Luna · Voice and messaging agent

It answers at 11 at night.

Luna is the navamkt agent. It answers WhatsApp and the phone in your business voice, handles the usual questions, books appointments and writes the lead into your CRM. Every reply carries its cost in plain sight.

You get me, not a queue · Reply <24 h · No sales deck

Luna · WhatsAppONLINE · AFTER HOURS

cuánto cuesta

Los paquetes de IA van de $6,000 a $13,000 MXN al mes. La instalación inicial cuesta $25,000 o $50,000 MXN según el paquete. Si contrata por año, recibe un 20 % de descuento.

$0.008 MXN · 1.1 s

oiga, ¿esto es un robot?

Sí, soy un asistente con inteligencia artificial de navamkt. Si prefiere hablar con una persona se lo paso ahorita mismo.

sin costo · la contestó sola · 9 ms
8 replies · 3 free$0.04 MXN

Real conversation, in Spanish as it happened · measured September 6, 2026

01

Why it exists

01

The Friday 9pm message that got answered on Monday.

02

The call that came in while you were with a customer.

03

The "how much is it?" nobody answered — already asked somewhere else.

Luna does not fix all of that. It answers the part that can be answered with what you already know about your business, at any hour, and flags anything outside that.

02

What it does

L·01

Answers WhatsApp

Through Meta's official API or whatever bridge you already run. Reads the message from any of the 13 providers it ships with, no setup.

L·02

Answers the phone

Inbound and outbound calls with neural voice. The same head that handles messages, on another channel.

L·03

Books appointments

Offers open slots, confirms and puts the appointment in. If the calendar says there is nothing, it does not invent one.

L·04

Writes the lead into your CRM

Reads the fields of the system you already use and proposes the mapping itself. Whatever it cannot map confidently goes to review, not to a guess.

L·05

Clones

The agent is a text file: name, voice, prices, hours, objections and what it must never say. Another file, another business.

03

What it costs

Where each message went

«what time do you close on Fridays»

asking about hours · 3316 µs

Answers it herself

«hey, where are you located?»

asking for the address · 290 µs

Answers it herself

«what services do you offer»

asking what you do · 206 µs

Cheap model

«that's expensive for what it is»

asking about price · 264 µs

Reasoning model

«my competitor uses another agency, why should I believe you?»

not confident · 311 µs

Reasoning model

Before touching the network, a 2 MB model reads the message and decides who answers it. It handled 2 of 5 on its own, in microseconds. And of 4 different ways of asking the same thing, 3 reused the answer it had already given.

The conversation above

$0.042

MXN · 8 replies, 3 of them free

Model usage goes on your own account, at cost. I do not resell or mark it up: you pay what it spends, and you can see it in your provider dashboard.

04

What it will not say

This will bring you about forty more patients a month.

✕ BLOCKEDpromises results · caught by the model

I guarantee you results from the first week.

✕ BLOCKEDpromises results · caught by a written rule

I guarantee it will be installed within four weeks.

✓ PASSES

The word "guarantee" is in both of the last two. What changes is what is being guaranteed: the work can be guaranteed, a number of customers cannot. A banned-words list cannot tell those apart — which is why there is a model trained to.

What it says instead

I'll be honest with you: I can't promise you a number of results. What I can guarantee is the work — what gets installed and how long it takes. Shall we go over it calmly?

You write the list of what is off limits. In a clinic, for instance, the first thing blocked is diagnosing.

05

How well it understands

92.6%

What the person wants

45 intents · 4,127 examples

100.0%

Whether what she is about to say is allowed

6 classes · 2,805 examples

98.9%

Which field of an unknown CRM is which

20 classes · 5,318 examples

93.2%

The same, for a dental practice

31 intents · 3,510 examples

«I already work with another agency»

17% confidence

When it is not confident it understood, it does not guess: it escalates to a person. Confidence is not there to brag, it is there to brake.

Accuracy is measured against phrases the model never saw in training, in the run of September 6, 2026. It is not a brochure number: it gets re-measured on every retrain.

06

It connects to what you already have

WhatsApp, wherever it comes from
Meta oficialTwilioEvolutionUltraMsgand 9 more

Each provider sends the message in a different shape. Luna reads all four with nothing configured, and for one it does not know there is a generic mode.

And your CRM's 10 fields
mailemail
funnel_stageetapa
cita_programadafecha_cita
pat_full_namenombre_completo

A model reads the fields of the system you already use and proposes the mapping. The one it is not confident about —clinic_branch_id, at 67%— goes to review instead of being invented.

07

What you need to install it

A business WhatsApp

Meta's official API, or the bridge you already use. If you have neither, it gets set up.

Your business information

Prices, hours, services, the usual questions and — most importantly — what must never be said. This is the part that takes time.

A model account

Usage goes on your account at cost. I do not resell or mark it up: you pay what it spends.

A decision about your data

Local models on your server, or a zero-retention API. Decided before install, not after.

08

Questions

Is this one of those button-menu chatbots?

No. There are no menus or "press 1". Luna reads what the customer wrote, works out what they are asking and answers with your business information. When she is not confident about the question, she does not guess: she hands it to you.

Does she get things wrong?

Yes, like any system. That is why she has two brakes: a model that measures how confident she is that she understood — below a threshold it escalates to a person — and a review of what she is about to say before saying it, which blocks whatever is off limits. Both are auditable, and the numbers are measured against examples the model never saw in training.

Does my customer data leave my infrastructure?

It depends on the setup. A good part of what Luna does runs on local models that never touch the internet. For the rest you can use your own zero-retention API account, or open models running on your server. That gets decided before install, not after.

What does it cost?

Installation is quoted based on what has to be connected: WhatsApp, phone, calendar, CRM. Model usage goes on your own account at cost — I do not resell it. In the run on this page, a full eight-reply conversation cost four Mexican cents.

Can I see it working first?

Yes. Message me on WhatsApp and I will show you the thing running, with the numbers on screen exactly as you see them here. It is not a sales deck: it is the system.

How long does it take to install?

It depends on how organized your business information is — prices, hours, services, what must never be said. That part takes the time, not the software. I will give you the range once I see your case.

I am not asking you to take my word.

Message me and I will show you Luna running, with the numbers on screen exactly like here. If it fits, we look at what it would take in your business.

Reply <24 h · No lock-in · Model usage goes on your account, at cost