Skip to main content
DialNexa agent functions are tools the agent can use during a live conversation. Use them when the caller expects the agent to do something, not only say something: end the call, book time, check or update data, or call your own API.
Give the agent tools like you would give a person tools: only the ones needed for the job, with labels that make misuse unlikely.

​
Before You Begin

Define the exact caller request that should trigger the function, required inputs, expected response, timeout behavior, and what the agent should say when the action fails.

​
Function Types In The Builder

The Functions panel shows configurable actions and the platform-managed End Call behavior. If the action is a dashboard-managed provider such as WhatsApp or email, also read using integrations in agents. If the action should run after a workflow branch rather than during the live call, use workflow Application nodes.

​
Platform-Managed End Call

End Call appears automatically on new agents and has no edit or delete controls. Its displayed description explains the behavior; editing a custom function description does not configure hangup detection. Use Call Settings for silence and duration limits, and test the intended closing behavior. The End Call row alone does not disable predictive preprocessing or require a tool-calling model. Those rules apply when you add callable functions such as calendar booking or an external API action.

​
Function Descriptions Matter

The model relies on descriptions to decide when to call a function.

Name for the model

Use a short function display name that describes the action.

When to call it

Write a description that tells the agent the exact moment the tool is appropriate.

What data is needed

Use dynamic variables or schema fields for required arguments.

What to say while waiting

Custom functions can use a holding message when the API takes time.

​
Add A Function Safely

1

Choose the action

Pick built-in function or custom function based on the job.
2

Write a precise description

Tell the model when to use it and what outcome it creates.
3

Configure required inputs

Add credentials, endpoint fields, event type, timezone, or API schema as needed.
4

Test the call path

Ask the caller something that should trigger the function and confirm the result.
5

Review the call log

Check transcript and function-related evidence after the call.

​
Function Mistakes

The dashboard rejects duplicate function names. Use clear unique names.
The model may call the tool too early, too late, or never.
A working API can still create awkward silence. Use timeout and holding messages wisely.
Check version state before assuming an edit changed live behavior.

​
Verify The Function

Test a successful call, missing input, provider rejection, and timeout. Confirm the agent explains the result correctly and the call or invocation evidence contains enough detail to diagnose failures.

Custom Functions

Configure external API calls.

Post-Call Analysis

Extract results after tools run.

Testing Agents

Trigger tools in realistic calls.

Using Integrations In Agents

Connect provider actions when the agent needs a managed integration.