Create your first voice agent

For the complete draft, publish, activation, version, duplication, and deletion model, see Agent lifecycle.

This guide takes you from the Agents page to an inbound voice agent that is published, connected to a phone number, and ready to receive calls.

At the end, you will understand the difference between saving a draft, publishing a version, and starting an agent. These are separate actions in Wirevox.

Before you begin

You need:

• access to a Wirevox workspace;
• permission to create and update agents in that workspace;
• available capacity under the workspace's agent limit; and
• workspace credits if you want to run a browser voice test.

To receive real calls, the workspace must also be able to provision or manage phone numbers. The Wirevox interface currently makes phone-number provisioning available on the Starter plan and above.

note

This guide covers the inbound voice-agent flow currently available in the creation dialog. Outbound Agent is displayed as Soon and cannot be selected there.

1. Start a new agent

1. Select the workspace where the agent should be created.
2. Open Agents from the main navigation.
3. Select New Agent.

If the workspace has no agents yet, select Set up first agent instead. Both controls open the same agent-type dialog.

If Wirevox displays Plan Limit Reached, the workspace already has the maximum number of agents allowed by its current plan. Remove an unneeded agent or change the subscription before continuing.

2. Select Inbound

The Select Agent Type dialog currently contains:

Option	Availability	Purpose
Inbound	Available	Answers incoming voice calls.
Outbound Agent	Soon	Displayed but not selectable.
AI Chat Agent	Beta	Handles website chat and follows a different configuration flow.

Select Inbound.

3. Choose a template

Wirevox opens Choose a Template. Templates can be filtered by language and industry.

Select a template that is close to the agent's intended role, or select Blank to begin with a general-purpose English voice assistant. Then select Create Agent.

A voice template can prefill:

• the agent name and avatar;
• the system prompt;
• the primary voice and voice-delivery settings;
• prompt variables and their initial values; and
• recommended capabilities for that use case.

important

A template can recommend functions, but it does not create functions, connect integrations, or attach functions automatically. Those actions require separate configuration.

After creation, Wirevox opens the agent editor. A newly created inbound agent is paused and has no published version or assigned phone number.

4. Rename the agent

The template name becomes the initial agent name. To change it:

1. Select the edit icon beside the name at the top of the editor.
2. Enter a name of up to 25 characters.
3. Press Enter or select outside the field to save it.

Use a name that helps workspace members identify the agent. The caller-facing identity should also be defined clearly in the system prompt.

5. Review the system prompt

The main Prompt Editor contains the agent's editable system prompt. Replace template placeholders and verify that the instructions accurately describe the real business and intended call flow.

At minimum, the prompt should define:

• who the agent is and which business it represents;
• the primary outcome for each call;
• how the agent should speak and ask questions;
• facts it may rely on and what it must not invent;
• when it should use each attached function;
• required confirmations before consequential actions;
• escalation and fallback behavior; and
• privacy, safety, and business boundaries.

Do not leave sample facts in a production prompt. If a template contains variables such as {{business_name}}, {{business_hours}}, or {{faqs}}, give each variable an accurate value in Configuration → Variables.

The prompt toolbar can insert configured variables and attached functions. Prompt warnings identify referenced variables that have no value, functions that are not attached, and other disconnected references.

6. Check the voice configuration

Select Configuration in the editor header. For a first agent, review these sections:

General

• Pipeline Architecture selects Modular or Realtime operation.
• Timezone controls time-sensitive behavior.
• Model or Realtime Model selects the conversational model available for the chosen pipeline.

Changing between Modular and Realtime also changes which compatible model and voice configuration is active. Keep the template's default unless you understand the requirements and cost of the alternative.

Variables

Add values referenced by the system prompt. Variable names support letters, numbers, and underscores and are inserted with double braces, for example {{business_hours}}.

Language & Voice

• Primary Language determines the main language configuration.
• Primary Voice selects the voice used for speech output.
• A secondary language and voice can be configured when the selected runtime supports them.

Changing the primary language may cause Wirevox to choose a compatible default voice when the existing voice does not support that language.

Additional sections control call timing, voice delivery, audio environment, knowledge, and conversation behavior. They can remain at their template defaults for an initial test unless the use case requires a specific change.

7. Add knowledge or functions only when needed

An agent does not need a knowledge base or function to complete a basic conversational test.

