Mark.
An AI that remembers you, thinks about you and about what matters.
AI got stronger.
And you still open it only when you feel like it.
It writes code, draws pictures, puts drafts together. But check the stats: you open it a couple of times a week, not every day. And it's no longer about model quality. It's about the form — it forgets, lives apart and notices nothing until you knock on its door yourself.
It forgets you, from scratch, every single time.
It isn't built into your daily flow.
It doesn't notice what matters while you're busy.
That's why even the best model remains a tool of occasional requests. Powerful, precise, fast — but passive. If you don't remember to ask, it never shows up in your day. If you don't re-explain the context, it answers from zero. While you were busy, something important changed — and nobody told you, because nobody was watching.
Mark is built differently. It isn't trying to be yet another place for questions.
How we tried to remember.
First there were notebooks. Then digital notes — with search, tags, sync. Then came Zettelkasten: interlinked cards, the “second brain” — the idea that notes should live, not just sit there.
But all these tools are passive. A notebook waits for you to open it. Search works only when you ask something yourself. A “second brain” links your notes — but it doesn't read them for you. Doesn't watch videos. Doesn't listen to podcasts. Doesn't keep an eye on what interests you while you're busy with something else.
The next step is a brain that works while you're busy with something else.
A new episode dropped — it already tracks the dub and will let you know. You forwarded an article you never got around to reading — it gave you the gist, knowing your interests. A forty-minute video — compressed into two minutes of reading. A two-hour podcast — distilled into key points tuned to your focus.
That leaves you with one job: send more. The rest is its work.
Mark is not another AI chat.
It's your working layer right inside Telegram.
A strong AI answers the first question well. Mark gets more useful on day 10 and day 30 — it remembers you, finds what you need in your own past and keeps running the things you don't want to think about every minute.
Remembers your working context
People, plans, constraints, habits and the links between them. Not a message history — a living working memory of you.
Finds things in your own past
Files, voice notes, links, old conversations become a personal archive you can pull from in plain language.
Keeps watch over what must not be lost
It watches, checks, monitors and gathers. It doesn't rely on you remembering to open the chat at the right moment.
Comes back only when there's a reason
It doesn't behave like a notification feed. It writes when there's a real occasion: a signal, a result, a reminder, an answer it found.
Every class of product has its own job.
General-purpose AI is great when you need a fast answer. Dev agents — when the work is code. Mark covers a different layer: it keeps your personal context, searches your past and comes back on its own when a useful signal appears.
General-purpose AI deals with the moment and answers now. Dev agents execute a task in the IDE or terminal. Mark — keeps your context and brings back what's useful later.
General-purpose AI — on the web and in apps. Dev agents — in the code editor and terminal. Mark — right inside your Telegram conversation.
General-purpose AI — the history of conversations inside the product. Dev agents — the project, the code, the tools. Mark — your history, files, plans, constraints and recurring patterns.
General-purpose AI — when you open the chat yourself. Dev agents — during a work session. Mark — on request, and later, when there's a reason to write.
General-purpose AI — almost nothing; that's not the core scenario. Dev agents — only within an active session. Mark — watches, checks and writes only on signal.
General-purpose AI — from the quality of a new prompt. Dev agents — from deep access to code and processes. Mark — from accumulated knowledge about you and your history.
Memory, search, background work and signal quality — the four layers that add up to Mark's everyday value. Below — what each of them looks like in practice.
It doesn't remember the chat.
It remembers you.
Mark's memory is not a log of recent messages. It's a working model of your context: people, plans, constraints, habits and the links between them. Every fact carries a confidence level — certain or probable — and a pointer to the moment you said it. Old and rarely used facts fade in weight on their own. Memory stays alive instead of turning into a junk pile.
Memory only matters when it changes behavior. Not “the bot knows a lot” — but “the bot won't try to schedule a meeting on Friday, because it remembers I don't take calls on Fridays.” Products that look alike on the outside are fundamentally different underneath.
It works,
while you're busy with something else.
The second layer is background work. You describe a task in plain language. Mark launches a watcher — it follows a schedule or a condition, stays silent while everything is normal, and writes only when something important happens. One such agent already changes the habit: a product that works while you sleep just feels different.
On the free plan you get one active watcher. That's enough to feel the difference. On the paid plan there are several, and they work together, cross-checking against your memory and past signals.
Your life gets
a search bar.
The third layer is search. Conversations, files, voice notes, links, videos — everything you've ever sent or said becomes your personal index. You ask in plain language and get not an abstract answer, but a pointer to a specific moment in your own past.
What you've already seen, said or sent no longer gets lost. For many people this is the first real aha moment: Telegram starts working not as an endless feed, but as personal memory with fast access.
And the fourth — the most important.
Mark knows how not to write. It remembers what it has already sent, sees what you reacted to, and tells a strong reason from a weak one. For Mark, silence is a decision just like a message. Most proactive products optimize the number of touches. Mark has a different optimizer — signal quality. And that can be shown in numbers.
