How to build your own AI assistant in Telegram — and when a ready-made one is simpler
You can put together a simple AI bot in Telegram in one evening: a bot from @BotFather, a server and an AI model over an API. It becomes an assistant later, once it can remember, understand voice messages, send files and set reminders. If you don't code and are deciding whether to build it yourself, first estimate how many hours a month maintenance will take and what your hour is worth. Below: what an assistant is made of, how to build one, what it costs and when a ready-made one is simpler.
We build an assistant like this ourselves — more about it in a separate section at the end.
What an AI assistant in Telegram is made of
From the outside it is an ordinary chat. Inside there are six parts, and you will have to build or connect each one.
- The bot. You create it for free with @BotFather using the /newbot command. The bot gets a key that your program uses to control it. Telegram calls this key a token, and you keep it like a password.
- The server. The program has to run around the clock, or the bot goes silent. Usually this is a rented server or a computer that is never switched off.
- The AI model. The program sends it your message and gets a reply back. You pay by the amount of text. That is also counted in tokens, but these are different tokens: units of text, not the bot's key.
- Memory. The AI model remembers nothing between requests on its own. The program stores the conversation history, your rules and facts about you, and adds them to every request.
- Tools. For the assistant to do work, not just answer, you connect it to spreadsheets, a calendar, search, and file and image generation.
- Voice and scheduling. A separate service transcribes voice messages. A scheduler sends reminders at the right time without you asking.
Three ways to build it yourself: a no-code builder, your own code or an open-source agent
A no-code builder, if you don't program. You assemble a workflow from blocks: “message received → request to the AI model → reply.” A popular option is n8n. You can install it for free on your own server for your own use, or take the cloud version, from €20 a month billed annually.
Your own code, if you have a developer. The bot is written in Python, for example with the aiogram or python-telegram-bot libraries, and you connect the AI model, memory and tools yourself. Full control, but all the maintenance is on you.
An open-source agent, if you are comfortable with the command line. This is a ready-made assistant program that you install on your own computer or server: OpenClaw or Hermes Agent. Both are free (MIT license), connect to Telegram and run on a schedule. You need a server, your own API keys and attention to security: the agent runs commands on the machine where it is installed.
There is also a way without building anything: a ready-made assistant on a subscription. More on that at the end.
Step by step: a simple assistant on n8n
The shortest route for people who don't code:
Create the bot. In Telegram, open @BotFather, send /newbot, and choose a name and a username. Save the token.
Get an AI model key. With a foreign provider, first check whether it works in your country and how you can pay. For example, Russia is not on OpenAI's list of supported countries. YandexGPT (by Yandex) and GigaChat (Sber's AI model) can be paid for in rubles.
Rent a server and install n8n on it, or sign up for n8n cloud.
Build the workflow. A Telegram Trigger block receives the message, an AI block prepares the reply, and a Telegram block sends it back.
Add simple memory. Write instructions about yourself and your rules, and store the latest messages of the conversation.
Lock down access. Make the bot reply only to your Telegram ID. Otherwise anyone who finds it can use it, and you will pay for the AI model.
Test it on real tasks and add what is missing: voice messages, files, reminders, calendar.
You can get through the first six steps in an evening. The last one never ends: every new feature means a new block, new bugs and new costs.
Telegram limits you discover in practice
- A bot can't message first. Until a person writes to the bot or adds it to a group, the bot can't start a conversation with them, as the Telegram documentation says. Reminders will only work after you press “Start.”
- Files. A bot can download a file you send it up to 20 MB and send files up to 50 MB (Bot API). A long meeting recording won't get through to such a bot. Only running your own Bot API server removes this limit.
- Voice messages arrive as files. They contain no text: the bot downloads the recording and sends it for transcription. For example, Yandex SpeechKit costs about 0.65 ₽ (under 1 cent) per minute.
- Speed. No more than one message per second to a single chat (bot FAQ). That is enough for a personal assistant. But if a reply is split into several messages, you need pauses between them.
