AI voice agents

AI voice agents are automated agents that can handle inbound calls autonomously, assist human agents during live calls with real-time coaching, and perform actions like looking up orders or checking account status. They integrate directly into the UltraCart Calls system alongside human agents.

Overview

AI voice agents serve two primary roles in UltraCart Calls:

• Autonomous call handling -- an AI agent answers an inbound call, has a natural language conversation with the caller, collects information, performs actions, and routes the call to a human agent or queue when needed.

• Real-time coaching -- an AI agent listens to a live call between a human agent and a caller, then provides whisper suggestions that only the human agent can see.

AI agents can be assigned to queues just like human agents. They can answer calls as a first responder, serve as a backup when no human agents are available, or coach human agents in real time. Every AI interaction is tracked with session data, tool call logs, and cost breakdowns in the call record.

AI agent capabilities

Autonomous inbound handling

When assigned to a queue, an AI agent can answer inbound calls and carry on a natural conversation with the caller. The AI agent can:

• Greet callers and collect information (name, order number, reason for calling)

• Look up orders and subscription details

• Check customer account status

• Perform actions like pausing, resuming, or canceling subscriptions

• Generate coupons

• Create support tickets

• Transfer the call to a queue, specific agent, extension, or IVR menu when it can't resolve the issue

Real-time coaching

When configured as an AI coach for a queue, the AI agent listens to the conversation between the human agent and the caller. It provides real-time whisper suggestions that appear in a coaching feed panel on the agent's screen. The caller never hears the AI coach.

Whisper suggestions have priority levels that affect their visual styling:

• Info -- general information and context

• Suggestion -- recommended actions or talking points

• Warning -- alerts about potential issues or sensitive topics

Configuring an AI agent

AI agent configuration requires Admin permissions (either Calls Admin or Chat Admin).

Creating an AI agent

AI agents are created through the AI Assistants section of the UltraCart CRM, not the Calls settings. Navigate to AI Agents > Agents to create and configure AI agents.

General settings

• Display Name -- the name the AI introduces itself as during calls (3-100 characters)

• Profile Image -- an avatar image displayed in webchat interactions

Instructions

AI agents have five instruction tabs that control their behavior across different channels:

AI voice personality

Under Voice Instructions, select the AI agent's speaking voice:

• Ara (default)

• Rex

• Sal

• Eve

• Leo

The voice personality determines how the AI sounds during phone calls.

Agent capabilities

Toggle individual capabilities on or off to control what the AI agent can do:

Orders and subscriptions:

• Look up order information

• Look up subscription information

• Update subscription credit card

• Pause, resume, cancel, or delay subscriptions

Live agent and ticket creation:

• Transfer chat to a live agent

• Create support tickets (via email, UltraCart Task, or Zoho Desk)

Item data and coupons:

• Access storefront and item data

• Generate coupons

Knowledge base

Upload documents (.txt, .md, .pdf) to give the AI agent additional context. The AI uses these documents to provide more accurate and personalized responses. Knowledge base documents can be downloaded or deleted from the agent's configuration page.

MCP servers

Connect external tool servers using the Model Context Protocol (MCP). Each MCP server provides additional capabilities the AI can use during conversations. Configure:

• Server URL

• Authentication (basic auth or header-based)

• Priority (determines the order in which servers are consulted)

Each configured server shows a live status indicator (available or unavailable).

Adding AI agents to queues

AI agents are assigned to queues the same way as human agents:

1. Navigate to Calls > Settings > Queues.

2. Edit a queue.

3. Under Queue Members, toggle on the AI agent.

4. Configure the AI priority and timeout (see below).

5. Select Save.

AI priority settings

When a queue contains both human and AI agents, the AI Priority setting controls when the AI answers:

AI timeout

When priority is set to First or Backup, the AI timeout defines how many seconds the AI agent has to handle the call before automatically routing to a human agent. Set a timeout that gives the AI enough time to greet the caller and attempt resolution, but doesn't leave callers waiting too long if the AI can't help.

AI coaching during calls

Queue-level coaching setup

To enable AI coaching for all calls in a queue:

1. Navigate to Calls > Settings > Queues.

2. Edit the queue.

3. Under AI Coach Agent, select an AI agent from the dropdown.

4. Select Save.

All calls taken from this queue now have AI coaching available.

Using AI coaching during a call

1. During an active call, select AI Coach from the softphone controls.

2. If multiple AI agents are available, select one from the dropdown.

3. Select Start Coaching.

4. The AI coaching feed appears, showing "Listening to conversation..." while the AI begins analyzing the call.

5. As the conversation progresses, whisper suggestions appear in the feed. Newer messages appear at the top.

6. Select Stop to end the coaching session.

Whisper suggestions are rendered as formatted text and can include bullet points, bold text, and other formatting. Each suggestion is styled by priority level for quick scanning.

AI agent transfers

During autonomous call handling, an AI agent can transfer the caller to:

• A specific queue

• A specific human agent

• An extension

• An IVR menu

Transfers happen seamlessly -- the caller is connected to the new destination just as they would be during a human-initiated transfer.

AI engagements in call records

Every AI interaction is tracked in the call record and visible in the call history detail view.

Engagement data

Each AI engagement records:

• AI agent name and identifier

• Engagement type (primary for autonomous handling, coaching for whisper coaching)

• Start and end timestamps

• Tool call history -- every action the AI took during the call, including tool name, parameters, result, duration, and success/failure status

• Whisper history -- for coaching engagements, every suggestion the AI provided to the human agent

Cost tracking

AI engagements include a cost breakdown:

• Billed minutes and cost per minute

• Per-engagement cost

• Total AI cost aggregated across all engagements for the call

AI costs appear alongside call and transcription costs in the call record's financial section (visible to admins only).

Multiple AI agents per call

A single call can involve multiple AI agents. For example, an AI receptionist might answer and greet the caller, then transfer to a support queue where a different AI agent provides coaching to the human agent. Each AI agent's engagement is tracked separately in the call record.

Related pages

• Call queues -- assign AI agents and configure coaching per queue

• Transfers and conferencing -- AI coaching feed during calls

• Call history -- review AI engagements, tool calls, and costs

• Permissions and roles -- admin permissions required for AI configuration

• Agent management -- AI agents in the agent list