Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.sigmamind.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Agent Builder Prompting Guide

Everything you need to build your AI agent - no coding required.

​
What is this guide?

This guide walks you through building an AI voice agent from scratch - step by step. You’ll learn how to write instructions for your agent, connect it to external tools, set up call transfers, and pull useful insights from your conversations.
By the end of this guide, you’ll be able to:
  • Write clear agent instructions from scratch
  • Use variables to personalize conversations
  • Connect your agent to external tools and APIs
  • Set up call transfers to human agents
  • Export and use post-conversation insights

​
Section 1 — Writing Your Agent’s Instructions

Think of your agent’s instructions like a job description. You’re telling it who it is, how to behave, and what to do — clearly and specifically.

​
The 5-Part Structure

Use these five sections every time you write an agent prompt. Copy and fill in the blanks: 
## Identity 
You are [name], a [role] for [company/purpose].

##Style
Keep responses [short/friendly/professional]. Use simple words. Ask one question at a time.

##Rules
Always do: [list your must-dos] Never do: [list your must-nots]

##Task
Step 1: [First thing the agent should do] 
Step 2: [Second thing] 
Step 3: [Third thing]

##Objection Handling 
If the user is confused: [what to do] If the user is upset: [what to do]
Example Prompt: 
## Identity
You are Maya, a friendly support assistant for Bloom Skincare.

## Style
Be warm and helpful. Keep answers short — 2 to 3 sentences max.
Always greet the user by name if you have it.

## Rules
Always ask for the order number before looking anything up.
Never promise a refund without checking the policy first.
If you don't know the answer, say so and offer to escalate.

## Task
Step 1: Greet the user and ask how you can help.
Step 2: Collect their order number.
Step 3: Answer their question or escalate if needed.

## Objection Handling
If the user is unhappy: Apologize, acknowledge the issue,
and offer to connect them with a human agent.
Write your instructions like you’re training a new employee on their first day. Be specific. Don’t assume they’ll figure it out.

​
Section 2 — Using Variables to Personalize Conversations

Variables are placeholders that automatically fill in real information during a conversation like the customer’s name, their account number, or today’s date. How Variables Work Instead of a static greeting like “Hello customer”, you write: 
Hello {{customer_name}}, thanks for calling {{company_name}}!
When the conversation starts, the agent automatically fills in the real values: 
Hello Sarah, thanks for calling Bloom Skincare!
Where to Use Variables
  • Greetings: "Hi {{customer_name}}, how can I help you today?"
  • Confirmations: "I've updated your order {{order_number}}."
  • Escalations: "I'll connect you with a human agent now, {{customer_name}}."
  • Summaries: "Thanks for calling {{company_name}}. Have a great day!"
Variables only work if they’ve been passed into the conversation or extracted automatically during the conversation itself. If a variable isn’t available, the agent will say the placeholder literally like. Always make sure your variable names match exactly what you’ve passed in.
Variables work in double curly braces. For a full reference, see the Exract variables guide.

​
Section 3 — Connecting to External Tools & APIs

Your agent can do more than just talk — it can look up information, book appointments, send data to other systems, and more. It does this through tools and webhooks.

​
What Are Tools?

A tool is an action your agent can take. For example:
  • Look up an order status
  • Book an appointment in your calendar
  • Transfer a call to a human agent
  • Trigger a refund in your payment system
SigmaMind supports three types of tools you can add to your agent:
  • Voice Tool — controls the call itself. Use this to transfer a call to a human (cold or warm transfer), end the call automatically, or navigate an IVR phone menu using keypad inputs.
  • Extract Variables — captures information from the live conversation (like a name, email, or date) and stores it as a variable to use later in the same call.
  • Integration Tool — connects to an external app or API you’ve set up in the Tool Library (like a CRM, booking system, or custom API).

​
Step 1: Tell the Agent When to Use a Tool

Before setting anything up in the dashboard, write the trigger conditions into your agent’s instructions. This tells the agent when to call each tool. 
If the user asks for their order status → use the check_order_status tool

If the user wants to book an appointment → use the book_appointment tool → confirm the date and time with the user after booking

If the user asks for a refund → use the escalate_to_support tool → do NOT process refunds yourself

