> ## Documentation Index
> Fetch the complete documentation index at: https://dialnexa.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Using Integrations In DialNexa Agents

> Add integration-backed functions to DialNexa agents with clear action descriptions, required fields, credentials, and call evidence.

DialNexa agents can use integration-backed actions when those actions are available to the agent function surface. Use this when the agent should perform a specific external action during the call, such as sending a message after consent or triggering a follow-up.

<Tip>
  Do not give a caller-facing agent an integration action just because the action exists. Give it a reason and a rule.
</Tip>

## Before You Begin

Connect and validate the provider, identify the exact caller intent that should trigger the action, define required fields, and decide what the agent says when the action fails.

## DialNexa Agent Integration Responsibilities

| Layer                  | Responsibility                                                                             |
| ---------------------- | ------------------------------------------------------------------------------------------ |
| Integration connection | Stores and validates credentials at the organization level.                                |
| Action catalog         | Defines the provider action, inputs, and label.                                            |
| Agent function         | Exposes a selected action to the model with a clear description.                           |
| Prompt                 | Explains when the agent should use the action and what consent or information is required. |
| Call History           | Shows conversation evidence around whether the action should have run.                     |
| Invocation evidence    | Shows whether the provider accepted the action and which fields were missing or failed.    |

## When To Use An Integration During A Call

| Scenario                                                        | Good agent action?                                          | Reason                                                                         |
| --------------------------------------------------------------- | ----------------------------------------------------------- | ------------------------------------------------------------------------------ |
| Caller asks for a WhatsApp confirmation and the number is known | Yes, Wati can send a message if the function is configured. | The caller intent is clear and the required field exists.                      |
| Caller asks for an email summary and the email is confirmed     | Yes, Resend can send a follow-up email.                     | The action matches the live request.                                           |
| Workflow should follow up after every completed call            | Usually no, use a workflow instead.                         | Post-call automation is more reliable outside live conversation.               |
| Action depends on slow external data                            | Be careful.                                                 | Slow actions create silence during the call unless the prompt handles waiting. |

## Add An Integration Action To An Agent

<Steps>
  <Step title="Connect the integration">
    Set credentials and validate the connection first.
  </Step>

  <Step title="Choose the action">
    Pick the action that matches the caller-facing outcome, such as a Wati message or Resend email.
  </Step>

  <Step title="Write a strict function description">
    Tell the model when to use the action, what values it needs, and when not to call it.
  </Step>

  <Step title="Collect required values">
    Use dynamic variables if the data is known before the call, or prompt the agent to ask the caller when the data is missing.
  </Step>

  <Step title="Test with caller phrases">
    Test direct requests, vague requests, no-consent cases, and missing-field cases.
  </Step>

  <Step title="Review call and invocation evidence">
    Confirm transcript, action timing, redacted request, provider response, and failure path.
  </Step>
</Steps>

## Agent Integration Risks

<AccordionGroup>
  <Accordion title="Action runs too early">
    Tighten the function description and prompt. Require explicit caller intent or required data before the model can call the action.
  </Accordion>

  <Accordion title="Action never runs">
    Make the trigger condition clearer and test with direct caller language that should call the action.
  </Accordion>

  <Accordion title="Missing field error">
    Add dynamic variables, collect the value during the call, or move the action into a workflow after post-call extraction.
  </Accordion>

  <Accordion title="Credential expired">
    Validate the integration connection again before editing the agent prompt.
  </Accordion>

  <Accordion title="Live call goes silent">
    Check integration latency. If the action does not need to happen before the agent replies, move it to a workflow.
  </Accordion>
</AccordionGroup>

## Verify The Agent Integration

Test a successful action, missing field, ambiguous caller request, and provider failure. Confirm the action runs only when intended and that the agent communicates the result accurately.

## Related Reading

<CardGroup cols={2}>
  <Card title="Agent Functions" icon="settings" href="/docs/agent-settings/functions">
    Expose actions to agents.
  </Card>

  <Card title="Connecting Integrations" icon="plug" href="/docs/dashboard-integrations/connecting-integrations">
    Set up credentials.
  </Card>

  <Card title="Integration Functions" icon="activity" href="/docs/dashboard-integrations/integration-functions">
    Review invocation evidence.
  </Card>

  <Card title="Call Detail Page" icon="panel-right" href="/docs/monitoring/call-detail-page">
    Review action context.
  </Card>
</CardGroup>