A personal AI assistant in Telegram that does the work
Search for an AI bot for Telegram and you mostly find two things: chat windows that forward your message to a model and send back the answer, and code on GitHub you can host yourself if you have an evening and a server. Both answer when you ask. Neither does anything while you are not asking.
The useful version works the other way round. It knows your notes and your tasks, it runs the small jobs you gave it on a schedule, and it messages you — the briefing at eight, the price that moved, the threshold that was crossed. Telegram is just where you read it, because your phone is already open there.
This is what that looks like in Aplite, where Telegram is one of the places an agent can deliver to.
Three things a chat bot does not do
It works from your own material. Every Aplite workspace is a folder of markdown notes the agent can read — projects, lists, the plan for the week. Ask it in Telegram what is left on a project and it answers from the note, not from the internet.
It writes to you first. An instruction like “every morning at eight” becomes a task with a schedule. The agent runs it, and the result lands in your Telegram chat without you opening anything.
It asks before it acts. Anything that changes your files or sends something on your behalf goes through approvals: you see what the agent wants to do and decide.
Five things to ask it
Each of these is one sentence. Copy it into the chat, and the agent sets up the task, the schedule and the delivery to Telegram itself.
Every morning at eight, send me my day on Telegram — calendar, weather, and what’s waiting in my notes.
The morning briefing: one message before your day starts.
Watch the price on this product page and message me when it moves: [link]
The price watcher checks the page once a day and stays quiet until the number changes.
Watch my tickers against the levels I set and message me only when one is crossed.
A market threshold alert — your levels, not a feed of every tick.
Every morning, brief me on my topics — a few lines per story with the source link underneath.
A news digest on the subjects you pick, with links you can follow from the phone.
Every morning at seven, turn my Todoist into a realistic plan for the day.
The daily plan from Todoist: overdue and due-today tasks, put in an order you can actually get through.
More routines like these — and what each one needs connected — are on the use cases page.
Setting it up
- Create a workspace. Sign in with Google and pick your agents — the workspace is ready in about a minute. The agent already answers in the web chat.
- Make a bot. In Telegram, open @BotFather, send
/newbotand copy the token it gives you. - Connect it. In Aplite, open Skills → Telegram, paste the token and press Validate. The workspace restarts to pick up the bot — about a minute.
- Say hello. Send
/startto your new bot — a Telegram bot cannot message you until you have written to it once. - Approve the chat. A pairing request appears under Telegram in Aplite; press Approve. Only chats you approve can talk to your agent.
From then on, anything you ask for “on Telegram” goes there, and you can message the agent from your phone like any other chat. The Skills and connections page has the details, including how to pause the channel without disconnecting it.
Your bot, your workspace
If you have looked at building this yourself, the usual recipe is an n8n workflow: a Telegram trigger node, a model node, a memory node, one more workflow per schedule. It works, and every new routine is another diagram to wire. Here a new routine is one more sentence.
The bot you create is yours: you name it, and it talks only to your workspace and to the chats you approved. Each workspace runs in its own container on the open-source OpenClaw runtime, so the setup that self-hosting guides spend an evening on — server, config file, pairing — is already done. The models are your choice too: the agent runs on any model from the catalog, and you see what each answer cost.
Try it in your workspace
Get started