A Telegram support bot
that answers from your docs, not a script
Most support bots fail the same way: a rigid menu for the three questions someone expected, and a dead end for everything else. An AI support bot reads your actual knowledge base and answers in sentences, then hands off cleanly when it is not confident.
What it is and who needs it
An AI support bot in Telegram answers real questions in sentences, grounded in your actual documentation, pricing, policies, FAQ, rather than a fixed menu tree that dead-ends the moment someone asks something slightly off-script. It suits any business getting the same twenty questions over and over in a support inbox: a shop, a service business, a membership community, anything where a human is currently typing near-identical answers all day.
It is not the right build for an agent that needs to take action, issue a refund, change an order, cancel a booking, that is an AI agent runtime with scoped tools and approval rules, a different and more involved build. A support bot that only answers questions is simpler, cheaper and faster to ship.
What is inside
The bot answers from a knowledge base you control: product pages, policies, pricing, past answers, anything you want it to know. When a question falls outside that knowledge, or the bot’s confidence is low, it hands off to a human with the full conversation attached, instead of guessing or looping the customer in circles. Every exchange is logged so you can see what people actually ask and where the knowledge base has gaps.
Admins edit the knowledge base themselves through a simple panel, no developer needed to fix a wrong price or add a new policy. We built exactly this pattern for a visa consulting centre: a client-facing AI consultant, an admin bot that edits the knowledge base live, and an alerts bot, with the client’s own staff updating content without ever touching code, and the model migrated to a newer version once with zero downtime.
How we build it
We start from your actual support history, the questions people really ask, not a guess at what they might ask, because the gap between the two is usually where a bot either earns trust or loses it in the first week. Every answer is grounded in a specific document, so the bot cannot wander into territory it was never given, and we test it against adversarial questions, someone deliberately trying to get a wrong or made-up answer, before launch, because cooperative testing alone misses that failure mode.
Rate limits and a spend cap keep AI costs predictable regardless of traffic spikes, and a hand-off rule sends anything genuinely uncertain to a human rather than letting the bot bluff its way through. We deploy with monitoring so a stuck or erroring bot gets flagged immediately, not discovered days later in a pile of unanswered messages.
Timeline and price
| Tier | Price | Timeline | What changes |
|---|---|---|---|
| MVP | from $1,200 | 2 to 3 weeks | One core flow, one channel, enough to go live and test demand |
| Production | from $2,800 | 3 to 6 weeks | Full flow, admin tools, integrations, monitoring, ready for real volume |
| Full control, handover-ready | from $4,760 | 5 to 8 weeks | Same build, plus documentation, test coverage and a handover package so any developer can take over, not just us |
Running cost is typically $20 to $80 a month in AI usage depending on volume, separate from hosting.
What you own at the end
The knowledge base, the conversation logs, the model gateway configuration, and the full source code. If you want to switch AI providers later, the gateway pattern lets you do that without rebuilding the bot around a single vendor.
Related
Pairs with a Telegram channel management bot if support questions also come through your channel’s comments, and with a website chatbot for the same knowledge base on your site. See help center and knowledge base and AI content moderation for related pieces. For the real build this pattern comes from, see the visa consulting centre AI support bots case study.
Answering the same questions all day? Get in touch with a sample of your support inbox and we will show you what a bot could handle.
FAQ
How much does an AI support bot cost?
A bot answering from a knowledge base of a few dozen articles, with hand-off to a human, starts at $1,200 and is live in two to three weeks. A multi-language bot with a full admin panel and analytics runs $2,500 to $5,000.
How long does it take to launch?
Two to three weeks to a working bot once we have your documentation or FAQ content. Building that content from scratch first adds time, and is worth doing properly rather than rushing.
What AI model does it use?
Claude or GPT depending on your needs and budget, behind a gateway that lets us switch models without rebuilding the bot, the same approach that let one of our support bots migrate models with zero downtime when the old one reached end of life.
Who owns the bot and the conversation data?
You do. Conversations are logged in a database you control, the knowledge base lives in your own admin panel, and the repository is yours from the first commit, not locked inside a SaaS bot builder.
What stops the bot from making things up?
It answers only from your documents, with a hard rule to say 'I don't know, let me get someone' rather than guess when the knowledge base does not cover a question. We test this specifically before launch with questions designed to tempt a wrong answer.