Skip to main content

Overview

Tools let your AI agents do more than answer questions. With tools configured, your agents can schedule appointments, transfer calls, update systems, and connect to your existing business workflows — all during a live conversation.
Tools execute automatically based on conversation context and the prompt you provide to the agent. You define when and how tools should be used through prompt engineering.

What Are Tools?

Tools are pre-configured capabilities that your agent can invoke during conversations to accomplish specific tasks. When a customer requests something like booking an appointment or speaking to a specialist, your agent executes the appropriate tool automatically.

How Tools Work

During conversations, your agent decides whether to respond directly or execute a tool: The agent uses tool names and descriptions to understand what each tool does. These details help the model select the right tool at the right moment. Best practices:
  • Give tools clear, descriptive names (e.g., “book_appointment” not “tool1”)
  • Write detailed descriptions explaining what the tool does
  • Add explicit guidance in your agent prompt about when to use each tool
While tool names and descriptions tell the agent what a tool does, your agent prompt should specify when to use it. For example: “When a customer asks to speak to a human, use the transfer_to_support tool.”

Available Tool Types

In the agent editor, the Tools tab add menu includes:
  • Transfer Call
  • Calendar Booking
  • Custom Action
  • Web Search (Alpha)
  • Calculator (Expert Mode)
  • MCP Server (Expert Mode)
Call ending is no longer a separate addable tool. Configure it in ConversationCall End with Allow AI to hang up.
For tool-specific setup, see Transfer Tools, Calendar Booking, Custom Action, Web Search, Calculator, and MCP Servers. For live variable and result flow, read Conversation Data Flow.

When Tools Execute

Tools execute during the conversation when triggered by your agent based on its prompt. Unlike traditional IVR systems that follow rigid scripts, AI agents use contextual understanding to determine when tools are appropriate.

Trigger Mechanisms

Instruction-Based Triggers:
Conditional Triggers:
Multi-Step Workflows:
Tools execute in real time during the call. Ensure connected systems are reliable and respond quickly to avoid awkward pauses in conversation.

Configuring Tools

All tool configuration happens in your agent editor under the Tools tab.
1

Navigate to Tools

Open your agent in the editor and click the Tools tab
2

Choose Tool Type

Browse the available tools and click Add to configure a tool
3

Configure Tool

Fill in the tool-specific configuration form:
  • Name: Give your tool a clear, descriptive name
  • Description: Explain what this tool does
  • Tool-specific settings: Configure parameters based on the tool type
4

Save

Save the tool to add it to your agent’s toolkit

Tools Table

Configured tools appear in a single table in the Tools tab. Typical columns include:
  • Name - Tool display name
  • Type - Tool type (for example Transfer, Booking, Custom, Web Search, MCP Server)
  • Category - Subtype or provider detail
  • Details - Key configuration details
Use Add to create more tools, click a row to edit, and use the remove action to delete.

Referencing Tools in Your Prompt

To use tools, reference them by exact name in your agent’s prompt:

Direct Reference

With Conditions

With Parameters

Tool names are case-sensitive and must match exactly as configured. If you rename a tool, update all references in your prompt.

Configuration Best Practices

Begin with basic tools before adding complex integrations. Add one tool at a time, test thoroughly, then add the next.Example progression:
  1. Add Transfer to Support
  2. Add Booking or Web Search
  3. Add Custom Action
  4. Add MCP Server only if you need external tool catalogs
Tool names are critical because the model uses them to understand what each tool does. Use descriptive, action-oriented names that clearly convey the function’s purpose.Why this matters: The model relies on function names and descriptions to detect when a function needs to be called and choose the right tool for the task.Good names:
  • “Get Customer Account” - Clear action verb + specific target
  • “Transfer to Billing Department” - Specific destination included
  • “Book 30-Minute Consultation” - Includes relevant details
Poor names:
  • “Tool 1” - No context about what it does
  • “Transfer” - Too generic, unclear where
  • “API Call” - Doesn’t describe the tool
Tool descriptions help the model understand what the tool does. The description should explain the tool’s purpose, what it returns, and what parameters it uses.Best practices from OpenAI function calling:
  • Clearly describe what the tool does and what it returns
  • Explain what parameters or data it uses
  • Use precise language that guides the model’s understanding
  • Keep it concise yet comprehensive
Example:
Note: Describe what the tool does in the description. Specify when to use it in your agent prompt.
Test each tool in the agent test interface before going live:
  • Verify tool executes correctly
  • Test success scenarios
  • Test failure scenarios
  • Verify error handling
  • Check conversation flow
Configure fallback behaviors for when tools fail. Instruct your agent what to do when tools don’t work.
Ensure agents gather required data before executing tools. Don’t attempt to book appointments without email addresses or transfer calls without explaining why.
Use appropriate authentication for all custom tools. Never expose API keys or credentials in URLs or unencrypted fields.
  • Use Bearer tokens for API authentication
  • Use Basic auth over HTTPS only
  • Store sensitive credentials securely
  • Rotate credentials regularly

