Replies built from your own business information

What you set up answers first

Each message goes through a fixed order: a stop request, your flows, a request for a person, your keyword automations, menu words, greetings and thanks, ready answers, and only then the AI. What you built yourself always answers before the AI.

• A stop or «no marketing» request comes before everything else
• Your trade pack's flows answer before a feature's own keywords
• A flow's «default reply» comes after the ready answers and before the AI

Hello, thanks and «ok» answered at once

A greeting, a thank-you or a short «ok» gets an instant reply without AI: a newcomer's hello gets your welcome and menu, a thank-you gets two next steps, and an «ok» right after a form link points back to the form.

• Only when the whole message is a greeting or thanks: a hello with a question is answered as a question
• 👍 👌 🙏 count as «ok»
• An «ok» never gets a menu it has just seen

Ready answers for your trade

For 152 trades, the questions customers of that trade usually ask come with ready answers in three languages, switched on with your trade pack.

• 2,756 ready entries across 152 trades
• Four kinds: a fact built from your own data, a general answer true for any business of the trade, a short clarifying question with a form or team button, and your own answer to the trade's question
• No number is ever written into a template: numbers come only from your data

A general answer until you write your own

For trade questions only you can answer, a general answer that is true for any business («it depends on …; tell us this and our team confirms») goes out until you write your own. Your answer always wins.

• 27 general topics for 148 trades
• Steps back when your own information already covers that topic

Your own FAQ, answered instantly

Every FAQ question and answer you write is available to the AI at once and used word for word. Mark it instant and it is sent without AI, in your own words, for other ways of asking it too.

• A customer of another language gets it translated just before sending
• Each FAQ: «instant» or «AI only»
• Answers added by the smart setup are labelled as such

Two questions in one message

A message like «hours and prices?» gets both answers separately. If not every part has a ready answer, the whole message goes to the AI or your team; no message is half answered.

• At most two ready answers per message
• Long messages (a story, several questions) are left to the AI

Understands meaning, not just words

When the words do not match, the meaning of the message is compared with the ready questions: «till when are you open today» is «opening hours». It handles slang, typos, Finglish and Gulf Arabic, on this computer, without internet.

• A small meaning model runs on this computer
• The panel shows how long it takes per message
• On by default; switch it off and only words are compared
• Also used to find the right piece of your business information
• Not on computers weaker than normal: there, words only

Example hours are never told to customers

The app's example opening hours are not used in answers or given to the AI until you confirm or change them, with a «these are our real hours» button.

What is still missing, listed

Each ready answer shows the languages it is ready in and what information the others need, with a direct link to where you fill it in.

• Needs listed: hours, address, map link, verified contacts, phone, WhatsApp, Telegram, email, website, payment methods, real prices, your own FAQ answer

Try a question

Type a question and see what would be sent at once, in how many milliseconds, and whether it was found by meaning. Nothing is sent to anyone.

10 editable default messages

System direct messages are ready; each can be switched on or off, has its own text per language, and a repeat interval per person.

• Welcome
• Please wait
• Default reply
• After hours
• Handoff to the team
• Back to assistant (off by default)
• Photo or voice received
• Story mention
• Story reply
• Error
• Repeat interval 0-720 hours per message (e.g. default reply once a day, after hours every 12 hours, story mention once a week)
• Up to 900 characters per text
• Every send logged
• A language without your own text gets the built-in text in that language, never the Persian one

Business hours with several ranges a day

Set one or more time ranges for each weekday in your time zone. After hours a special message tells the next opening time («tomorrow at 9»).

• Starts from your country's usual working days, 09:00-21:00
• The page shows whether you are open now and the next opening time

No wrong ready answer

Matching is strict on purpose: when two topics are close, or your information is missing, no ready answer goes and the message continues as usual.

• Never another topic in place of the one asked

«One moment» only when needed

With the local engine the «one moment» message goes at once; with a cloud AI only when the reply takes over 8 seconds. Not repeated if a message went in the last 5 minutes.

• Shows typing while the reply is written

Typos, Finglish and the wrong keyboard

Keywords are found even with a typo, a suffix, Finglish («gheymat») or typed on an English keyboard («rdlj» for «قیمت»), and each match says how it was found.

