Vozis Chat is an AI agent that talks with your customers on WhatsApp. It is not a menu-based WhatsApp chatbot, the kind that asks people to type 1, 2 or 3 and gets stuck the moment someone writes something unexpected. It understands what the customer wrote or said, replies naturally and steers the conversation toward a goal: answering a question, booking an appointment, qualifying a lead, sending a proposal or passing the contact to a sales rep.

The idea is simple. The first layer of conversations, which today takes up a large share of your team's time, is handled by AI, with quality and at any hour. Your team steps in where it makes a difference: to negotiate, close and take care of the cases that call for a person.

As with everything at Vozis, you don't get a tool to set up on your own. Our team builds the agent, writes the instructions, connects the official number, integrates your systems, tests everything and tracks the results. You follow it all in the platform.

What separates good WhatsApp service from bad comes down to details. A real person replies in short messages, one idea at a time, ends with a question to keep the conversation moving, notices when the customer sent three messages in a row and answers the right one. They also wait for the customer to finish typing before replying, and show up as "typing…" while they think.

Vozis Chat was built to talk that way. Each of these behaviors is an agent setting, tuned by the Vozis team to fit your audience. The result is a conversation the customer experiences as real service, not as an automated form.

Every conversation follows the same path. Knowing this path shows you where each setting makes a difference:

1. The contact comes in. The customer messages your company's number, taps an ad or receives a message template sent by a campaign or by the cadence flow.
2. The agent reads, listens and understands. Text, voice note, photo or PDF. It waits for the customer to finish typing before it replies.
3. The conversation happens. The agent follows its instructions, draws on the knowledge base and the company's tools, sends materials and replies by text or by voice.
4. The data is captured. Company, budget, interest, Tax ID or any detail the operation needs.
5. The conversation is classified. Qualified lead, appointment confirmed, price objection, not interested or whatever makes sense for your business.
6. The next step fires. The conversation goes to a sales rep, the team gets notified, the CRM is updated or the cadence flow moves on.

Every chat agent starts with its identity: the role (the internal name, such as "Customer Service Agent") and the display name, which is what it calls itself in the conversation. "Hi, this is Sophia from Aurora Clinic."

Right below sits the agent's intelligence: the provider and the language model that understand and write the messages, and the maximum length of each reply. On WhatsApp, short replies work best, so this limit is usually kept low. The Vozis team picks the model for each operation, balancing quality and speed.

Many people would rather send a voice note than type. Service that ignores voice notes misses a large share of conversations. Vozis Chat listens to the customer's voice notes and can also reply by voice, using the same human-sounding voice as the Vozis Voice agents.

Listening

When the customer sends a voice note, the agent transcribes it before replying. The transcript goes into the conversation history, so the agent understands the voice note in the context of everything already said, and your team can read what was said without having to listen.

Speaking

With voice replies turned on, the team decides when the agent uses voice. The most common mode is mirror: if the customer sent a voice note, the agent replies with one; if they typed, it replies in writing. It can also always reply by voice or always by text. The voice is chosen from the Vozis female and male voices and has the same settings as on calls: model, speaking speed, stability and clarity.

Quoting the customer's message

When the customer sends several messages in a row, the agent can reply by quoting the message it is answering, just as we do on WhatsApp when we tap "Reply". If the customer sent only one message, the reply never quotes. If they sent several, the agent quotes only when the reply addresses one of them in particular, for example when two different questions came in.

Interactive buttons

For questions with only a few options, the agent can offer buttons for the customer to tap instead of typing. Up to three short options become buttons; four to ten options become a list to pick from. This speeds up questions like "What time works best: morning, afternoon or evening?" and cuts down on typos. In the agent's instructions, you only need to say which questions have options; the technical format is built automatically.

On WhatsApp, people break what they want to say into several messages. If the agent replies right after the first one, it cuts the customer off mid-thought and answers only half the question. That is why there is a wait before reading: the agent waits a few seconds (from 5 to 60) with no new messages before reading everything and replying at once.

