Telegram AI Bot: Every Option in 2026 and What Each Costs
Four ways to get an AI bot on Telegram in 2026: public bots, built-in AI, self-hosting, or managed. Real costs, platform limits, and what breaks.
Telegram AI Bot: Every Option in 2026 and What Each Costs
A Telegram AI bot is an assistant that answers inside your Telegram chats, and in 2026 there are four ways to get one: use a public bot somebody else runs, use the AI features Telegram now builds into the app, build and host your own, or pay for a managed private assistant such as Tessera’s Founder plan. Which option is right for you comes down to a single question: do you want to babysit a server?
This guide covers all four, with the real costs and the failure modes nobody puts in the sales page.
Why Telegram Became the Default Home for AI Agents
On 7 May 2026 Telegram shipped an update that changed what a bot can do. Three of the changes matter for AI assistants.
Guest AI bots. You can now get a response from an AI bot in any chat by tagging its @username, even in chats where the bot is not a member. The bot only sees the message that tagged it and the replies to that message, which keeps the rest of the conversation private.
Bots can answer other bots. Telegram’s wording is that bots “are now able to respond to other bots, not just users”. That turns a bot into something an automated workflow can call, not just a person.
Streaming responses. Bots can stream a reply as it is generated rather than holding it until the whole message is ready. An assistant answering a long question now feels like a chat instead of a delay.
Add what was already true and the appeal is obvious. Telegram charges no per-message fee for bots, getting a token takes minutes with no business verification, and your bot needs no inbound ports open.
Telegram vs WhatsApp for AI Assistants
The channel choice is worth settling before you build anything.
Meta banned general-purpose AI chatbots from the WhatsApp Business API on 15 January 2026, which pushed ChatGPT, Copilot and Perplexity off the platform. From 1 October 2026 Meta also charges per service message on WhatsApp, with its own AI exempt from the charge.
Telegram does neither. That is the whole reason serious assistant projects defaulted to it this year. The full comparison is in AI on WhatsApp.
The Four Ways to Run a Telegram AI Bot
| Option | Setup effort | What you pay | What breaks |
|---|---|---|---|
| Public bot someone else runs | None | Free or a subscription | Owner shuts it down, changes limits, or reads your data |
| Telegram’s built-in AI | None | Included, some in Premium | You get only what Telegram ships |
| Build and self-host | Hours to days | Server plus a variable token bill | You are on call for it, forever |
| Managed private assistant | Minutes | Flat monthly fee | Less low-level control than self-hosting |
Option 1: Public Bots Somebody Else Runs
The fastest option. You search for a bot, press start, and you are talking to a model. Third-party aggregator bots are the most common kind, bundling access to several commercial models behind one chat, usually with a free tier and a paid upgrade.
The tradeoff is that you are a guest. You do not control which model answers, what happens to the conversation, or whether the bot exists next month. For casual questions that is fine. For anything touching client data or business process, read the operator’s privacy policy before you type.
Option 2: Telegram’s Built-In AI Features
Telegram now ships AI capability in the app itself, including the guest-bot tagging described above alongside productivity features in chats and channels. There is nothing to set up and nothing to run.
The limit is the obvious one. You get the feature set Telegram decided to build. It is not your assistant, it does not know your business, and you cannot give it your own instructions or connect it to your own tools.
Option 3: Build and Self-Host Your Own
What is genuinely free. Creating a bot through @BotFather costs nothing and takes minutes. The Telegram Bot API itself has no per-message fee, a point worth repeating because it is unusual among messaging platforms.
What is not free. Your bot is a program that has to be running whenever someone messages it, which means an always-on server. A minimal virtual private server starts around $5 per month, and that figure is where most guides stop. It is also where most cost estimates go wrong, because a bot that only relays messages is not the assistant people actually want. Once you add persistent memory, document handling and tool use, you are past the smallest instance.
The variable that surprises people. The server is the predictable cost. The model is not. Every message your assistant processes consumes tokens, billed by whoever serves the model, and the bill scales with how useful the assistant becomes. We worked that arithmetic through in how much OpenClaw costs to run.
What Self-Hosting Costs After Launch
You are now operating a production service for yourself, and that part has no price tag on it.
Dependencies need patching. The process needs to come back up after a reboot. Your bot token and model API keys are credentials that grant control of the assistant, so they need to be stored properly and rotated when they leak. When a model provider changes an endpoint, your assistant stops answering until you fix it, and you find out because someone messages it and gets nothing back.
None of this is hard for an engineer. It is simply never finished. If you want to try it anyway, our OpenClaw setup guide walks through the install honestly, including the parts that take longer than five minutes.
Option 4: A Managed Private Assistant
The managed option exists because most people who want an AI assistant on Telegram do not want to become its system administrator.
Tessera’s Founder plan is our version of this. It runs a private Hermes assistant on your Telegram for a flat €55 per month, with the AI included rather than billed per token. There is no server for you to provision, no dependency to patch, and no separate model invoice that grows with usage. The models are open source, primarily Qwen3.6-35B-A3B, and thinking mode is not rate limited.
Being fair about the tradeoff: you get less low-level control than running everything yourself. If your requirement is a custom fork with your own patches, self-hosting is the honest answer, and we would rather say so than sell you the wrong thing. If your requirement is an assistant that answers reliably every day without becoming a maintenance job, that is what a managed plan buys.
Other managed options exist, and we compare them in OpenClaw alternatives.
Platform Limits Every Telegram Bot Hits
Whichever option you choose, Telegram’s own rules apply.
Privacy mode. By default a bot in a group does not see every message. It receives commands, replies to itself, and mentions. An administrator can disable privacy mode to give the bot full visibility, which matters because an assistant that sees only a fraction of a conversation gives worse answers. Decide this deliberately, since it is a real privacy choice for everyone else in the group.
Rate limits. The commonly documented limits are roughly one message per second to the same user, around thirty messages per second across different users, and about twenty messages per minute into the same group. A personal assistant will never come close. A bot broadcasting to a large audience will, and Telegram sells a higher-throughput broadcast tier for that case.
No per-message fee. Telegram does not bill you for messages your bot sends or receives. Your costs live entirely in hosting and in whoever serves the model.
Which One Should You Pick
Use a public bot if you want to ask a model a question occasionally and nothing you type is sensitive.
Use Telegram’s built-in features if you want convenience with zero setup and no custom behaviour.
Self-host if you are technical, you want total control of the stack, and you accept being permanently on call for it.
Go managed if you want a private assistant that works every day and you would rather spend your time on your business than on uptime. That is the case the Founder plan was built for.
FAQ
Is the Telegram Bot API free in 2026?
Yes. Telegram charges no monthly fee and no per-message fee for bots, and creating a bot with @BotFather is also free. Your costs come from hosting the bot somewhere and from whoever serves the AI model behind it.
Can a Telegram bot read all messages in a group?
Not by default. Privacy mode means a bot only receives commands, mentions and replies to its own messages. A group administrator can turn privacy mode off to give the bot access to everything.
What is the cheapest way to run a Telegram AI bot?
A free public bot costs nothing. For your own bot, the floor is a small virtual private server at roughly $5 per month plus the model cost per token. That floor is misleading, because the token bill grows with how much you use the assistant while a flat managed plan does not.
Do I need a server to have an AI assistant on Telegram?
Only if you build it yourself. A self-hosted bot needs an always-on machine. A managed assistant runs on someone else’s infrastructure, so you connect it to Telegram and use it without provisioning anything.
Can Telegram bots talk to each other?
Yes. Since the May 2026 update, bots can respond to other bots rather than only to people, which makes it possible to chain bots into automated workflows.