AI Voice Agents

An AI that answers phone calls in a natural voice — grounded in your knowledge, connected to your tools, and ready to hand off to a human when it matters.

TL;DR

• What it is — a configurable AI persona you can route phone calls to. It listens, speaks in a natural voice, checks your knowledge base, calls your tools (availability, booking lookup, payment link), and transfers to a human department when the conversation needs one.
• Who it's for — hospitality operators who want to scale front-line voice coverage without hiring more agents, and teams that want a consistent after-hours voice presence.
• Top outcome — every routine call answered within a ring, with the complex ones cleanly handed to the right human and the whole conversation logged for review.

At a glance

How to find it

Sidebar: Settings → Calling → [your account] → AI Agents.

Direct URL: https://lodgestory.com/crm/settings/calling/{accountId}/ai-agents

This is a companion tab to the regular Agents tab on the same calling account. Human agents are on one tab; AI agents on the other.

📸 [SCREENSHOT: ai-voice-agent-landing.png — AI Agents list with voice preview and tool count per persona]

What are AI Voice Agents?

The problem they solve

The volume of phone calls a hospitality business receives is lumpy. Morning check-out and evening check-in produce spikes that no reasonable staffing plan fully covers. Nights and weekends bring a steady trickle that a human team can't profitably staff. An answering machine loses most of those callers. A generic call-centre handoff loses them to brand confusion.

AI voice agents answer the call immediately, in a voice you choose, grounded in the content you've fed them, and handle the conversations that don't need a human. "What time is check-out?" "Is the spa open on Sunday?" "Can you send me a payment link for ₹12,000?" — all answered in seconds, 24/7, without a human picking up. When the caller wants something more — "I want to change my booking dates" — the agent transfers to the right human department with the context already preserved.

What you get

• Per-persona voice selection. Each AI voice agent has its own system prompt and its own voice. A warm concierge voice for the luxury brand's number; a crisp operations voice for the support line.
• Grounded in your knowledge. Attach documents, FAQs, rate sheets, and property details. The agent cites from what you've given it rather than making things up.
• Tool-calling during the conversation. Let the AI look up a booking, check availability for specific dates, generate a payment link, or open a ticket — all during the call, with the results folded into its reply.
• Clean handoff to humans. Define the phrases or intents that trigger a transfer ("I want to speak to a human", "I want to make a complaint"), and the call routes to the right department with the transcript carried forward.
• Full conversation transcripts. Every turn — caller and agent — is stored and searchable. Use them for QA, training, and compliance.
• Multilingual. Agents can listen and speak in multiple languages; configure the language(s) per persona.
• Recordings, too. The full audio is recorded like any other Lodgestory Voice call.
• Concurrency at scale. An AI agent can field many parallel calls at once, so a volume spike doesn't create a queue.

How they're different

• Voice-specific, not a text agent reading aloud. Voice agents are tuned for the rhythm of phone conversations — turn-taking, interruptions, natural pauses, clarification prompts — rather than being a chat bot awkwardly wired to a phone.
• Grounded output by default. When you attach knowledge, the agent pulls from it before generating. This dramatically reduces hallucinations compared to a bare AI.
• Tools first-class. Tools aren't an afterthought — they're part of the persona configuration. The agent decides when to call one, you see the call in the transcript, and the result shapes the reply.
• Handoffs preserve context. Human agents picking up a transferred call see the full transcript before they say hello. No re-explanation.
• Sits alongside human agents, not instead of them. Route easy calls to AI, hard calls to humans, with the choice configurable per number and per menu key.

Customer scenarios

1. After-hours front desk. A villa operator routes their main number to an AI voice agent between 11 PM and 7 AM. The agent answers questions about check-in time, Wi-Fi, pool hours, and amenities. Booking changes transfer to the on-call manager's mobile. Morning staff see the overnight transcripts in the Calls workspace.
2. Payment collection line. A hospitality group exposes a dedicated number for payment follow-ups. The AI agent greets the caller, looks up their booking by phone number, confirms the amount owed, generates a payment link on request, and transfers to accounts if there's a dispute.
3. High-volume reservations hotline. During peak season, a property's reservation line routes first to an AI agent. It handles the 70% of callers who just want dates and rates, and transfers the 30% who want to actually book to a human sales agent.
4. Multilingual support. A resort with international guests configures two AI personas — one for English + Spanish, one for Hindi. The voice menu routes Key 1 to English/Spanish, Key 2 to Hindi.

