Appointment booking bot
The bot clarifies the service and specialist, offers free slots from your calendar, reminds clients and reschedules visits in chat. We start with a free review.
12–25 days
11 services
AI for business: chatbots, product descriptions and model integration
This area covers three kinds of work, and what separates them is not the model but where its answer ends up. A chatbot talks to your customer: it answers from your catalogue and knowledge base, and in the fuller setup it takes the cart, Nova Poshta delivery and payment right inside the conversation. Description generation works on the catalogue: it rewrites product cards from your own attributes and returns a feed ready for import. Integration puts the model inside your system, with answer validation, a spending cap and a log. All three share one limit: without data a model gets things wrong, so we build a frame in which it answers only from what you gave it and passes everything else to a person. The first review of your task is free.
- Bot accuracy
- 89.7%
- on 45 questions
- LEO Chat reference set on the product's own knowledge base, measured 4 July 2026; your bot is tested on your questions
- Chatbot channels
- Telegram · website widget
- Instagram, Viber and WhatsApp are not supported yet — if that is your main channel, we will tell you up front
- Product descriptions
- Feed in, feed out
- XML, YML, CSV or Excel; each description is built from your attributes, with no invented properties
- Timelines
- 7–22
- days for chatbots
- model integration takes 10–30 business days, catalogue descriptions 10–40; the review gives the exact figure
- Payment in the chat
- Monobank · LiqPay · cash on delivery
- through the store's own merchant account: the money goes to its account, and the confirmation and Nova Poshta tracking number arrive in the same chat
Chatbots
All 9Bot for answering common questions
The bot answers only from your knowledge base and makes nothing up: 89.7% accurate answers on a reference set of 45 questions. Launch in 7–15 days, free review.
Bot with in-chat payment
The buyer fills a cart, picks a Nova Poshta branch and pays by card or cash on delivery without leaving the chat — through your store's own merchant account.
Chatbot for an online store
The bot answers from your catalog, builds the cart, arranges Nova Poshta delivery and takes payment through your merchant account. First, a free catalog review.
Instagram chatbot
A bot for Instagram Direct: answers from your catalog, a private reply to a comment, handover to a manager. Built around the 24-hour window and Meta app review.
9–20 days
OLX auto-reply bot
The bot reads OLX chats through the official API and drafts replies from your catalog for a manager to approve. OLX limits and risks are reviewed upfront.
10–21 days
Telegram chatbot
A Telegram bot answers from your catalog and guides buyers to an order: cart, Nova Poshta branch, payment via your own provider. Launch in 8–16 days in total.
Viber chatbot
We build a Viber bot for your store: answers from the catalog, orders with Nova Poshta. We go over the Viber application and monthly fee before signing.
11–24 days
Website chatbot
The widget answers from your catalog and takes orders right in the chat. The script is 19 KB gzip, measured on 7 September 2026. We start with a free review.
Content
Integrations
Who it's for
Situations where this service delivers results
Scenario 1 of 4
Your manager answers the same questions every day
Stock, delivery, sizes, opening hours — dozens of identical messages eat the time that complex orders need. The bot answers the repetitive ones from your knowledge base and hands anything unusual to a person together with the chat history.
We'll review your situation in a free auditScenario 2 of 4
Customers write in the chat but order somewhere else — or nowhere
The conversation breaks off at “tell me how to pay” or at the jump to the site's cart. A bot with in-chat ordering collects the cart, the Nova Poshta branch and payment through the store's own merchant account, then posts the tracking number back into the chat.
Scenario 3 of 4
A catalogue of thousands of items with supplier copy
The descriptions came with the price list and repeat across dozens of stores, and there are neither the hours nor the budget to write them by hand. Rule-based generation covers the volume; items without attributes are set aside on a separate list rather than filled in from imagination.
Scenario 4 of 4
Someone already runs the model through a chat by hand
A colleague pastes data into a chat and copies the answer back into your system. At hundreds of items that becomes a job of its own. Integration moves the model into your process: with an answer schema, a queue, caching and a monthly spending cap.
