Knowledge you don't have to remember to save
Trail is a knowledge engine. It takes what you already produce — documents, pages, conversations, decisions — and lets a language model read it and write it up as a graph of cross-referenced Neurons. Not search results. Knowledge.
The difference from ordinary search is worth holding on to: a new fact updates what is already there, instead of becoming one more fragment. The more you put in, the smarter it gets about you — rather than merely bigger.
Trail is the universe's shared memory, a second brain if you like — the place an insight stays put, so it doesn't have to be had twice. It has now gained two new ways of reaching that knowledge: Ambient, which watches along while you work, and Web Clipper, which saves the page you are on.
Ambient — capture that costs nothing
The knowledge hardest to get hold of is the kind nobody writes down. A realisation in the middle of a piece of work. The reason you chose one solution over the other. The thing you worked out at four in the afternoon and had forgotten by Friday.
It exists only in the head of the person who had it — until someone bothers to write it down. Nobody does. Not out of laziness, but because the price of saving is higher than the value in that moment.
Ambient removes the price. It sits in the menu bar on your Mac and watches along with what you are working in. It gathers your activity into working windows and files what you work out into your Trail, without you writing anything. Afterwards you can search your own working day.
Setup takes a minute: get the app, sign in with your Trail account, choose which knowledge base it may write to. After that there is nothing more to do — and that is the whole point.
It is built to have access without abusing it. The text on screen is read on your own machine using Apple's text recognition; the picture of your screen never leaves it. There is a deny list — passwords, banking, messages — and it is checked before a screenshot is taken at all. It does not capture keystrokes.
And you can always see what it filed. Everything lands in a queue you can review, and a Neuron can be archived again. Trail's own premise holds here too: the model proposes, you decide.
Web Clipper — save the page, not the address
You read something useful. You save a bookmark. Six months later you go looking, find the bookmark, click — and the page has changed. The content you remembered is gone, and you could never have searched it anyway.
Web Clipper saves the content. A button in the browser puts the page you are on into the right customer's knowledge base, as a source Trail reads and writes up alongside everything else it knows. After that you can search it — and find it again through everything it connects to.
Its most useful property is also the least obvious: it works behind a login. It reads the page as your own browser sees it, after you have signed in — it does not fetch it from a server somewhere else. So it reaches things an ordinary crawler never gets to: customer portals, internal wikis, documentation behind a paywall, a thread in a closed forum.
It touches only the tab you are on, and only in the moment you click.
Why they belong together
They are two funnels into the same memory, and they cover opposite ends of how knowledge actually arises.
Ambient takes what you do. The clipper takes what you read. One is your own work; the other is the world outside. Both become Neurons in the same graph — and a graph does not merely get bigger with more sources, it gets better connected.
Both can write to several customers' knowledge bases, each kept apart from the others. A consultant can clip from two customers' portals in the same afternoon without a single word landing in the wrong place.
Getting started
What used to require someone remembering to write it down now happens by itself — and what you read and saved can be found again, because it is content and not a link.
Read more about Trail and the rest of the flagships, or talk it through with Christian if you want to know what it means for your own knowledge.