Prerequisites
Before you begin, make sure you have:- A DialNexa account with access to the dashboard
- A funded wallet (phone number purchase and call minutes are billed from your wallet balance)
If your wallet balance is zero, phone number provisioning will fail and test calls will not complete. Add funds at Settings > Billing before continuing.
Create Your First Agent
Open the Agents tab
In the left sidebar, click Agents. If this is a new workspace, the list is empty.
Create a new agent
Click Create agent in the top right. Choose Single Prompt Agent to start. Single prompt agents are the fastest to configure and best for focused tasks like appointment reminders or lead qualification.Give the agent a name that describes its purpose, for example
Appointment Reminder Bot. The name is internal and visible only to your team.Configure the conversation stack
The builder opens on the Stack tab. Set the following:
Leave Speed, Stability, and Temperature at their defaults for now.
| Setting | Recommended for first agent |
|---|---|
| Language | English (or your primary language) |
| Voice Provider | ElevenLabs |
| Voice | Pick any voice from the dropdown and click the play button to preview |
| LLM | OpenAI GPT-4o Mini (fast, cost-effective for simple tasks) |
| Transcriber | Deepgram Nova 3 (multilingual default) |
Write the system prompt
Click the Prompt tab. The prompt is the agent’s full operating instruction.Paste the following starter prompt, then edit it for your use case:Notice the
{{appointment_date}} and {{appointment_time}} placeholders. These are dynamic variables you can pass per-call. For a test call you can supply fixed values.Buy a Phone Number
Open Phone Numbers
In the left sidebar, click Phone Numbers (under Calls or the top-level menu, depending on your plan).
Search for a number
Click Buy Number. Choose your country, select a number type (local or toll-free), and search. A list of available numbers appears with their monthly cost.
Assign the Agent to the Phone Number
Open the phone number settings
In the Phone Numbers list, click the number you just bought or click the row action menu and select Edit.
Set the inbound agent
Under Inbound Configuration, select your agent from the dropdown. Select the published version (the one you just published, not Draft).
Make a Test Call
Call the number
Dial the phone number from any mobile or landline. The call should connect within a few seconds and you will hear your agent’s welcome message.
Verify the agent responds correctly
Speak naturally. Confirm the agent responds according to your prompt. If you used the sample prompt above, the agent will attempt to confirm an appointment.
What to Check if the Call Does Not Work
Call goes to voicemail or dead air
Call goes to voicemail or dead air
The phone number may not be fully provisioned yet. Wait 60 seconds and try again. If the problem persists, confirm the number status in Phone Numbers shows Active.
Agent does not respond or response is wrong
Agent does not respond or response is wrong
Open the agent in the builder and confirm a published version exists. Drafts are not served. If the version is published, check that the phone number is assigned to that version, not an older one.
Call fails immediately with an error
Call fails immediately with an error
Check your wallet balance. A zero or negative balance will cause calls to fail at connection. Add funds and retry.
Dynamic variables appear literally in speech (e.g. 'double brace appointment date')
Dynamic variables appear literally in speech (e.g. 'double brace appointment date')
For test calls, enter variable values in the test call modal or set fallback defaults in Agent Settings > Fallback Settings. Variables without a value are passed as-is, which the agent will speak literally.
Next Steps
Outbound Calls
Place a single outbound call and review the result.
Types Of Agents
Choose the right agent type before configuring the stack.
Prompts And Welcome Messages
Write prompts that produce consistent, on-brand conversations.
Provider Selection Guide
Choose the right voice, LLM, and transcriber for your use case.