Data distillation.
It is very important to keep in your second brain, which I hope you have, not only everything that has happened to you in life, where, loosely speaking, it has access to your email, Telegram, WhatsApp and so on.
It is also important to keep everything that happened in all your sessions, in all your LLMs: that is, you have ChatGPT, Grok, Claude, maybe some DeepSeek and so on.
And you have had a large number of sessions.
You were looking for something, booking a therapist, trying to buy some mouse or monitor, writing a letter to your landlord. And all of that is quite valuable information.
All the deep research you have run is certainly very valuable too.
And I have set it up so that all of my sessions, every single one, both the coding sessions like Claude Code and Codex and the ones from an ordinary browser, when you are simply running deep research, all of them land in my storage, in my vault.
Entities are distilled out of them.
That is, loosely speaking, from sessions that held a long, long slab of knowledge, data and text, out of that long slab of text we distil literally a few sentences: what it was about, what pain we solved, and what we actually achieved.
And it is this distillate that new sessions search. Every time I start a new session, I have set it up so that the new session always tries to work out whether we have worked on this topic at all before.
And very often it turns out that the knowledge was already in my storage.
Loosely speaking, over the last week I asked my Claude and Codex 20 to 30 times which meetup I should go to next in San Francisco.
And if I did not have this memory of mine, it would search for the event from scratch every time, look up an event for me and so on. Whereas here my cofounder goes into my storage, looks at which events we have already registered for and which ones it has already recommended, checks whether anything new has appeared, and gives me a very clear recommendation.
And again, when you have many computers, and I have many computers, you can ask from one computer, then walk away from it and ask the same thing from another.
And the other computer, since you have a single shared memory area between your own machines, will see what you were discussing on the other one and give you the information you need.
That is, a session can start on one computer and finish on another.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20261022.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20261022.md. Just hand this link to your coding agent (Claude Code, Codex, Cursor) and it will figure everything out: it is written for machines.
🔗 All our channels and contacts in one place: https://linktr.ee/PaloAltoAI
Invented by Mycroft and Tony Dzi (Anton Dziatkovskii), Palo Alto AI Research Lab. Proudly made in Silicon Valley.
Top comments (0)