I want to make outbound call and if agent have some information about user than that would be great.

To pass input variables like

first_name

to a Vapi Workflow for outbound calls, include them in the

customer

object when creating the call:

json
{
  "phoneNumberId": "YOUR_PHONE_NUMBER_ID",
  "assistantId": "YOUR_ASSISTANT_ID",
  "customer": {
    "first_name": "John"
  }
}

In your workflow prompts, use

{{first_name}}

to reference this variable and personalize the conversation. Tip: Make sure your workflow nodes use the exact variable name in the

{{variable_name}}

format for dynamic insertion. [See more in the docs](https://docs.vapi.ai/assistants/personalization) Sources: - [Personalization with user information](https://docs.vapi.ai/assistants/personalization) - [Workflows documentation](https://docs.vapi.ai/workflows) - [Outbound sales example](https://docs.vapi.ai/examples/outbound-sales)

@Vapi Support Bot I'm using websocket connection with Vapi's workflow

To pass input variables like

first_name

to a Vapi Workflow over websocket, use the

assistantOverrides

parameter with

variableValues

in your

.start()

call:

js
const assistantOverrides = {
  variableValues: { first_name: "John" }
};
vapi.start("YOUR_WORKFLOW_ID", assistantOverrides);

In your workflow, reference

{{first_name}}

to personalize the conversation. Tip: Make sure your workflow nodes use the exact variable name in the

{{variable_name}}

format for dynamic insertion. See [Web SDK docs](https://docs.vapi.ai/sdk/web#overriding-assistant-configurations) for more details. Sources: - [Web SDK documentation](https://docs.vapi.ai/sdk/web#overriding-assistant-configurations) - [Personalization with user information](https://docs.vapi.ai/assistants/personalization)

Integrate your CRM to fetch data, personalize assistant prompts with that info, automate call triggers via tools like Make.com, and ensure real-time data is available during the call..

One more -- there is such a high latency in Workflow than Assistant. Is there any way we can reduce them ? (I'm building voice agent for hindi support)

Use streaming STT/TTS, semantic VAD, faster or smaller LLMs like GPT-4o mini, cached responses, and host services regionally, while avoiding HTTP-based delays by switching to WebSockets where possible..