Build

Tools & function calling

Tools let your agent take actions mid-conversation — look something up, send a message, transfer the caller, or call your own systems — instead of just talking.

What tools are

A tool is a function your agent can call while a conversation is running. When the agent's LLM decides a tool is needed, it calls the tool, gets a result back, and folds that result into its reply. This is the same "function calling" pattern used across modern voice and chat AI, and it works identically on the phone, on the web chat widget, and in the dashboard.

Tools turn a talking agent into an acting one. Common uses:

Fetch live data (an order status, an availability slot) at the moment the caller asks.
Take an action the caller requests (send a confirmation, book an appointment).
Hand the conversation off — to a human teammate or another agent.
Reach into your own backend or a third-party service through an MCP server.

Creating and attaching a tool

You manage tools from the Tools section in the left nav. Attaching a tool to an agent happens in the Tools tab of the agent builder.

The two steps are separate on purpose: you create a tool once, then attach it to as many agents as you like.

Open Tools in the left nav and create a new tool.
Define what the tool does so the agent knows when to call it and what to pass.
Open the agent you want in the agent builder, go to its Tools tab, and attach the tool.
Save the agent.

Tip: An agent only knows about the tools attached to it in its Tools tab. If an agent isn't calling a tool you expect, confirm the tool is attached to that specific agent and that the agent was saved.

Testing function calling

To see a tool in action, attach it to an agent and test the agent itself: use Test chat for a chat agent, or place a call to a voice agent, then say something that should trigger the tool. Review what happened afterward in History (past conversations, with transcripts) and Call logs.

Tip: Make the tool's description clear about when it should be used — that description is what the agent's model reads to decide whether to call it.

Built-in capabilities

Callable ships built-in capabilities you can give an agent without writing anything custom:

Transfer to human — the agent hands the live call off to a person.
Send SMS — the agent sends a text message during the conversation (for example, a confirmation or a link).
End call — the agent ends the call when the conversation is done.

These pair well with a scripted Conversational Flow, which also has dedicated Transfer and End nodes for stepping a structured conversation to a hand-off or a clean close. See Conversational flow for building flows and Agent teams for multi-agent hand-offs.

Adding MCP servers as tools

You can add MCP (Model Context Protocol) servers as callable tools, which lets your agent use whatever functions that server exposes. Add the MCP server as a tool, then attach it to an agent the same way you'd attach any other tool — in the agent's Tools tab. From the agent's perspective the MCP server's functions become tools it can call mid-conversation.

Next steps

Conversational flow — script structured conversations with Transfer and End nodes.
Agent teams — have multiple agents hand off to each other.
Knowledge base — ground your agent's answers in your business content.
Integrations — connect LLM, voice, and data providers.

← Knowledge Base

Conversational Flows →