Guide
How to run an AI assistant from Telegram
A chat app is already where you think out loud. That makes it a natural place for an assistant, and a risky one if you connect the wrong thing.
Written by Tallis. Last updated 2026-09-18.
Why a chat app suits an assistant
An assistant earns its keep in the small moments: the thought on the walk home, the follow-up you remember in a queue, the file someone sends you at nine at night. A chat app is already open for those moments. There is no dashboard to visit and no tab to find.
Telegram in particular gives an assistant three useful things: fast delivery on every device, a place to drop files and voice notes, and buttons, so an approval can be a single tap instead of a typed sentence.
What to check before you connect one
Who can read the chat
A bot chat is not end to end encrypted. Whoever runs the bot can read what you send it. That is true of every assistant bot, including ours, and it is the first thing to be honest about. Ask where the messages are stored, for how long, and whether they are used for training.
Whether your space is your own
A single shared workspace behind many customers is a different risk from a private space per customer. Ask which one you are getting, and ask what separates one customer's files, memory and logged-in sessions from another's.
What it is allowed to do
Reading your calendar is not the same as writing to it. Drafting a reply is not the same as sending one. Get the list of actions it can take without asking, and make sure you agree with it before you connect an account.
How you get your data back
Ask how to export, how to disconnect an account, and how to delete everything. If those three answers are vague, that is your answer.
The limits of a bot chat
Chat is a good door and a poor filing cabinet. Long documents, wide tables and side-by-side comparison are painful to read in a phone-sized message, which is why an assistant that lives in chat should still be able to hand you a file or a page.
A bot also cannot see your history in a chat before it arrived, and it cannot search old messages the way you can. If context matters, paste it or attach it.
Finally, a chat is a conversation, not a contract. Anything consequential, a send, a purchase, a change to a live system, deserves an explicit preview and an explicit approval inside that conversation.
A sensible first week
- Start with reading and drafting only. Ask for an inbox summary, a research brief, a first draft. Nothing leaves your hands.
- Give it one recurring job you already do by hand, and compare the result with your own work.
- Only then connect an account that can act, and keep every action behind an explicit approval while you learn what it gets right.
- After a fortnight, widen the rules for the jobs that were consistently correct, and leave the rest asking.
What this looks like in Tallis
You write to Tallis in Telegram or a web chat. It has a private space of its own for you, with its own memory, files and logged-in sessions. Actions that reach the outside world are previewed exactly and wait for your tap, unless you have written a rule that says otherwise.
Common questions
Is a Telegram assistant private?
The transport is Telegram's, and the bot operator can read your messages to it. Privacy comes from how the operator stores and separates your data, not from the chat app. Ask for that in writing.
Can it work in a group chat?
Ours does not, in the first round. Group chat changes who can instruct the assistant, and that is a permission question we would rather solve properly than early.
Do I need my own API key?
Not with Tallis. Model access is included, and you do not choose a model from a menu.
What if I prefer a browser?
There is a web chat as well. It is the entry point for testers during the beta.
Put the assistant where you already talk.
Telegram or web chat, a private space of your own, and a tap before anything is sent, unless you have written a rule that says otherwise. Join the waitlist.
Join the waitlist