• Steps: exact, with a Persian ending, wrong keyboard layout, a phrase with one extra word in between, Finglish, typo
• Spaces and half-spaces ignored: «ثبتنام» = «ثبت نام»
• Arabic articles and endings, English endings
• Emoji keywords
• Match where: anywhere, at the start, whole message
• Modes: smart, exact, any message

«I don't want prices» is not a price question

Not wanting or not needing is recognised in Persian, Arabic and English, while «I don't have one» or «I don't know the price» still count as a request.

• Exclude words can veto a whole message

Sensitive and emergency triage

35 ready message intents

The app recognises what a message is about: price, booking, order status, shipping, payment, complaint and more, in Persian, Finglish, Arabic including Gulf dialect, Arabizi and English.

• Price, booking, status, shipping, payment, talk to a person, contact, order, occasion, return/refund, availability, urgent, renewal, trial, documents, cancel/reschedule, wholesale, size/fit, consultation, complaint, location/hours, portfolio, sell to us, gift, kids/family, authenticity, discount, register, greeting, thanks, stop, start, yes, received, no
• 2,019 built-in words
• The longest match wins: «shipping cost» is shipping, not price
• Guard phrases: «precious stone» is not a price question

Tune intents yourself

Add or remove words for each intent, rename it, switch it off or reset it; changes apply at once to replies and automations. A test box shows a message's intents.

• Up to 200 of your own words per intent and script
• Your edits come back if a module is switched off and on again

Synonym lexicon

Synonym groups, colloquial forms and Finglish let one keyword catch all its forms. Groups are editable and also used to search your business information.

• 7 built-in groups, 259 terms: price, sign-up, consultation, form, discount, payment, online
• Persian, colloquial, Finglish, Arabic, Arabizi and English terms
• Modules add their own groups
• Built-in: 7 groups, 259 terms
• Modules add their own trade words
• Built-in words in Persian, colloquial Persian, Finglish, Arabic, Arabizi and English
• Each word can belong to only one group; up to 80 words per group
• A test box shows which group catches a sentence

Look-alike words are not confused

Typos are accepted only in long words, real look-alike words are never treated as one, and an English message is never read as Finglish.

• Typos from 6 letters in Persian/Arabic, 7 in Finglish, 8 in English, never in Arabizi

Your trade's own words

The installed trade pack and the enabled modules add their own words and intents to message recognition.

Every reply cites its source

The AI states only facts found in a piece of your information and records which pieces it used. A reply with a number and no source is flagged.

Finding the right piece by words and meaning

For each message the related pieces are found by words, synonyms and meaning together: «do you come to the house?» finds the home-service piece with no shared word.

• Up to 5 pieces per message
• Without the meaning engine: words only

A short price list always in view

Services and products with prices are always in the local model's instructions, however short or long your text; the rest of a big catalogue comes with each message.

• Up to 12 items in the short list

Content in three languages, prices in the customer's

Write each service's name and description, the welcome and your business text separately for Persian, Arabic and English; prices are written with the currency and digits of each language.

Write the assistant's instructions yourself

You can write the assistant's role, tone and sales rules yourself or go back to the default; without it the trade pack's text or the app's neutral text is used.

• Up to 3,000 characters
• Every truth rule of the app still applies on top of it
• Your trade pack's rules for your country are added to the instructions

Conversation memory

The AI knows the last 12 messages and the details the customer gave (like name and city), which prefill the form. With a cloud AI a short summary of the chat is also refreshed every 6 turns.

• Enabled modules add their own fields
• An email address or phone number in a message is picked up by the app itself, without AI
• With the local engine the summary is written when the chat goes to your team, so it never holds up a reply

The AI can act, not only talk

Besides replying, it can show the price list, send the form, start a payment for a package, send the right file from your media library, show the menu, save details or hand the chat to your team.

• At most two actions per reply
• Enabled modules add their own actions
• Every action is checked against the allowed choices: a wrong package or file name is dropped

Large catalogues fit

A cloud AI reads up to 24,000 characters of your information at once; above that your business's own information comes first, then products as far as they fit, and a piece left out comes with the message.

Your details filled into your texts

In the welcome message and your business text you can write {brand}, {staff}, {phone} or {whatsapp}; they are filled from your settings in the customer's language, and a line that names a contact you have not set up is left out.

• Also: {telegram}, {email}, {website}, the business name per language, your country's emergency number
• Your own words for customer, offering, booking and order can be used too

Knows what day it is

Today's day and date in your business's calendar and time zone (the Persian solar calendar in Persian) go with every message, so «are you open tomorrow?» is answered right.