How it fits with Lodgestory Voice

• Routing targets. Phone numbers, voice menus, and department failovers can all target an AI voice agent.
• Departments — the handoff destination. AI agents transfer to a specific department's queue when the conversation escalates.
• Voice Menus (IVR) — a menu key can route to an AI voice agent; the agent can also start a call and hand back to the menu for further routing.
• Call Recording & Voicemail — AI calls are recorded and transcribed like any other call.
• Agents — human agents sit on a sibling tab. Calls can flow between AI and human agents freely.

Core concepts

Quick Start — your first AI voice agent in 10 minutes

Step 1 — Create an AI voice agent

Settings → Calling → [your account] → AI Agents → + New. Pick a name, a voice from the library, and a language. Write a short system prompt: "You are the after-hours concierge for Acme Villas. Answer questions about check-in, amenities, and the spa. If the caller wants to change a booking or speak to a human, transfer to the Front Desk department."

📸 [SCREENSHOT: ai-voice-qs-1-create.png]

Step 2 — Attach knowledge

Upload your property handbook as a PDF, plus a short FAQ document with the most-asked questions and their answers. Save.

📸 [SCREENSHOT: ai-voice-qs-2-knowledge.png]

Step 3 — Attach tools

Pick from your organisation's tool library — for example, checkAvailability and transferToHuman. Save.

📸 [SCREENSHOT: ai-voice-qs-3-tools.png]

Step 4 — Route a phone number to the agent

Settings → Calling → Numbers. Pick a number, set routing type to AI Agent, select your new agent, save.

📸 [SCREENSHOT: ai-voice-qs-4-route.png]

Step 5 — Test the call

Dial the number from your personal phone. The AI answers in the voice you picked, responds to your question, and (if configured) transfers you to the right human department.

📸 [SCREENSHOT: ai-voice-qs-5-test.png]

Step 6 — Review the transcript

Open the Calls workspace. Your test call appears with a transcript icon. Click it to see the turn-by-turn exchange between you and the agent, including any tool calls it made along the way.

📸 [SCREENSHOT: ai-voice-qs-6-transcript.png]

How it works

1. A call arrives on a number routed to an AI voice agent.
2. The AI greets the caller in its configured voice.
3. The caller speaks. The AI listens, transcribes, and understands the intent.
4. If answering requires data, the AI calls the relevant tool (availability, booking, payment). It weaves the result into its reply.
5. If answering requires information from your documents, the AI pulls the relevant passage from your attached knowledge and cites it.
6. If the conversation hits a handoff trigger, the AI says a brief handoff message and transfers the call to the configured department. The human picking up sees the full transcript before saying hello.
7. On hangup, the call is recorded, the transcript is stored, and the call-event card appears in the caller's contact timeline.

Everything the agent says is generated in the moment — you don't script the conversation in advance. The system prompt, voice, knowledge, and tools shape it; the caller drives it.

Features in depth

Voice library

A set of preset voices across major languages and accents, each previewable in the admin UI. Pick one per agent. Enterprise customers can add a custom cloned voice for brand consistency.

Tip: preview the voice on your actual greeting text before publishing. A voice that sounds great saying "Hello, how can I help?" might pronounce your property's name oddly; hearing it in-context lets you rewrite the greeting or pick a different voice.

Knowledge library

Upload PDFs, plain text, or paste content directly. The agent retrieves relevant passages before responding. Typical contents: property handbook, FAQs, rate sheets, cancellation policies, amenity details, local recommendations.

Tips:

• Keep knowledge documents short and structured. Bullet lists, clear headings, Q&A format. The agent will paraphrase; the more structured the source, the cleaner the paraphrase.
• Update as things change. A stale rate sheet produces confidently wrong answers.
• Avoid contradictions — if two documents disagree, the agent may mix them.