Simulated human typing completes the effect. Before sending, the agent shows as "typing…" for a time proportional to the length of the reply, up to 12 seconds. The reply no longer arrives instantly with a machine feel; it takes on the rhythm of a person on the other end.

The instructions are the text that teaches the agent to be part of your company: who it is, how it talks, what the goal of each conversation is and what to do in each situation. It is the equivalent of training a new customer service rep, only written out in great detail. A good set of instructions often runs past 10,000 characters and is organized into blocks:

• Identity and tone: who the agent is, who it works for and the style of the conversation (warm, consultative, direct).
• Message format: rules like "one idea per message", "no more than three short sentences" and "almost always end with a question".
• Goal: what counts as success, such as booking an appointment or qualifying the lead for the sales rep.
• Right and wrong examples: replies the company wants and replies to avoid.
• Language: the agent detects the customer's language and replies in it.

These instructions are written by the Vozis team based on real conversations from your company and refined with every results cycle.

While it chats, the agent relies on tools: the connections to your company's systems. With them, it stops merely answering and starts getting things done: it checks open slots on the calendar and books the appointment, looks up an order status, pulls an invoice copy, finds an available property in your system or logs the lead straight into the CRM.

The Vozis team builds these integrations to fit, with CRMs like HubSpot, Salesforce, Pipedrive and Zoho, calendars like Google Calendar, ERPs and your company's own systems. The more the agent can do on its own, the less your team needs to step in.

A WhatsApp conversation is more than text. The customer sends a photo of a document, a screenshot, a PDF with a quote. Vozis Chat reads these files and uses their content in its reply: it understands what is in the screenshot, checks the details on a document, summarizes a PDF.

In the other direction, the agent sends materials from the company's knowledge base at the right moment: the menu, the price list, photos of a property, the contract, the sales deck. Each material has a description that tells the agent when to send it, for example "send when the customer asks for photos of the apartment".

The knowledge base is the same one used by the Vozis Voice agents, organized into groups (like "Catalog", "General" or "Menu") and linked to each agent. So what was agreed on a call can become a PDF on WhatsApp seconds later.

Every conversation gets a classification, created for your operation, with a name, a color and a description that tells the agent when to use it. At a clinic, for example: Appointment booked, Just looking for information, Price objection, International patient, Not interested and Conversation in progress.

In chat, a classification can also trigger an action the instant it is applied:

• Turn off AI: hands the conversation to a human. The AI stops replying to that contact until someone hands control back.
• Stop cadence: cancels the next steps scheduled for the contact, such as follow-ups.
• Qualified lead: the conversation counts as qualified in the results.
• Add to do-not-contact list: blocks calls, WhatsApp and RCS for that number and ends the cadence.

While it chats, the agent fills in the fields your operation needs with what the customer shared. Each field has a type, which helps the AI record the information the right way:

• Text: names, addresses and open-ended answers.
• Number: amounts, quantities and money.
• Date: dates in month, day and year format.
• Yes or no: straight answers.
• List: one option from the ones you define.
• Tax ID: a tax identification number, validated by check digit.

The values appear in each conversation's summary and flow into the CRM, the webhook and the email notifications. Your team gets the lead with the information already organized, without having to read the whole conversation.

Actions alert the right people and systems when a conversation reaches an important classification. They fire when the classification changes to one of the ones you selected:

• Custom webhook: sends the conversation and contact data to your system (CRM, ERP or other), with automatic retries if it fails.
• WhatsApp group alert: posts a message in a team group, for example "new qualified lead", with the conversation summary.
• Email notification: sends an email to the chosen address with the collected data.

Next to the settings sits a test window. The team chats with the agent as if they were the customer, by text or by voice note, and sees right away how it replies. Every change shows up in the test in real time, before it reaches any customer.

Just like with Vozis Voice, every change stays in draft until it is approved and published as a new version. The version history lets you compare what changed and roll back safely.