Booking requests get the form

When a customer asks for an appointment or visit and a small model handed it to the team for no reason, the form is sent instead; a plain price question gets the price list.

No made-up numbers or times

Any other number, an hour, a duration («takes 2 hours»), a count or a day, must come from your information or the customer's own words. A local reply with a made-up number is not sent; a cloud reply waits for your approval.

• Numbers written as words are read too, in all three languages
• Your opening hours are understood in every form (20:00 = 8 pm)

No made-up colours

A colour in a reply must be in your information or the customer's words, in any language; «we have no blue» is no claim.

• 11 colour families recognised in Persian, Arabic and English

No made-up discounts or freebies

A discount, percentage, free item, refund, VIP status or «the manager approved» that is not in your information is never sent. When the customer asks for a discount or refund, your team decides and gets the chat.

• Same offer recognised across languages: your «free delivery over 100 AED» covers «ارسال رایگان» and «التوصيل مجاني»
• A polite «no discounts» is never a promise

No foreign links or phone numbers

A link that is not your business's, even written as «shop dot com», shortened, or someone else's Telegram or WhatsApp, and a phone number or email that is not yours or the customer's, are stopped.

• Your own links and contacts still go out as buttons
• Card numbers are blocked in replies

Old numbers in your text never reach customers

Before the AI reads your business text, any line with a phone number or a WhatsApp or Telegram link that is not one of your verified contacts is left out, so an outdated number is never passed on.

• Also applied to answers the smart setup builds from your website and chats

Customers cannot give the bot orders

Customer text is always data, never instructions: forged labels like «system:» are defused and the message reaches the AI in its own block. The translator translates a sentence that orders it around and never obeys it.

• The same applies to profile details and summaries the customer influenced

No leaked instructions or secrets

A reply that repeats the bot's instructions, talks about them, or states a key or password is not sent. Internal tool names are removed from the text.

No fake «booked» or «ordered»

The AI never says an order is placed, a booking confirmed or a payment received, and never says a time is free or full; the form goes instead and your team confirms.

• «After you fill the form, the team confirms» is allowed: a next step, not a claim
• A booking feature that really booked can say so

No drug doses, no guaranteed results

Medicine or supplement doses and diagnoses are blocked, and sentences like «100% guaranteed results» are removed; «we can't guarantee» is fine.

Sensitive topics: fixed safe reply and priority handoff

With the safety module, emergency, medical, legal, financial and under-18 messages get a fixed safe reply without AI and reach your team with high priority. Custom rules can be added.

• Emergencies: the country's emergency number
• Sets: emergencies, medical topics, legal questions, financial regulation, age and minors
• Each set on or off, with your own words and replies in Persian, Arabic and English
• Priority: normal, high, urgent
• Without this module there is no stop; general limits such as no diagnosis or dosing stay in place

Sensitive and emergency triage

A blocked reply never goes, no one is left unanswered

When the checks block a reply, the customer gets a fixed safe reply and your team gets the chat. If the customer asked for the bot's own instructions, they only hear politely that this is internal. An upset customer gets an apology and a person, not price cards.

• Fixed replies for: a promise only your team can make, an unknown link or number, a made-up «done», a health topic, internal information, an upset customer, and a general «our team will answer»
• A made-up price is replaced by your price list, a made-up «booked» by the form

Unsure replies wait for your approval

In automatic mode, a reply the checks are unsure of waits, flagged, in the approval queue while the customer hears «we'll reply here soon». In manual approval mode every AI reply is a draft.

• A flagged or corrected draft is never sent on its own by the auto-send timer

«Why did the bot say this?»

Every AI reply is logged: the customer's message, the information pieces used, tools, the result of every check, model, time and cost. Corrected or rejected replies can be filtered.

• Today's replies, average time, cost today and corrected count at the top
• The last 150 replies
• Each sent message links to its log entry
• Per reply: the model's raw text, the active automation mission and a link to the person's record
• How much text each reply used, and how much came from the cloud service's cache
• Today: replies, average time, estimated cost, corrected or made-safe replies
• Filter: corrected / rejected only
• also: provider, model, seconds and cost, how much text went to the model and how much of it was reused from earlier requests

Right language, right length

A reply not in the customer's language is not sent, and the local model is asked once more in that language. A reply over 700 characters is cut at its last full sentence.

