The bliq team
From advice to action
Booking the table, ordering the gift, tracking the flight — and the rule that anything spending your money or speaking in your name stops and asks first.
How bliq remembers you
The things you only said once are the things worth keeping. A look at what bliq stores, why it stores so little, and how to see and delete any of it.
The bliq team
bliq
The single most tiring thing about talking to a computer is that it does not remember you talked to it.
You explain that Clara is your sister. Tomorrow you explain it again. You mention you do not eat meat and it recommends a steakhouse the following week. Every conversation begins with a paragraph of context you have already typed twenty times, and after a while you stop bothering and just do the thing yourself.
So the first real problem we had to solve was not intelligence. It was memory.
What is worth keeping
The instinct is to store everything. It is the wrong instinct — a system that remembers every word is functionally the same as one that remembers nothing, because the useful line is buried under ten thousand useless ones.
What is worth keeping is small and specific:
- The people. Who they are to you, when their days are, how you talk about them.
- The standing facts. Where you live, what you will not eat, which airline you avoid, which pharmacy is yours.
- The preferences you would never write down. Aisle seat on long flights, window on short. No calls before ten. Nothing scheduled the day after a red-eye.
- The open loops. The thing you said you would deal with, that you have not dealt with.
Almost all of it arrives once, in passing, in the middle of a sentence about something else. "I can't do Thursday, that's the dentist." That is the whole mechanism — you were not filing a fact, you were just talking.
You can look at all of it
There is a list. It is the actual memory — not a summary of it, not a friendly rendering of it. Everything bliq knows about you is a line on that list, and each line says where it came from.
That matters for two reasons.
It has to be correctable
Memory picked up from conversation will sometimes be wrong. You said something sarcastically. You changed your mind in April. Somebody else was using your phone.
If the only way to fix a wrong fact is to argue with a chatbot about it, the wrong fact stays forever. So it is a list, and you delete the line, and it is gone.
It has to be legible
An assistant that quietly accumulates a model of you that you cannot inspect is not something we would want to use, regardless of how well it worked. "It just knows" is not a feature, it is an unexamined liability.
If you cannot see what it knows, you cannot know what it does not know.
What it deliberately does not do
A few things we decided against, and have kept deciding against:
- It does not read everything and keep the interesting bits. Connected apps are used for the job at hand. A flight lookup is a flight lookup, not a harvesting opportunity.
- It does not build a profile for anyone else. Nothing about you is sold, brokered, or shared with advertisers. There is no version of bliq that is ad-supported.
- It does not train anyone's model on you. Not ours, not a provider's. Your life is not a dataset.
- It does not keep what you deleted. Deleted is deleted, not hidden from the interface.
The longer version of all four is in our privacy policy, and we tried hard to write it in sentences rather than clauses.
Why this is the foundation
Everything else bliq does depends on this part working.
It cannot text you first if it does not know that the 9th matters. It cannot book the table if it does not know there are four of you and one does not eat fish. Proactivity and action are both just memory, applied.
Get the memory right and the rest is mostly follow-through. Get it wrong and you have built another chat window.
Let bliq handle the rest.
It knows you, checks in on you, and quietly gets the small things done. On iPhone today.