​
Step 2: Set Up the Tool in the Tool Library

SigmaMind has a built-in Tool Library where you can connect any external app or API — no coding required. If the app is already in the marketplace Cal. com, HubSpot, Shopify, Stripe, Zoho, Pipedrive, and more):
  1. Go to Tool Library in the Agent Builder
  2. Find the app and click Configure Authentication
  3. Enter your API key or token from that app
  4. The app’s actions (like book_appointment or get_available_slots) will become available to add to your agent
If your app isn’t in the marketplace (custom CRM, internal system, or any other API):
  1. Go to Tool Library → click Add New Integration
  2. Give it a name (e.g. My Booking System)
  3. Click Configure Authentication and enter your credentials
  4. Click Add New Tool and fill in the action details — give it a name, paste the API endpoint URL, set the method (GET, POST, etc.), and add any required headers
  5. Use variable_name placeholders in the URL or request body wherever the agent needs to insert live values (e.g. order_number, customer_email)
  6. Click Run Test to verify it’s working

​
Step 3: Add the Tool to Your Agent

Once your tool is ready in the Tool Library, attach it to your agent from inside the Agent Builder.
  1. Click Add Tool inside your agent
  2. Select the tool type (Voice Tool, Extract Variables, or Integration Tool)
  3. Choose the specific tool you want to use
  4. Add a short description telling the agent when to trigger it — e.g., “Call this when the user asks to book an appointment.”
  5. Click Add to Agent
For a full reference, see the Add Tool guide.

​
What is a Webhook?

A webhook is a way for SigmaMind to automatically push information to your system the moment something happens in a conversation — when a call starts, ends, or has been analysed. Instead of you going to check what happened, SigmaMind sends the data to you instantly. You can use this to update your CRM, trigger a workflow, log analytics, or notify your team — all without any manual work. SigmaMind currently supports three webhook events:
  • conversation_started — fires when a new conversation begins
  • conversation_ended — fires when a conversation is completed
  • conversation_analysed — fires when call analysis is ready after the conversation

​
How to Set Up a Webhook

  1. Open your agent and go to Agent Settings → Webhooks
  2. Click Add Webhook and fill in the following:
    • Webhook Name — a label for your own reference
    • URL — the endpoint where you want to receive the data (your CRM, Zapier, Make, Slack, or your own server)
    • Secret API Key — your SigmaMind API key, found under Settings in your dashboard
  3. Select which event should trigger it
  4. Click Submit
From that point on, SigmaMind will automatically send a data payload to your URL every time the selected event fires.
Note: One webhook URL can be reused across multiple agents. For a full reference, see the Agent Webhooks.

​
Section 4 — Exporting Post-Conversation Insights

After each conversation ends, SigmaMind can automatically analyse it and extract useful information - like what the customer wanted, whether the issue was resolved, or how the call went. This is done through Post-Conversation Analysis, and the results can be sent to any external system using a webhook.

​
Step 1: Enable Post-Conversation Analysis

  1. Open your agent in the Agent Builder
  2. Click Agent Settings from the side panel
  3. Expand the Post-Conversation Analysis section
  4. Toggle it ON
Once enabled, SigmaMind will automatically analyse every conversation after it ends and extract whatever insight fields you’ve set up.

​
Step 2: Add Insight Fields

Insight fields tell the AI what to extract from each conversation. You can add as many as you need.
  1. Click Add New Field
  2. Choose the type of insight
  3. Give it a name
  4. Write a simple instruction telling the AI what to look for
  5. Click Submit
There are four types of insight fields you can use:
TypeWhat it returnsExample use
TextA written responseConversation summary, reason for call
SelectorOne value from a list you defineCall outcome: Billing / Technical / Sales
BooleanTrue or FalseWas an appointment booked?
NumberA numeric valueSentiment score from 1 to 10
Example insight fields you might set up:
  • Name: conversation_summary | Instruction: Write a concise summary of the conversation, including the main topic and final outcome.
  • Name: call_outcome | Instruction: Select the final outcome from the predefined options.
  • Name: appointment_booked | Instruction: Return true if an appointment was booked during the conversation, otherwise return false.
  • Name: sentiment_score | Instruction: Rate the overall customer sentiment from 1 to 10 where 1 is very negative, and 10 is very positive.

​
Step 3: Send the Data via Webhook

Once your insight fields are set up, connect a webhook to automatically push the extracted data to your system after every call.
  1. Go to Agent Settings → Webhooks
  2. Click Add Webhook
  3. Enter a name and your destination URL.
  4. Click Submit
SigmaMind will now send the extracted insights to your URL automatically after each conversation. You can pipe this data into your CRM, a Google Sheet, Zapier, n8n, or any custom application.
For a full reference, see the Post-Conversation Analysis guide and Webhooks guide.