• Use Configuration → Knowledge Base to select an existing workspace knowledge base.
• Use Functions in the editor header to attach existing reusable functions.
• If a template shows a suggested setup, Create functions opens the workspace Functions page and Attach functions opens the attachment panel.

Attaching a function makes it available to the runtime. It does not, by itself, teach the agent when to use it. Add explicit function rules to the system prompt and test successful, failed, and ambiguous cases.

8. Confirm the draft is saved

The inbound editor saves configuration changes automatically after a short delay. The header shows Saving..., Save failed, or the most recent saved time.

Before testing, leaving the editor, or publishing, wait until saving finishes. Selecting Test, Publish, or the back control also attempts to flush the latest pending changes first.

Saving changes updates the draft only. It does not change the published configuration used by an active agent.

9. Test the draft in the browser

1. Select Test in the agent header.
2. Allow microphone access when the browser asks.
3. Select Start Live Test.
4. Speak with the agent and review the live transcript.
5. Select End Call when finished.

The browser test uses the current draft configuration, so you can evaluate unpublished prompt, model, voice, knowledge, and function changes. It does not require an assigned phone number.

warning

Browser voice tests consume workspace credits at the agent's calculated rate. The editor header shows the current draft cost in credits per minute. A test cannot start when the workspace has no usable credits or is locked because of a failed payment.

danger

Connected functions can execute real operations during an owner browser test. For example, a calendar function may create or change a real event. Use test accounts, safe destinations, or non-destructive inputs where appropriate.

Test at least:

• the expected opening and main request;
• unclear or incomplete caller input;
• a request outside the agent's scope;
• every attached function, including a failure response; and
• the closing or escalation path.

10. Publish the draft

When the draft is ready:

1. Select Publish.
2. Optionally enter a Version Name and Version Note.
3. Select Publish Version.

Wirevox saves pending changes and validates the draft before publishing. Configuration errors must be fixed before publishing can complete. A successful publish creates the version that live conversations can use and enables version history.

Publishing does not start the agent and does not assign a phone number.

11. Assign a phone number

Open Numbers from the main navigation.

1. Acquire a number if the workspace does not already have an available one.
2. In the number inventory, find the number to use.
3. Under Assigned Agent, select the newly published agent.

The number and agent must belong to the same workspace. Wirevox will not assign a number to an agent that has never been published, and one agent cannot have two different numbers assigned at the same time.

12. Start the agent

Return to Agents or open the agent editor, then select Start or Start Agent.

Wirevox requires both of the following before an inbound voice agent can start:

• at least one published version; and
• an assigned phone number.

After the status changes to Active, calls to the assigned number can be routed to the agent. Select Stop or Stop Agent to return it to Paused without deleting its configuration or published history.

Understand the lifecycle

State or indicator	Meaning
Draft	Editable configuration used by the editor and browser testing.
Unpublished changes	The saved draft differs from the currently published version.
Published version	Immutable version selected for live operation.
Paused	The agent is not currently active, even if a version is published.
Active	The published agent is started and can handle traffic through its assigned connection.

You can continue editing while an earlier version is active. Those edits remain in the draft until you publish again.

Troubleshooting

New Agent is unavailable or shows a plan limit

The workspace has reached its plan's agent limit. Check the current plan or remove an agent that is no longer needed.

The prompt contains warnings

Open the warning list in the Prompt Editor. Add missing variable values, attach referenced functions, or remove instructions that reference unavailable capabilities.

The browser test does not start

Confirm that microphone access is allowed, the workspace has usable credits, and the workspace is not payment-locked. A browser test does not require a published version or phone number.

Publish fails

Read the configuration error shown by the publish dialog or notification. Wirevox blocks publishing when the draft fails configuration validation.

A phone number cannot be assigned

Publish the agent first. Also confirm that the number belongs to the same workspace and that the agent does not already have another number.

Start Agent is disabled or fails

Confirm that the agent has a published version and an assigned phone number. Publishing and starting are separate operations.

Next steps

• Use Voice agents as the complete production guide for inbound runtime, configuration, context, testing, delivery, monitoring, and troubleshooting.
• Read Understand the agent editor for every inbound editor surface and lifecycle control.
• Learn how to write an effective system prompt before expanding the agent's role or attaching consequential actions.
• Learn the platform vocabulary in Core terminology.
• Review the relationship between drafts and live operation in How Wirevox is organized.