• The instructions to the local model are written in the customer's language, so a small model answers in it

Learned answers are checked too

A ready answer built from your team's past chats is checked before it goes: nothing about one person, no made-up amount, link, card number or guarantee.

• Checks: phone · email · reference or tracking code · street address · percentage · amount · «your order» wording · customer names
• The business's own published phone, address and prices are not treated as personal

With the local engine nothing leaves your PC

When replies are written by the built-in engine or Ollama, messages and business information stay on this computer and the AI itself needs no internet.

• The engine accepts requests only from this computer, with a new random key each run

Team notes are masked too

A note the AI writes for your team from the customer's words masks card numbers, IBANs, links and phone numbers, and cannot raise the priority on its own.

One assistant on Instagram, WhatsApp and Telegram

The same AI, ready answers and checks answer direct messages on Instagram, WhatsApp and Telegram, and each reply is shaped to what that app shows best. You can try the assistant in the panel as any of the three apps without sending anything.

• Comments are answered by your automations, not by the AI
• Instagram and WhatsApp's 24-hour reply window is respected

Gulf Arabic for Gulf businesses

For a business in the UAE, Saudi Arabia, Kuwait, Qatar, Bahrain or Oman, Egyptian words a small model sometimes writes become their Gulf forms («عايز» to «تبي»).

Formal or warm tone

Some trades call customers «bro» or «dear». In formal mode (the default) these words are removed before sending; in warm mode they stay. The rest of the sentence is untouched.

Replies in your team's own voice

A «tone card» is built from your team's real replies: how formal, how many emoji, greetings, closings and pet phrases. After your approval the AI writes in that voice.

• Style only: any phrase carrying a fact (a price, a day, a number) is dropped
• Every truth rule comes first
• On your PC it needs the largest model; any cloud AI works
• Learns how you write, never what you said: a phrase with a price, date, time or discount is dropped
• The polite form of address in Persian always holds
• Needs at least 3 replies from your team

Reads the customer's mood and style

Each message shows whether the customer is upset, in a hurry, formal, playful, a short texter, a voice user or wants more detail; the reply's length and tone follow.

• Anger is weighed, never read from one word: a happy «عالیه» with many exclamation marks is not anger, and a customer's own problem («my acne») is not anger at you
• Style tags on the customer's record: upset, short texter, wants detail, formal, voice user

No selling to an upset customer

An upset customer is not shown buttons, the form, the price list or the payment page unless they ask; calming comes first.

One question per message

A reply answers first and asks only one question; what the customer already said is never asked again. If the AI asks five things at once, your short form goes instead and any price in that sentence is kept.

• A sentence with a number in it is never dropped whole: only the offending part

No repeated replies

If the model repeats its last reply word for word, it is asked once more; the second reply is used only when it is new and keeps any price the first had. An offer or question just made is not repeated.

Buttons that fit the moment

Under an AI reply only the button needed appears, often none: yes/no under a yes/no question, services named in the reply, the form after a price, or a person when the customer asks again. A button just shown or tapped does not come back.

• Service choices: WhatsApp list up to 10, Instagram quick replies up to 13, Telegram rows up to 10
• Tapping yes/no is read as the customer's answer

Booking times right after the answer

When a customer wants to book and you use the appointments or reservations feature, the days and times from your calendar follow the reply at once as choices, instead of a button to tap first. Not for an upset customer, and not when the chat goes to your team.

• Understood in Persian, Arabic and English («نوبت می‌خوام», «احجز», «book»)

AppointmentsStays, tables and rentals

Form and price list not sent again and again

A form, price list or menu just shown is not sent again; the reply points to the one above unless the customer asks for it.

Never «message us on Telegram» on Telegram

A sentence sending the customer to the same Instagram, WhatsApp or Telegram they are writing on is removed; if nothing is left, «we'll answer right here» is said.

No robotic phrases

Phrases like «as an AI…» and «I hope this helps» are removed, and it does not greet again in an ongoing chat.

• A sentence like «tap the button below» is removed when no button follows

Tone variety for older cloud models

One setting chooses whether replies are precise and consistent or more varied. It only affects models that accept it (older Claude models and many OpenAI-compatible services); the newest Claude, GPT and Gemini models ignore it.