Tool library

Tools are shared with the CRM's AI Agents. Define a tool once (e.g., checkAvailability(dateRange) hitting your property-management system), attach it to any AI — voice or text — that needs it. The tool library lets voice agents reach the same capabilities chat agents already have.

Built-in tools worth enabling by default:

• transferToHuman(department) — the handoff primitive. Without it, the agent can't escalate.
• lookupBookingByPhone() — starts every call with the caller's booking context if they're an existing guest.
• sendPaymentLink(amount, bookingReference) — lets the AI collect outstanding payments over voice.
• createTicket(category, description) — captures a task for humans to follow up on.

Handoff rules

You configure when the AI should transfer a call and where to transfer it. Three common triggers:

• Caller intent — "I want a human", "speak to manager", "this isn't helping". The AI detects the intent and calls the transfer tool.
• Conversation length — after N turns without resolution, escalate.
• Specific tools failing — if a booking lookup returns nothing, escalate rather than guess.

The transfer destination is a department (queue of humans), not a specific person. The queue's ring strategy picks the right human.

Transcript review

Every call an AI agent takes is stored with a full transcript. Search across every conversation — find every time the agent discussed the cancellation policy, find every caller who asked about the pool, find every handoff to a given department.

Transcripts also surface in reports so you can quantify how often the AI answered vs. transferred, and on which topics.

Multilingual

Per-agent language configuration. An agent can also detect the caller's language on the fly and switch, if you've configured multiple languages in its persona. Voice selection is per-language — the same agent can have a Spanish voice and an English voice and use the right one per turn.

Concurrent calls

AI voice agents can answer many calls at once — the cap is per plan, not per agent. A sudden spike doesn't cause callers to wait in a queue.

Session duration cap

Configure a maximum call length per agent (e.g., 15 minutes). If a call goes past the cap without a handoff or hangup, the agent politely ends it and invites the caller to call back. Prevents runaway costs on pathological calls.

Roles & permissions

Connections

• Phone Numbers — direct routing target.
• Voice Menus (IVR) — menu keys can route to an AI agent.
• Departments — the handoff destination for escalated calls.
• Agents — human-agent sibling tab on the same calling account.
• CRM AI Agents — shared tool library. Tools defined once work in both voice and chat.
• Calls workspace — where AI call transcripts show up alongside human calls.
• Call Recording & Voicemail — AI calls are recorded like any other call.

Limits a user will run into

Errors & FAQ

FAQ

• Can I have multiple AI voice agents on the same account? Yes. Common pattern: one for front-desk enquiries on the main number, one for payment collection on a dedicated line, one for after-hours only.
• Can I schedule AI agents only for certain hours? Yes — route the number through a time condition first. Office hours go to a human department; after-hours go to the AI.
• Does the AI interrupt if the caller is still speaking? No — it waits for a natural pause, the same way a human does.
• Can it handle callers who are hard to understand? Most callers work fine. Heavy background noise or very strong regional accents may cause the agent to ask for repetition more often.
• Does the AI ever say it's an AI? You control this. Some operators want full transparency ("Hi, I'm Lodgestory's AI assistant — how can I help?"); others prefer a neutral "Hi, thanks for calling Acme Villas." Configurable in the system prompt.
• What happens if a caller asks a sensitive question (legal, medical, financial advice)? Good practice is to add an explicit guardrail to the system prompt: "If asked for legal, medical, or financial advice, politely decline and offer to transfer to a human." The agent will follow the guardrail.

What's new

• Apr 2026 — Multilingual voice agents with automatic language detection.
• Mar 2026 — Shared tool library between CRM AI Agents and Voice AI Agents — define a tool once, attach anywhere.
• Jan 2026 — AI Voice Agents general availability. Custom system prompts, voice library, knowledge attachments, tool-calling, and department handoff.

Related modules & next steps

• Agents — human-agent configuration on the sibling tab.
• Voice Menus (IVR) — a common way to route some calls to AI, some to humans.
• Departments — the handoff destination.
• CRM AI Agents — the sibling module for chat AI; shares the tool library.
• Call Recording & Voicemail — where AI call audio and transcripts are stored.