No new app to get adopted
The bot lives in Telegram or WhatsApp, which your team and your customers already use. No training, no app to install: you write, it is recorded.
AI agents
Your team enters data, makes requests and receives its alerts from the messaging app it already opens all day long, and your customers get an answer on WhatsApp. Every exchange goes straight into your CRM or your database, with no re-keying.
The bot lives in Telegram or WhatsApp, which your team and your customers already use. No training, no app to install: you write, it is recorded.
An expense, a reading or a request is entered in the field, as it happens, and lands in the right tool. No more notes to copy out or receipts lost in a pocket.
The questions that keep coming back on WhatsApp get their answer without tying anyone up, and your team keeps its time for the requests that matter.
The bot sorts and prepares; it does not decide for you: the steps that commit you are validated with one button.
Everything is logged and linked to your CRM: you know who asked for what, when, and what came of it.
Without AI, the bot costs only its hosting. With AI answers, each message goes to the cheapest model that does the job, and spending is measured and capped: no nasty surprise on the bill.
We first look at the tools you already use, what connects to them and the messaging app your team really uses, then together we choose the three scenarios that save you the most.
We build the bot and connect it to your tools.
Two or three of your users try it in real conditions.
The whole team adopts it, and we adjust based on real conversations.
Telegram for your team: simple, reliable and with no platform fees. WhatsApp for your customers, with Meta’s rules: a dedicated number and message templates, which we set up for you.
It replaces data entry and sorting, not judgement. Your team spends less time copying and stays in control of decisions.
Yes, when a request cannot be settled in a single step. The bot then works in a loop: it plans, acts, checks its result and starts again until the goal is reached. We cap its number of turns, limit its rights to the planned scenario, and anything that commits the company (a refund, a quote, a booking) waits for a person to validate it. Otherwise, a fixed sequence is enough, safer and cheaper.
Because a message does not always need the same one, and the price follows the model. A small model works out what the person wants and files the request, a model good at extraction reads the photographed receipt, and only a hard question goes up to the most powerful one. An answer already given comes back from the cache instead of being paid for again. It is the method we applied on Palmora Property, a real-estate portal in six languages: switching to a suitable AI model divided the translation cost by 60, and 95% of segments come from the cache.
Exchanges are logged, personal data is masked in the logs, and consent is handled on the WhatsApp side as Meta requires.
20-minute diagnostic
We look at your situation, identify what can be automated or improved, and tell you plainly whether we can help.