01 The problem
You open fifty tabs researching something, close them, and come back a week later with no memory of what you found or what you had ruled out. Browser history does not help: it is a list of URLs with no record of what you were trying to do.
So the extension watched what you browsed, and every page you stayed on long enough became an intent-rich memory rather than a link — “comparing observability vendors, leaning self-host” instead of datadoghq.com/pricing.
It worked. I killed it anyway.
02 How it works
03 Find a threshold that works
A browsing session, and the capture rule applied to it. Drag the dwell threshold and watch what survives — the whole bet was that time-on-page separates intent from noise.
The extracted line is what the memory layer stores instead of a URL. That is the part that worked. The part that killed it is underneath.
04 Why I killed it
- There is no good threshold. Play with the dial above. Set it low and you are capturing someone’s inbox and their banking tab. Set it high and you lose the research trail that was the entire point. No value cleanly separates the two, because dwell time does not know what a page is.
- Passive capture is a different product than it looks like. “Remembers your research” and “logs everything you read” are the same system described two ways, and only one of them is what I wanted to install.
- I was the first user. The test that ended it was noticing I did not want it running on my own machine. If I would not run it, shipping it to other people was not defensible.
- What I kept. The idea that memory should store intent rather than artifacts survived — it is the thing Yaad does with a life and Hindsight does with predictions, in both cases over data the user explicitly handed over.
Honest limits. It worked well enough that killing it was a judgement call rather than a forced one, and a version with explicit opt-in per site would probably be fine. I have not built that, and listing this as dead is more honest than listing it as paused.