• Not sent to Claude 4.7 and later, GPT-5 / GPT-6 and the o-series, or Gemini 3 and later

Bot disclosure on the first message

A line saying the customer is talking to an automated assistant is added once, to the first message; you write it in each language.

«I want a real person»

A request to talk to a human is recognised in Persian, Arabic and English and the chat goes to your team at once with a ticket; «I don't want an operator» does not count.

• Short messages of up to 8 words with the «talk to a person» intent
• Modules add their own words (e.g. «coach»)

The AI knows when to hand over

The chat goes to your team with a ticket when the customer asks for a discount or refund, the reply promises the team will answer, the reply was blocked, the topic is sensitive, or the AI cannot be reached.

• Tagged on the customer's record
• Priority normal or high (sensitive topics, failures)

If the team is silent, the assistant steps in

After a handoff, if nobody on the team replies within 5, 10, 20 or 30 minutes, the assistant answers politely, knowing the team has the request. While the team is talking, the bot stays silent.

• 10 minutes recommended; «never» keeps the bot silent
• Outside business hours it does not wait
• Silent for 30 minutes after a team member writes
• Button taps are always answered
• «Never» together with «never hand back» is refused: no customer left without an answer
• Wait time: 5, 10 (recommended), 20 or 30 minutes, or never
• While a colleague is active, the bot stays quiet for 30 minutes after their last message
• The assistant knows the team already has the request

A three-line note for your team

The AI writes a note on each customer's record: what they want, what is known, what is still open. It is written when the chat goes to your team, and with a cloud AI every 6 turns too. Writing it never delays a customer's reply.

• Written in your business language
• Sensitive answers left out
• On your PC it needs the medium model or larger; any cloud AI works
• Refreshed after every 6 assistant replies and when the chat is handed to the team
• Personal details are masked in the summary
• Needs the AI assistant to be set up

No AI? Still nobody is left unanswered

If the AI is not set up, cannot be reached, or this computer's model is not good enough to reply, the customer hears at once that the message arrived and the team will answer, and a ticket opens.

Handoff after hours

When a chat is handed over after hours, the after-hours message with the next opening time and your contact buttons go with the reply.

• Contact buttons for the app the customer is on are left out

Four model sizes

From very small to large, each for a level of computer; the smaller ones come with the app and the largest downloads with one press.

• Qwen3.5 0.8B, 2B, 4B and 9B
• The 9B is downloaded (about 5.7 GB); the disk space is checked first
• Measured customer replies right: 53% (0.8B), 62% (2B), 81% (4B), 86% (9B)
• Gemma 4 E2B can be chosen as another small model; models copied from Ollama are marked untested or for comparison

The right model chosen for your PC

The app measures memory, processor and graphics card, picks the right model and times a test write; if it is slow it steps one model down. If the hardware changes, it chooses again.

• Four levels: too weak for a reply model, weak, average, strong
• A change is acted on only when two separate checks see it
• Your own model choice is never changed
• «Measure this computer again» button
• Weak: 4 GB memory or more · Average: 15 GB memory and 8 processor threads, or a 4 GB graphics card · Strong: an 8 GB graphics card and the large model downloaded
• Level 0: under 4 GB memory, or an old processor with no graphics card: no AI model; ready answers, flows and team replies still work
• Level 1: 4-6 GB memory: Qwen3.5 0.8B; 6 GB or more: Qwen3.5 2B (on the processor)
• Level 2: a processor with 8+ threads and 15+ GB memory, or a graphics card with 4+ GB: Qwen3.5 4B

Live speed check and model advice

The writing speed of real replies is measured. After 5 replies a smaller model is suggested if it is slow, a bigger one if it is very fast, and the top one on a strong computer; nothing changes until you press.

• Strong computer: a 12 GB graphics card, or 64 GB memory with 24 threads
• A slow graphics card falls back to the processor for that run and is tried again later
• Very strong means a graphics card with 12+ GB, or 64+ GB memory with a 24+ thread processor

Ollama support

Instead of the built-in engine you can choose Ollama and any model installed in it. If the built-in engine fails and Ollama has the model, the reply comes from Ollama.

• Model size is read from Ollama, so the same quality minimums apply
• From the System page Ollama models can be downloaded, removed, speed-tested or chosen; the model in use cannot be removed
• Disk space is checked before a download; the download needs a confirmation and can be cancelled
• The model in use cannot be removed by mistake
• «Quick speed test» measures a model's real speed on this computer

Claude, ChatGPT and Gemini

Connect a cloud service with your own key; the key is tested with a real request, your account's models are listed with one recommended, and errors are explained plainly (invalid key, no credit).