What's included
Complete list of work and what you get as a result
- A free review of the task that tells you where a model is genuinely needed and where a rule or a database lookup is cheaper
- Testing on your real data before any promise about quality: reference questions for a bot, a trial batch for descriptions, a prototype for an integration
- An answer boundary built into the system: the bot answers from the knowledge base and hands unknowns to an operator with the full chat history
- Product descriptions built only from your attributes, checked for repeats inside your catalogue and uploaded with the option to roll back
- For a model integration, the provider's key and billing in your name: you see the costs directly and can stop them without us
- Safeguards for production: schema validation of answers, a retry queue, caching of repeated requests, a monthly spending cap
- A request log with personal data masked, and the cost of a single operation tracked
- Instructions for your staff: how to extend the knowledge base, generate descriptions for new items, change a prompt or stop processing
- Work under a written contract, with a 30-day warranty from the day the acceptance certificate is signed
Process steps
Transparent stages with approval at every step
Task review
We look at your enquiries, catalogue or process and separate the steps where a model helps from the ones where a rule will do. At the same point we record which data would go to the model provider.
Testing on your data
A bot is run against real questions, descriptions get a trial batch, an integration gets a prototype with the cost of one request measured. This step decides whether the rest makes sense.
Frame and integration
The “answer / hand over to a person” boundary, the answer schema, the queue and spending limits, connecting channels or uploading to the catalogue.
Launch and handover
A run on the real volume, spot checks by a person, instructions and access. The 30-day warranty starts on the day the acceptance certificate is signed.
When this service isn't right
What's not included — so there are no surprises at delivery
- A guarantee that the model will never make a mistake: errors are reduced by checks and data, not by promises
- Generating demand: the bot answers people who have already written in and does not bring new customers by itself
- Instagram, Viber and WhatsApp as bot channels — for now only Telegram and the website widget work
- Training a custom model on your data — that is a different scope of work and a separate budget
- A legal opinion on sending data to an external provider: we give you the technical list of fields, your lawyer draws the conclusion
- Copy for the home page and category pages: that calls for a person, not generation
FAQ
Questions asked before starting
Which of the three jobs do I actually need?
It depends on where the model's answer has to land. With a customer in a conversation — a chatbot. In catalogue cards — description generation. In your system, an email or a database — model integration. Sometimes none of them: at the review we say plainly where a rule or a search does the job without a model.
Can the bot make up an answer?
A model without data gets things wrong, so the bot answers only from your knowledge base and passes anything outside it to an operator. On a reference set of 45 questions on 4 July 2026, LEO Chat gave 89.7% accurate answers and none outside the knowledge base. That is a measurement of our product on its own data; your bot is tested on your questions before launch.
Will the bot replace a manager?
No. It takes the repetitive questions and routine checkout, while custom pricing, disputed returns and unusual cases are still closed by a person. And the bot only works with people who have already written in: it does not create demand.
Who pays for requests to the model?
It depends on the job. In a model integration, you do, directly to the provider: the key and billing are in your name, and we do not resell requests. The provider charges by the volume of text in and out, so we show the cost of one operation on a sample of your data during the review, before any contract. Catalogue descriptions in Textory are paid for by the actual volume of text. Terms for a chatbot are given after the review.
Will our data go to third-party services?
If the model is external, some of it will. So at the review we list the fields that would have to be sent and those that can be removed or replaced; personal data is masked in the log. When data must not leave your premises at all, there is the option of a model on your own server — a different scope of work.
How do LEO Chat and Textory fit in?
They are our own products, and two of the three jobs are built on them. LEO Chat is a chatbot with in-chat ordering for Telegram and websites; its card is in our Products section. Textory rewrites catalogue descriptions and publishes a feed that LEO Chat picks up on a schedule, so the bot answers from the new descriptions.
What happens to items with few attributes?
We generate nothing for them: three fields only produce filler. Such items go on a separate list — data first, descriptions second. In Textory an item with no description, or one shorter than 20 characters, is marked as skipped: there is nothing to rewrite.
What is the GEO score of a description?
A deterministic 0–100 score calculated by Textory: a direct answer in the first sentence, question subheadings, fact density, entities, an attribute list and length. The model does not grade itself here. The score does not guarantee a mention in AI answers — it only shows how structurally suitable the description is for being quoted.
How do leads from the bot reach our CRM?
Through a webhook or the public API. LEO Chat has no ready-made integrations with KeyCRM or SalesDrive yet — those are product plans, not features, and we will not present them as existing.
Not sure which service is yours?
Describe the task in your own words — we will tell you where to start and whether you need what you came for. If you do not, we will say so.
- Reply within 2 hours
- No commitment
- We work under a contract