Proactivity under control.
Here are the real numbers.
An internal audit of one active user, January–February 2026. The numbers show the main thing: Mark systematically filters out weak reasons to write — and keeps only the appropriate ones.
Situations the system evaluated at all as a reason to message the user.
Two-thirds suppressed as inappropriate, weak or too frequent. That's exactly how the rate-limiting layer works.
Actually reached the user after filters, pauses and appropriateness checks.
Behind these numbers isn't a single cooldown, but a whole layer that can decline a weak reason. Mark optimizes not for frequency of touches but for signal quality — and that builds trust more than any “writes first” promise ever could.
What this looks like
in an ordinary week.
Not marketing cases and not “the future of AI.” Just a short slice of how Mark shows up in the day of someone who already lives in Telegram. All examples are real patterns, only lightly anonymized.
BTC dipped 7% overnight, then recovered to −3% by the open. I didn't message you right away — the recovery was too fast. Sending this now so you have the picture of the day.
You asked about “that presentation with Igor's numbers” — it's from November 8. 11 slides, the main point on the third. Want a two-minute summary, or should I send the file?
I re-read your contract — it has auto-renewal 90 days before the term ends. The deadline is March 12. Set a reminder two weeks ahead?
Marina's birthday is next week. Last year you gave her a book by Taylor — she loved it. Repeat the genre or try something new?
In your chat with Victor, “when are we checking Bangkok” came up again. I remember your budget of up to $300 — there are three options under $280. Show them?
Over one week — five small decisions you didn't have to make yourself. That's not magic and not AGI. It's a simple idea: if you remember the context and keep the thread, half of the everyday load disappears.
Three types of people
Mark clicks with fastest.
Those for whom Telegram is already the work environment.
Founders, managers, freelancers. Their Telegram holds people, agreements, files and voice notes. Mark adds memory, search, lists and background signals to that flow. This is where the product slips into the daily rhythm instead of becoming “yet another tool.”
Those who care more about signal than conversation.
Crypto, stocks, commodities. They don't need an “AI trader” — they need one monitoring agent that watches around the clock, gathers context and doesn't bury them in noise. Here Mark sells as a layer of signals and short briefs delivered at the right moment.
Those who are often on the move.
Two-way voice translation, flight prices, time zone awareness, parsing documents in a foreign language. The value lands on day one — in a taxi, in a restaurant, at the airport. It's the best quick entry into the product.
Mark isn't trying to become an assistant for everyone first. First it has to win where Telegram is already the daily work environment — a narrower entry, but commercially more honest and stronger.
It knows a lot of practical things —
no extra apps needed.
Beyond memory, watchers and history search, Mark has a set of tools that extend it without requiring you to install other apps. They're not the core of the story, but they're often what closes the small everyday gaps and helps you feel the product sooner.
Files and documents
Parsing PDF, DOCX, XLSX, PPTX. Contracts, invoices, reports — key points, amounts and hidden pitfalls.
Voice ↔ voice
Multilingual voice translation. Handy when traveling, messaging with foreigners, or in live conversation.
Photos and video
A receipt becomes a table. A menu in a foreign language becomes an order plan. A forty-minute video becomes two minutes of reading.
Internet, news, markets
Smart web search, news monitoring, data on crypto and traditional markets — for when you need not a chat, but signal and context.
Lists and personal data
Shopping, watchlists, to-dos, preferences — in one place and at hand.
Creating media
Images, simple videos, music. A useful extra for cards, stories, ideas — not the core of the subscription.
All of this sits around the core. The essentials are memory, history search, background work and precise proactivity. The rest steps in to help when needed.
Free must let you feel the magic.
Pro — lift the ceiling once Mark has become part of your day.
With Mark you can't sell trust or basic usefulness. So Free has to be genuinely strong: memory, search through your own history, file handling and one active watcher. The paid layer starts where the product has already settled into your flow and you need more capacity, depth and control.
Memory across your conversations. Search through chats, files and voice notes. 1 active watcher. Basic limits on voice and media. Enough to understand whether you need this kind of product every day.
Several active watchers. More frequent checks, more sources, higher limits on files and voice. Deeper memory and control over what Mark remembers. A status screen: why it wrote and what it's doing right now. Plus daily and weekly digests.
Images beyond the limit, video and music on demand. No separate card, no extra subscription. Pay only when you actually need it.
Your data is yours alone.
Honest about what already exists and what's still being built.
We don't like selling trust with words. Below is what already works in the code at the time of publication — and what honestly remains in progress.
First value — in 2 minutes.
Habit — in 30 days.
In 2 minutes — the first answer from your own past. In 24 hours — the first background signal from Mark. And by day 30 it works as part of your ordinary day, not as yet another chat.
Send one message — and see for yourself.
@usemarkbot→