1. Overview
  2. Learning the Software - Beginner
  3. Customize AI Agents

Customize AI Agents

Steps to Customize Your AI Agents

Introduction

This guide explains how to customize and refine both the voice and chat aspects of your agents in WellGrow Software. You’ll learn how to configure voice settings (including cloning your own voice), choose your preferred Large Language Model (LLM), and fine-tune additional settings, such as talk speed, temperature, and sleep mode. By following these steps, you can create a more natural, personalized, and efficient experience for your users.

1. Accessing the Assistant Settings

Navigate to the Assistants tab

  • On the left-hand menu, click Assistants
  • .Select the assistant you want to customize (e.g., "Olivia Impactful Look").

Open the Settings

  • In the top-right corner of the selected assistant’s panel, click the little gear icon (Settings).
  • This will open the Assistant Settings panel.

2. Adjusting Voice Settings

Configure Voice Attributes

Under Voice Settings, you’ll see controls for:

  • Sensitivity to Interruption: Determines how easily the assistant will pause or stop speaking when interrupted.
  • Talk Speed: Adjust this to change how quickly the AI speaks. A setting near 0.92 often balances clarity and natural flow.
  • Temperature (Emotional Range): Higher values can yield more expressive or creative tones, while lower values keep the voice more consistent and “strict.”

Select a Voice

  • Click the Select Voice option.
  • Choose from a library of voices or select different parameters such as age, gender, and accent.
  • Voice Providers include Deepgram, Eleven Labs, OpenAI, Play.ht, and more.
  • Once you find a voice you like, click Add to Assistant to assign it.

Clone Your Own Voice (Optional)

  • Click Clone your voice and upload an audio file containing samples of your voice.
  • Wait a couple of hours for the system to process the file.
  • Once completed, your cloned voice will appear in your library to be selected.

Test the Voice in Voice Lab

  • Go to Voice Lab > Start a call.
  • Listen to the real-time output to confirm you’re satisfied with how the new voice sounds.

3. Configuring Chat (Autopilot) Settings

Select an LLM Model

  • In the Autopilot section, choose your language model (e.g., GPT-4.0, Anthropic Claude 3.5, etc.).
  • Each model may differ in style, creativity, and how it understands user prompts.

Set the Response Timing

  • Wait Time: Specifies how long the AI waits before responding. For testing, set this to 0 seconds to speed up the process. For more natural, human-like interactions, use 15–20 seconds.
  • Max Responses per Agent: Defaults to around 50. Increase if you want the agent to handle longer conversations (you can have unlimited messages).

Control Agent Creativity

  • Temperature: A lower setting (e.g., 0.1–0.3) yields more predictable, knowledge-base-centric replies. A higher setting (e.g., 0.7–1.0) allows more creative or variable responses.

Enable or Disable Sleep Mode

  • Sleep Mode stops the AI from responding if you (or a human team member) manually intervene in a conversation.
  • This is useful if you have a live support team available from 9 AM to 6 PM and only want the AI active outside those hours.
  • Leave it blank if you want the AI to remain active at all times.

4. Managing Call Settings

Set Maximum Call Duration

  • Prevent prolonged calls that could incur higher costs by specifying a time limit.

Choose Background Noise

  • Add realistic call center static, mountain ambiance, or outdoor noise to simulate real-world environments.
  • Adjust volume levels for the background effect as desired.

Call Initiation and Voicemail

  • Define who initiates the call (the user or the AI) under Call Initiation.
  • Enable Voicemail Detection to play a custom message (e.g., “Call me back when you can”) if the call goes to voicemail.
  • Leave the message field blank if you want no voicemail and immediate hang-up.

Incoming and Pull Webhooks

  • Use webhooks to submit extra information or custom variables to the conversation, enabling personalized interactions.
  • This allows the AI to reference user-specific details (e.g., previous orders or account information).

5. Testing Your Configuration

Perform a Test Call or Chat

  • Use the Voice Lab to test voice settings or Test Chat for autopilot responses.
  • Notice how the AI interacts, and listen for clarity, pace, and overall natural sound.

Evaluate Response Quality

  • Check if the AI correctly references your internal data and follows your configured guidelines.
  • If the answers sound off-topic or incomplete, you may need to adjust the temperature, reduce talk speed, or add more context in the knowledge base.

6. Refining Your Agent Settings

Adjust Parameters Based on Results

  • If the AI is too rigid or repetitive, slightly increase the temperature setting.
  • If it goes off-script, lower the temperature or add more relevant data to the knowledge base.

Update or Add More Data

  • Add further FAQs or detailed service descriptions to your knowledge base for more accurate replies.
  • Use real customer interactions to identify where the AI might be missing key information.

Retest and Iterate

  • Keep testing the agent after each tweak.
  • Aim to strike a balance between creative, human-like conversation and factual accuracy tied to your company’s specifics.

Key Tips for Refinement

  • Keep Temperature Low for Strict Adherence: If you want the AI to stick closely to scripted information (e.g., compliance or legal statements), set the temperature lower.
  • Use Real Calls or Chats to Identify Gaps: Record typical questions or scenarios and test them against the AI’s responses to pinpoint any inaccuracies.
  • Leverage Team Notes: If multiple team members are overseeing the AI, use Team Notes to communicate tweaks or important observations about the prompt.
  • Employ Webhooks for Personalization: Passing custom variables via webhooks can make calls feel more tailored and human-like.
  • Regularly Update Settings: Business offers, pricing, or operational hours might change. Keep your knowledge base and agent settings current to maintain optimal performance.

By following the steps above and continually refining the AI’s parameters, you’ll be able to create a highly efficient, natural-sounding virtual assistant that accurately represents your brand - whether it’s via voice calls or chat interactions.

Was this article helpful?