Story
I got tired of explaining everything again
Whisperer began with a simple thought: an AI that already heard your conversation should not have to ask what it was about. This is the story of how a single panel over a video call grew into a workspace where meetings, tasks, knowledge and calendar share one context.
Where it started
The first version appeared in late May 2026 and did very little: a panel over a call that heard both sides, kept a transcript and suggested an answer while the conversation was still going. Nothing else. That was enough to see the idea was alive.
I built it for myself. I did not need a service that mails a summary ten minutes after the call. I needed something that hears the conversation at the same time I do.
What was missing
Every strong model has the same weak spot: it knows nothing about you. Each conversation starts on a blank page. You explain what you are working on, who you agreed with and about what, what was decided last Tuesday — and tomorrow you explain it again.
This is not about context window size. The context of your work does not live in one file. Part of it is in a call, part in tasks, part in documents, part in a calendar, and part nowhere except your own head. Tools cut that context along their own borders, and the model got the narrowest slice of all: whatever you managed to type into a chat box.
That is why an assistant that can do everything helps so little in practice. It answers the question without understanding why you asked it.
An AI should not ask what the meeting was about when it sat through the meeting.
How the idea changed
At first it was a helper on the call: hear, transcribe, suggest. It became clear fairly quickly that the valuable part happens after the conversation, not during it — when something has to be done about what was said.
So notes arrived, then tasks, then a map of decisions. Then calendar and files, so you would not have to leave the product to act on what you had just agreed. Then a knowledge base, so an answer could rest on your documents instead of general facts from the internet.
By that point it was no longer a meeting assistant. Each piece on its own exists in a dozen products; the value here is that they read and write one shared context.
Why Leo appeared
A helper that only answers hits a ceiling fast. The next step is obvious and difficult at the same time: let it act.
Leo got the name in August 2026, but the name was not the point. It got memory that survives between conversations, tools inside your own services, and a rule to ask permission before changing anything. The distance between suggest and do is not model size. It is memory, permissions, and responsibility for consequences.
It matters to me that it stays legible: you can see what it remembers, you can see what it is using, and it never does anything irreversible in silence.
Why Whisperer
The name was not accidental. A whisperer speaks quietly and only when there is something to say. A good assistant does not compete for your attention: it stays silent until it is needed and shows up exactly when it is.
That is where the product decisions come from. The panel is invisible when you share your screen. No bot walks into the call asking to be let in. A suggestion appears when you asked for one. An AI that keeps reminding you it exists gets in the way of the work — and the work is the point.
I want to build an AI that understands not only what you are asking, but what you are working on.
What I am building now
Today Whisperer is meetings, calendar, tracker, knowledge base, files, memory and Leo in one context. Not a row of tabs sitting next to each other, but parts that read and extend the same thing.
A decision made on a call becomes a task on a board, a promise Leo will remind you about, and a fact you can still find six months later. No exporting, no copying across — because this is one product, not five.
Where this is going
What interests me is an AI that knows more than the question: what you are working on right now, what already happened, what was agreed, what is still open, what you may have forgotten, and what context will be needed next.
This is not about a smarter model. Models will get better without us. It is about giving them the right context at the right moment — while the person stays the owner of their own data.
It is too early to call Whisperer finished. But I know the direction exactly, and it will not change: a workspace where the AI understands your work instead of guessing at it from a single message.