Testing Tools

Some tools only work during a real phone call. Transfers to a phone number or SIP address cannot be tested in the web simulator — the transfer action requires an actual telephony connection. Use Test Agent → Phone call to test these.
Before deploying agents with tools, thoroughly test in the dashboard test interface:
1

Open Test Interface

Click Test Agent in the agent editor’s top-right corner
2

Start Web call

Click Start web call to begin a test conversation
3

Trigger Each Tool

Run through scenarios that trigger each configured tool
4

Verify Execution

Check that tools execute correctly and handle responses appropriately
5

Test Failure Cases

Simulate failures to verify error handling works as expected
6

Review Transcript

Examine the conversation transcript to ensure flow is natural and tools integrate smoothly

What to Test

For Transfer Tools:
  • Transfer executes to correct destination
  • Agent-transfer hold music plays if configured
  • Transfer message is appropriate
  • Cold transfer behavior and fallback handling work correctly
For Booking Tools:
  • Availability is retrieved correctly
  • Booking confirms successfully
  • Email/SMS notifications send properly
  • Timezone handling is accurate
For Custom Actions:
  • Connected systems respond successfully
  • Authentication works
  • Response data is available to agent
  • Error responses are handled gracefully
For All Tools:
  • Agent references tool by correct name
  • Agent gathers required information first
  • Conversation flow remains natural
  • Failures don’t break the conversation
Test calls use real integrations. If you’re testing a booking tool, it will create real appointments in your Cal.com account. Clean up test data afterward.

Common Use Cases

Customer Support Workflow

Appointment Booking Workflow

Sales Qualification Workflow


Troubleshooting Common Issues

Problem: Agent doesn’t use the tool even though it should.Solutions:
  • Verify the tool name in your prompt matches exactly (case-sensitive)
  • Verify the tool exists in the Tools table and any required integration is connected
  • Make your prompt more explicit about when to use the tool
  • Test in isolation by explicitly asking the agent to use the tool
  • Review the conversation transcript to see the agent’s reasoning
Problem: Agent verbally confirms it’s executing a tool (e.g., “I’m transferring you now”) but the tool doesn’t execute until the next conversation turn.Why this happens: The agent generates a response and executes the tool in the same turn, but only one can happen per turn.Solution: Prompt the agent to ask for user confirmation before executing tools:
This ensures the agent executes the tool immediately after receiving confirmation, not in the same turn as announcing it.
Problem: Custom tools fail with 401/403 errors.Solutions:
  • Verify credentials are correct and not expired
  • Check authentication type matches the connected system’s requirements
  • Ensure Bearer tokens include “Bearer” prefix if needed
  • Ask the owner of the connected system to confirm the connection works outside the agent
  • Review the connected system’s authentication requirements
Problem: Cal.com booking tool returns errors.Solutions:
  • Verify Cal.com integration is connected
  • Check event type exists and is active
  • Ensure meeting platform is configured in Cal.com
  • Verify scheduling windows allow requested dates
  • Check timezone configuration
  • Test booking manually in Cal.com to verify availability
Problem: Transfer tool executes but doesn’t connect.Solutions:
  • Verify destination is in correct E.164 format (e.g., +15551234567)
  • Ensure phone number is reachable (for phone transfers)
  • Verify SIP address is correct (for SIP transfers)
  • Test destination independently

Next Steps

Transfer Tools

Configure destinations, modes, routing patterns, and call end

Booking Integration

Set up Cal.com appointment scheduling with email and SMS notifications

Custom Action

Connect your agents to external systems and APIs

Conversation Data Flow

Understand how values move into live tools and post-call workflows

Prompt

Learn how to write an effective agent prompt

Prompt Engineering

Master advanced prompting techniques for reliable agent behavior

Common Questions

Yes. The agent selects tools based on the name and description you provide. Use clear, descriptive names like “Transfer to Sales” or “Book Appointment” — not “Tool 1” or “Action A”. Names are case-sensitive.
Built-in tools (Transfer Call, Calendar Booking, Web Search) require no code. Custom API Actions need an API endpoint to call, but you configure them entirely in the UI — no SDK or server-side code required.
The agent receives the error response and can handle it gracefully if your prompt includes fallback instructions. Check the conversation timeline to see the exact request and response for any tool call.