Skip to main content

​
Voice AI Agent Function calls

In one line: Function calls let your voice agent do things during a live call — not just talk. Mid-conversation, the agent can hand off to a human, look up availability, book a meeting, or hit your own API, then keep speaking naturally with the result.

​
What is a function call?

A voice agent’s core skill is conversation. But most useful tasks need it to reach outside the conversation — into a calendar, a CRM, a phone system, or your backend. A function call (also called tool calling) is how the agent does that. You describe an action to the agent as a function: a name, a plain-English description of when to use it, and the parameters it needs. During a call, when the caller’s intent matches that description, the agent:
  1. Decides the function is needed based on what the caller said.
  2. Extracts the parameters from the conversation (e.g. a date, a name, a phone number).
  3. Calls the function — UnleashX runs it against the real service.
  4. Uses the result — the agent speaks the outcome back to the caller and continues the conversation.
The key idea: you define the “what” (the function and its parameters), and the agent decides the “when” and fills in the “with what” from natural speech. No rigid menus or keypad trees.
Function calls run in real time, mid-call. The caller experiences it as a natural pause (“let me check that for you…”) rather than a handoff to a form.

​
How it works in a voice call

Because the whole loop happens during the call, latency matters. UnleashX streams the agent’s speech, calls the function, and resumes speaking as soon as the result returns — so the caller hears a brief, natural “one moment” rather than dead air.

​
Built-in function calls

UnleashX ships with ready-made functions for the most common voice-agent actions. Turn them on for an agent and it can use them immediately — no code required.

Transfer calls

Hand the live call off to a human or another number when the caller asks for a person, hits an edge case, or needs escalation. The agent transfers the active call to a configured destination and can pass along context.

Check calendar slots

Look up real-time availability on a connected calendar. The caller asks “what’s open Thursday afternoon?” and the agent reads back the free slots.

Book calendar slots

Create a booking on the connected calendar once the caller picks a time — capturing name, time, and any details, then confirming the appointment out loud.

Custom function calls

Define your own function that calls your API or workflow. Give it a name, a description, and parameters, and the agent will invoke it whenever the conversation calls for it. See below.

​
Custom function calls

When the built-ins aren’t enough, define a custom function to connect the agent to any service you can reach over HTTP — your order system, a lookup service, a payment link generator, or an internal workflow. A custom function is defined by three things: At call time, UnleashX sends the extracted parameters to your endpoint, waits for the response, and hands the result back to the agent to speak.
Write the description like an instruction to a new teammate: say exactly when to use the function and when not to. A clear description is the single biggest factor in whether the agent calls the right function at the right moment.

​
Setting up a function call

1

Open your agent

Sign in at https://www.tryunleashx.com and open the voice agent you want to extend.
2

Add a function

In the agent’s Function calls section, enable a built-in (Transfer, Check slots, Book slots) or choose Add custom function.
3

Configure it

For built-ins, connect the required service (e.g. a calendar or transfer destination). For a custom function, set the name, description, parameters, and the endpoint to call.
4

Test in the playground

Run a test call and speak the trigger scenario. Confirm the agent calls the function, extracts the right parameters, and speaks the result naturally.

​
Best practices

  • One job per function. Narrow, single-purpose functions are easier for the agent to choose correctly than one function that does many things.
  • Describe the trigger, not the mechanism. The description should say when to use the function in the caller’s terms, not how your backend works.
  • Mark optional vs. required parameters. If a parameter is required, the agent will ask the caller for it before calling — make sure that question feels natural.
  • Keep endpoints fast. Since the caller is waiting on the line, aim for responses in a second or two; return a graceful message the agent can speak if something goes wrong.
  • Confirm irreversible actions. For bookings, transfers, or anything that changes data, have the agent read back the details and confirm before it acts.