What it costs to build yourself: monthly expenses
Telegram bots are free. You will pay for the server, the AI model and voice transcription. Estimates as of October 2026:
| Cost | Comment | |
|---|---|---|
| Telegram bot | Free | The bot platform is free for both users and developers |
| Server | About 680 ₽ (about $8) a month, at minimum | Cheapest plan with a public IP address at Reg.ru, a Russian hosting provider. It may not be enough for n8n or an open-source agent |
| AI model: Claude Sonnet 5.5 | About $33 a month | $2 per 1M input tokens and $10 per 1M output tokens. A foreign service: check payment first (step 2) |
| AI model: YandexGPT Pro 5.1 | About 8,400 ₽ (about $104) a month | 800 ₽ (about $10) per 1M tokens |
| Voice transcription | About 0.65 ₽ (under 1 cent) per minute | Yandex SpeechKit, synchronous recognition |
| n8n cloud | From €20 a month | If you don't install n8n on your own server; price billed annually |
How we estimated the AI model: 100 messages a day, each with 3,000 input tokens (history and rules) and 500 output tokens. Yandex prices are in rubles including VAT; Anthropic prices are in dollars before taxes. Sources: Anthropic, Yandex AI Studio (in Russian), SpeechKit (in Russian), Reg.ru (in Russian), n8n, as of October 19, 2026.
The biggest expense isn't in the table: your time. An assistant has to be updated, fixed and extended. The author of this article spent more than 500 hours on his own system before he started offering it to clients.
Build it yourself or get a ready-made one: a comparison
| Build it yourself | Ready-made assistant on a subscription | |
|---|---|---|
| Launch | From an evening to several weeks, depending on the features | The provider sets the timeline, so ask in advance. They handle setup and configuration |
| Who fixes it | You | The provider: find out in advance how quickly they respond |
| Costs | Server, AI model, voice transcription and your time | Subscription |
| Data | Wherever you decide, even on your own server | With the provider: find out who can see it and whether you can export it |
| Flexibility | Any integrations, if your skills allow | Within what the provider offers |
When it makes sense to build it yourself
- You enjoy figuring things out and have the time for it.
- Your data must stay only with you, on your own server.
- You need custom integrations: your own CRM, internal company systems.
- You have a developer who will maintain the assistant.
When a ready-made one is simpler
- You need results this week, not in a month.
- You don't want to deal with servers, API keys and paying for foreign services.
- You need finished files, a calendar and reminders, not just answers in a chat.
- You want someone to contact when something breaks.
A simple test: if the maintenance time alone costs more than a subscription, building it yourself doesn't pay off, and the server and AI model come on top of that.
If you decide not to build: our subscription assistant
We make a personal AI assistant in Telegram, MAX and the browser: everything described above is already built and set up. The assistant remembers your business, understands voice messages, even long ones, sends finished documents, spreadsheets and presentations, and messages you first with reminders. It works with Google Sheets and Docs and Notion, and, with your permission, with Google Calendar and iCloud and Yandex calendars.
Your assistant runs on the same system as the author's own assistant. It is ready within 30 minutes of payment if the payment arrives before 6:00 p.m. Moscow time, otherwise the next morning. If something breaks, a person responds within 4 business hours.
The price is 5,000 ₽ (about $62) a month, paid monthly in rubles, with no long-term contract. Our trial bot in Telegram will do your first task for free, right in the chat. If this is your first subscription and you change your mind within the first month, we will refund the full payment, no questions asked. All the terms are on the home page, and how to choose an assistant is covered in our guide.
Sources
- Telegram: what bots can do, creating a bot, Bot API, limits.
- n8n: pricing, repository and license.
- Open-source agents: OpenClaw, Hermes Agent.
- AI models paid in rubles: YandexGPT (in Russian), GigaChat (in Russian).
- OpenAI: API countries, creating GPTs.
- All links checked on October 19, 2026.