iSales AI

Create AI Team

Building Your Custom AI Team: JSON Configuration Guide

The Create an AI Team section empowers you to design a bespoke team of AI agents, tailored precisely to your business needs. You achieve this by uploading a JSON (JavaScript Object Notation) file that meticulously defines the roles, capabilities, instructions, and resources for each member of your AI team, including its manager.

As indicated on the page, “Your AI team status” will show if a custom team is currently active (e.g., “Disabled”).

The Blueprint: Your AI Team’s JSON File

To create your custom AI team, you’ll construct a JSON file. This file acts as the blueprint, containing:

  • Job descriptions for your AI “employees” (agents).
  • Instructions for the chief AI manager who oversees the team.
  • Data sources available to the manager and agents.
  • Tools that your AI agents can use to perform tasks.

Crucial First Step: Before uploading, it is vital to validate your JSON file to ensure it’s correctly formatted. An invalid JSON file will not be processed correctly. You can use online JSON validators for this purpose.

Understanding the JSON Structure

Your AI team’s configuration is defined within a specific JSON structure. Let’s break it down:

{
  "manager": {
    "llm": "string",        // Name of the LLM model for the manager
    "agent_prompt": "string", // Main prompt that controls employee selection and task delegation
    "data": ["string"]      // List of data sources available to the manager
  },
  "agents": [               // List of AI employee objects
    {
      "llm": "string",            // Name of the LLM model for the employee
      "agent_name": "string",       // Job title/name of the employee
      "agent_prompt": "string",     // Detailed prompt/instructions for the employee
      "agent_description": "string",// Brief description of the employee and their role
      "data": ["string"],           // Data sources available to this employee
      "tools": ["string"]           // List of tools this employee can use
    }
    // Add more agent objects here if needed
  ]
}

1. The manager Object:

The AI team manager oversees operations, delegates tasks, and selects the appropriate agent for a given job.

  • "llm": (String) The name of the Large Language Model (LLM) that will power the manager. Choose one from the “List of LLM models” below.
  • "agent_prompt": (String) This is the primary instruction set for the manager. It should detail how the manager should analyze incoming requests, select the appropriate AI agent from the agents list for a task, and how to delegate work.
  • "data": (Array of Strings) A list of data sources the manager can access to make decisions. See “Available Data Sources” below.

2. The agents Array:

This array contains objects, where each object represents an individual AI “employee” in your team.

  • Each Agent Object:
    • "llm": (String) The name of the LLM model for this specific agent. Choose one from the “List of LLM models.”
    • "agent_name": (String) The job title or name for this agent (e.g., “Sales_Closer_Bot”, “Support_Tier1_Agent”).
    • "agent_prompt": (String) Detailed instructions, persona, and operational guidelines for this agent. This is where you define how the agent should perform its tasks.
    • "agent_description": (String) A brief description of what this agent does and its role. This helps the manager (and you) understand its specialty.
    • "data": (Array of Strings) Data sources this specific agent can access. See “Available Data Sources.”
    • "tools": (Array of Strings) A list of tools this agent is equipped to use. See “Available Tools for Employees.” If no tools are needed, use an empty array [].

Available Resources for Configuration:

List of LLM Models:

You must choose only one LLM name for the manager and for each agent.

  • "claude-3.5-sonnet" (Anthropic)
  • "claude-3.5-haiku" (Anthropic)
  • "gpt-4o-mini" (OpenAI)
  • "gpt-4o" (OpenAI)
  • "o1-mini" (OpenAI)
  • "o1" (OpenAI)
  • "deepseek" (DeepSeek)
  • "grok-2-1212" (Grok)

Available Tools for Employees (Agents):

Tools can only be assigned to agents, not the manager. If an agent doesn’t need tools, use an empty array: [].

  • "InternetSearch": Allows the agent to perform internet searches.
  • "AssistantRequestMessage": Enables the agent to send a request or escalate to an admin/human.
  • "CreateMeme": The agent can generate memes.
  • "SendFreePhoto": Agent can send a free photo.
  • "SendFreeVideo": Agent can send a free video.
  • "SendPaidPhoto": Agent can send a photo that requires payment.
  • "SendPaidVideo": Agent can send a video that requires payment.
  • "SendInvoice": Agent can generate and send an invoice.
  • "StopChatting": Agent can end the current conversation.

If you require a custom tool not listed, please contact iSales AI support.

Available Data Sources:

Data sources provide context and information that both the manager and agents can use for decision-making and task execution.

  • "CurrentTime": Provides the current UTC time.
  • "PurchaseHistory": Access to invoice history and their statuses.
  • "ChatHistory:X": Provides the chat history. Replace X with a number to specify how many recent messages to include (e.g., "ChatHistory:10" for the last 10 messages).

Navigating the “Create an AI Team” Interface:

  • 📄 Instructions for creating a JSON file for your AI team: This link or button will provide the official, detailed guidelines and specifications directly from iSales AI for structuring your JSON file. Always refer to this for the most current and precise information.
  • 📤 Upload JSON Instruction: Click this button to select and upload your prepared and validated JSON configuration file from your device.
  • ↩️ Back: Returns you to the previous menu.

By carefully constructing your JSON file with clear prompts, appropriate LLMs, tools, and data sources, you can build a powerful, custom AI team ready to tackle your specific business challenges. Remember to validate your JSON before uploading!