUR_OPENAI_API_KEY"  # Replace with your actual API key

@app.route("/chat/completions", methods=["POST"])
def chat_completions():
    data = request.get_json()
    # Extract relevant information from data (e.g., prompt, conversation history)
    # ...
    
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo-instruct",
        messages=[
            {"role": "system", "content": "You are a helpful assistant."},
            # ... (Add messages from conversation history and current prompt)
        ]
    )
    # Format response according to Vapi's structure
    # ...
    return jsonify(formatted_response)

if __name__ == "__main__":
    app.run(debug=True, port=5000)  # You can adjust the port if needed

2. Run the Script: Execute the Python script using python app.py in your terminal. This will start the Flask server on the specified port (5000 in this example).

3. Expose with Ngrok: Open a new terminal window and run ngrok http 5000 (replace 5000 with your chosen port) to create a public URL that tunnels to your local server.

​

Step 2: Configuring Vapi with Custom LLM

1. Access Vapi Dashboard: Log in to your Vapi account and navigate to the “Model” section.

2. Select Custom LLM: Choose the “Custom LLM” option to set up the integration.

3. Enter Ngrok URL: Paste the public URL generated by ngrok (e.g., https://your-unique-id.ngrok.io) into the endpoint field. This will be the URL Vapi uses to communicate with your local server.

4. Test the Connection: Send a test message through the Vapi interface to ensure it reaches your local server and receives a response from the OpenAI API. Verify that the response is displayed correctly in Vapi.

​

Step 3: Understanding the Communication Flow

1. Vapi Sends POST Request: When a user interacts with your Vapi application, Vapi sends a POST request containing conversation context and metadata to the configured endpoint (your ngrok URL).

2. Local Server Processes Request: Your Python script receives the POST request and the chat_completions function is invoked.

3. Extract and Prepare Data: The script parses the JSON data, extracts relevant information (prompt, conversation history), and builds the prompt for the OpenAI API call.

4. Call to OpenAI API: The constructed prompt is sent to the gpt-3.5-turbo-instruct model using the openai.ChatCompletion.create method.

5. Receive and Format Response: The response from OpenAI, containing the generated text, is received and formatted according to Vapi’s expected structure.

6. Send Response to Vapi: The formatted response is sent back to Vapi as a JSON object.

7. Vapi Displays Response: Vapi receives the response and displays the generated text within the conversation interface to the user.

By following these detailed steps and understanding the communication flow, you can successfully connect Vapi to OpenAI’s gpt-3.5-turbo-instruct model and create powerful conversational experiences within your Vapi applications. The provided code example and reference serve as a starting point for you to build and customize your integration based on your specific needs.