How to Create an AI Agent in RecallSync

This guide walks you through the process of creating a new AI Agent in RecallSync. An agent is your core automated sales and communication entity, and it consists of two main parts: the Primary Agent (defining its overall purpose and resources) and the Channel Agents (defining how it communicates on specific platforms like Voice, Email, or Instagram).

Written By ReCallSyncAI

Last updated 5 months ago

Prerequisites

  • You must have an active RecallSync account.

  • Relevant LLM API Keys (e.g., OpenAI) must be configured in Settings → Integration → LLM Keys (See [Integration Guide] for details).

Step 1: Create the Primary Agent

The Primary Agent serves as the container and central configuration for all associated Channel Agents.

  1. Navigate to the Agents section from the main sidebar. (0:05)

  2. Click the + Create Primary Agent button in the top right corner. (0:07)

  3. A pop-up window will appear, prompting you for the following details:

    • Name: Enter a clear, descriptive name (e.g., "Qualification Agent," "Outbound Sales"). (0:13)

    • Description (Optional): Describe the purpose and role of this agent.

    • Calendar Integration (Optional): Select a calendar (e.g., CAL) to integrate. This is used for scheduling appointments by the agent. (0:18)

    • Training Configuration (Optional):

      • Training URL: Configure the URL for managing the agent's primary data and knowledge base.

      • Training Token: Enter the required authentication token.

  4. Click Create to finalize the Primary Agent. (0:22)

A success message will confirm the creation, and your new Primary Agent card will appear on the Agents page. (0:23)

Step 2: Add a Channel Agent

A Channel Agent specifies which communication channel (e.g., Instagram, Voice Call) the Primary Agent will use.

  1. On your newly created Primary Agent card, click the Add Agent button. (0:26)

  2. A Create AI Agent pop-up will appear:

    • Name: Enter a name for the channel-specific agent (e.g., "Instagram Qualifier"). (0:32)

    • Description (Optional): Enter a brief description. (0:35)

    • Agent Type: Select the type (e.g., INTEGRATED or STANDARD).

    • Channel: Select the communication channel (e.g., EMAIL, WHATSAPP, INSTAGRAM, VOICE_CALL, LIVE_CHAT). (0:38)

    • Provider: Select the integrated service provider for that channel (e.g., GHL for GoHighLevel, NBN). (0:42)

    • Agent Status: Toggle the switch to Activate or Deactivate the agent.

  3. Click Create. (0:46)

The new Channel Agent will now appear listed under the Primary Agent card.

Step 3: Configure the Agent's Conversation Flow

The Agent Builder is where you define the AI agent's conversation logic, prompts, and actions.

  1. On the Channel Agent card, click Go To Builder. (0:52)

  2. The Builder interface will open with a default Start node. (0:55)

Option A: Building a Flow from Scratch

  1. Click the + Add Node button in the upper right. (0:58)

  2. Select a node type to add it to the canvas: (1:02)

    • Milestone: Define a key step or objective in the conversation flow (e.g., INITIAL, QUALIFIED).

    • End: Terminates the agent's workflow for a specific outcome.

    • Send Message: Sends a text/voice message to the user.

  3. Connect the nodes to define the conversation path.

Option B: Importing a Pre-Built Flow

  1. Click the Import icon (Cloud with up arrow) in the top-right corner. (1:17)

  2. In the pop-up, select a previously exported flow file (typically a .json file). (1:20)

  3. Click Open. The flow will be imported and displayed on the canvas. (1:21)

Step 4: Define Agent Settings (Model & Prompt)

For your agent to perform its role correctly, you must define its core personality and instructions.

  1. Click the Agent Settings icon (gear) in the upper right of the Builder interface. (1:37)

  2. In the pop-up window:

    • Model: Select the underlying Large Language Model that powers the agent (e.g., GPT-4o Mini).

    • AI Prompt: This is the most critical setting. Enter a detailed system prompt that defines:

      • Context: What the agent does (e.g., "You are a helpful assistant for booking meetings...").

      • Behaviour: How the agent should interact (e.g., "Keep responses under 30 words," "Be supportive and responsive").

      • Goal/Objective: The desired outcome of the conversation (e.g., "Find the discovery call time and book a 15-minute appointment").

      • Input/Output Conditions: Define specific conditions for handling user input. (1:39)

  3. Click Save Settings. (1:43)

Step 5: Test the Agent

Always test your agent's flow and prompt to ensure it behaves as expected.

  1. In the Builder interface, click the Test Chat button at the bottom center. (1:48)

  2. A side panel will open for a live chat simulation.

  3. First, select a Test Lead from your Leads list. (1:49)

  4. Begin typing messages into the chat box to interact with your agent and confirm its responses match your defined prompt and flow. (2:04)