• Anthropic Claude: Sonnet 5.5, Opus 5.5, Haiku 5.5
• OpenAI: GPT-6.1 Sol, GPT-6 Astra, GPT-6 Luna, GPT-5.4 mini
• Google Gemini: 3.8 Flash, 4 Argon, 3.5 Flash-Lite
• Any OpenAI-compatible service through its own web address
• Used only after its key passed the test
• A «test with a message» button

Local model steps in when the cloud fails

When credit runs out, the key stops working or the internet is down, the same message is answered by this computer's model and the panel shows why.

• One switch, on by default
• A missing key never breaks the bot: the local engine answers

Monthly cost cap for cloud AI

Set a monthly cap in US dollars; when reached, replies use the local model until the month ends. Without a monthly cap, a daily cap stops a surprise bill from a flood of messages.

• Daily cap $5 by default
• Cloud voice replies count toward the cap
• No customer is left unanswered when a cap is reached

Alerts at 80% and 100% of the cap

At 80% of the cap and when it is reached, you are told once a day in the panel bell and on Telegram.

Costs always visible

This month's cost with a progress bar, today's cost, the cost of each reply and an estimated cost per 1,000 messages for each service are shown.

• Estimates at the providers' list prices
• Text sent, received and reused for each reply

Each task only with a model that can do it

For each AI task the smallest model that does it right was measured. With a weaker model the task is not done and work goes on without AI: never a wrong note, translation or reply.

• Customer replies: from 2B
• Replies in your team's tone: 9B
• Team note, finding a product, translation: from 4B
• Understanding free answers, picking out details, sorting messages, reading hours and websites, learning your team's tone, spoken voice wording: from 2B
• Building a flow from a sentence, panel helper, describing photos: any model
• Any cloud AI runs every task

What this computer can do, explained

A plain-language page says what the app does on this computer, what it does not and exactly why (memory, graphics card, processor, disk), and which upgrade turns on which task.

• Upgrades suggested: more memory, a graphics card, a processor, or a cloud AI key
• Measured numbers shown

Works on a weak computer too

On a computer too weak for a reply model, menus, buttons, forms, ready answers and flows work and other chats go straight to your team, never a long wait.

• Under 4 GB memory, or a processor without AVX2 and no graphics card

Engine settings in your hands

On the System page choose whether the engine runs on the processor or the graphics card, how many processor cores it uses and how much of the chat it keeps in memory, start or stop it, and check any model file against its recorded fingerprint.

• Everything automatic by default
• Processor cores: automatic or a number you set
• Conversation memory from 512 to 32,768 text units
• One button switches replies between the built-in engine and Ollama
• Settings: run on the processor or graphics card, how much of the conversation it keeps in mind, processor cores, 1-4 chats at once
• Each model shows its job: writing replies, search by meaning, pictures, reading voice messages, making the voice

Step-by-step help to get a key

For Claude, ChatGPT and Gemini the panel shows where to sign up, add credit and create a key, with a link to that page, and warns at once when a pasted key has the wrong form.

• The key is stored only on this computer and is not shown again
• Only one key is needed

Lower cloud cost through caching

The fixed part of the instructions and business information is sent apart from each person's part so the cloud service keeps it; for Claude up to one hour.

Faster replies on an ordinary PC

Each language's instructions are read once and saved to disk, long history is trimmed to the model's room, the model is kept warm, and background notes never hold up a customer's reply.

• A built-in reply that takes over 90 seconds is given up and handled like any AI failure
• The newest 12 ready states are kept; older ones are removed by themselves

Engine test and model comparison

At one press the most common customer question is asked in Persian, Arabic and English with your real business information, and you see whether the reply would be sent and, if not, what the customer would get instead. The model contest on real trade questions (right answer, confident wrong, nothing made up, right language, time) is shown too.

• Reading and writing speed shown for each language
• A made-up price or number is shown as «would NOT be sent», with the safe reply the customer would get instead
• Blocks a made-up price or number that is not in your business info
• Questions on hours, address and prices across several trades, in 3 languages, through the real reply path with the number check
• Measured the same way real replies are made, with the same guard against made-up numbers
• Shown when a contest has been run on this computer