Vozis Chat uses official WhatsApp numbers, connected through Meta. On the channels screen, the team sees each connected number, the agent linked to it, the channel quality and how many contacts were reached in the last 24 hours.

This monitoring matters because Meta sets a daily limit on new contacts for each number, based on its reputation. Relevant, well-written messages sent to people who want to receive them keep quality high and the limit growing. The Vozis team tracks these numbers so the operation is never interrupted.

On the official WhatsApp Business API, when the company starts the conversation (or when the customer hasn't replied in more than 24 hours), the first message has to be a Meta-approved template. That is how Vozis Chat welcomes a new lead, confirms a meeting or re-engages a past customer.

Each template has a category (marketing or utility), a language and variables like {{first_name}} that personalize the message. It can also have an image or video header, a footer and buttons. Once approved, the template can be used in campaigns, in the cadence flow and by the team itself from the inbox.

As soon as the customer replies, the conversation opens up: the agent chats normally, with no templates needed.

Every WhatsApp conversation lives on a single screen. On one side, the contact list, with each contact's last message, date and classification. On the other, the open conversation, exactly as the customer sees it, with text, voice notes, photos and documents, and a marker showing which messages the AI wrote.

The list can be filtered by the agent handling the conversation, by the conversations assigned to you and by classification, such as "Qualified lead" or "Not interested". This helps everyone on the team quickly find what is theirs and what needs attention. Closed conversations move to the archive, keeping the list clean.

Vozis Chat doesn't replace your team; it organizes your team's work. At any moment, someone can click Take over and pick up from there, with the full history on screen. The AI stops replying to that contact until the conversation is handed back to it. That person can also change the classification the AI assigned, if they disagree.

The handoff also happens without anyone clicking. Through its instructions and classifications, the agent knows when to transfer: when the customer asks to talk to a person, when the negotiation reaches a point that calls for a sales rep or when the topic belongs to another department. And it transfers to the right person. A billing issue goes to finance, a technical question goes to support, a lead ready to buy goes to sales.

That way, the whole company can work from the same WhatsApp number, with each department handling what is theirs, and the customer never feels stuck talking to a bot that goes nowhere.

In operations where several people receive contacts, like a real estate agency with many agents, the question is always the same: who gets the next lead? Vozis distributes leads automatically, the way your operation prefers:

• Round robin: each new lead goes to the next person in line, and everyone gets the same number.
• Set order: the company defines the sequence and the priorities.
• By rules: by region, product, language or any criteria your operation uses.

That way, no lead sits waiting for someone to pick it up, and the distribution stays fair and balanced across the team.

No one needs to read an entire conversation to know where it stands. The Summary button opens a panel with the essentials: when the conversation started, how many messages were exchanged, the current classification with an explanation written by the AI, the collected data (such as company, budget and interest) and the classification history, showing who changed what and when.

This summary is what lets a sales rep take over the lead and carry on the conversation as if they had followed it from the start.

Vozis Chat is also a step in the cadence flow, the path that combines calls, WhatsApp, RCS, waits and conditions. A common example: the voice agent calls; if the contact doesn't pick up, they receive a WhatsApp message with a welcome template; if they reply, the chat agent carries on the conversation; if they don't reply within an hour, they get another call.

Voice and chat share the same knowledge base and the same contact history. A customer who started on a call continues on WhatsApp without repeating a thing.

• 24/7 customer service: every message gets an instant reply, including at night and on weekends.
• Lead qualification: people coming from ads are answered within seconds and reach the sales rep already qualified.
• Scheduling: appointments, visits and meetings booked straight into the calendar, with confirmation.
• Prospecting: message templates start the conversation and the agent guides it to a meeting.
• Collections and payment plans: reminders, invoice copies and negotiation, always courteous.
• Support and post-sale: questions resolved with the knowledge base and, when needed, passed to the right department.

Healthcare, real estate and construction, and financial and collections companies are the heaviest users, but any company that serves and sells on WhatsApp can benefit from AI customer service on WhatsApp.