We give agents memory.

Plenty of systems can store data. We care about something else: that an agent has a memory of its own — the kind experience grows out of. A database answers a query and stays as it was. Memory changes whoever carries it: skills grow out of what accumulates, caution out of mistakes, habit out of what keeps recurring. We work on making that real for an agent — so it grows through its own experience instead of starting from nothing every time.

This is about an agent that remembers you and remembers itself — what you have already tried and how it turned out, what you keep coming back to, and which of its own suggestions missed.

A personality grows out of memory, not out of a prompt. What an agent becomes is decided by what it remembers and how that is connected — accumulated skills, wins, misses, and the conclusions drawn from them. We build that layer as a system in its own right, because that layer is what carries the rest.

A mistake is material, not waste. Memory that keeps only successes is useless: learning happens on exactly what did not work. We record our own negative results alongside the positive ones, and we ask the same of the system.

Memory belongs to whoever it is about. Local by default — not as a privacy checkbox, but as the shape of the thing. Memory that lives on someone else's machine is no longer quite your memory, however good the privacy policy is.

A system owes the truth about itself. An honest "I couldn't" is worth more than a "done" with something quietly left undone. A large share of our work is closing the distance between what a system does and what it says it did.

A small independent lab. Our first product, An Airy Spirit — long-term memory for AI agents that runs on your own machine. Has been public since October 2026: anairyspirit.com

📍 Chile · seventhwalllab.com